Beautiful Danish Data

• Labour market participation, part-time and full-time work Employment and labour force status
  Participation means at least 481 annual hours (e.g. 37 hours for 13 weeks) on BFL pay slips together with annual income of at least 50,000 DKK; part time is 481-1,664 annual hours; full time is more than 1,664 hours. Monthly hours and income are scaled down by the share of the month registered on parental leave in the OF register.
  Registers: BFL (eIndkomst), OF
  Codes: thresholds 481 and 1,664 annual hours; 50,000 DKK
  working paper · Section 2; Supplemental Material A
• Annual labour income (BFL) Earnings and income
  12-month sum of monthly labour income from BFL pay slips, by calendar year for age profiles and centred on the birth month for event-study moments (e.g. income from January-February before a March birth counts as year -1); excludes income registered while on parental leave; deflated by CPI, in 1000 DKK. Estimation sample 2010-2018.
  Registers: BFL (eIndkomst), OF
  working paper · Section 2; Supplemental Material A
• Personal income (tax sample) Earnings and income
  Personal income from annual tax data: labour income from wage work, profits from own firms and labour market transfers; deflated by CPI.
  Registers: IND
  Codes: PERINDKP
  working paper · Supplemental Material A
• Couples / partners Family links, fertility and household
  Partners follow Statistics Denmark's family definition, including married and cohabiting couples; cohabitants are two adults at the same address registered as parents to a child, or two opposite-sex adults with an age difference under 15 years at the same address. Constructed from EFALLE in BEF; couple age is the woman's age. Estimation sample keeps couples with at most five years age difference.
  Registers: BEF
  Codes: EFALLE; FOED_DAG; KOEN
  working paper · Section 2, footnote 8; Supplemental Material A
• High skilled vs less skilled Education and test scores
  High skilled if at least 180 months of education (at least a bachelor's degree), measured as the highest degree earned in the sample period; less skilled otherwise.
  Registers: UDDA
  Codes: HFPRIA >= 180 months
  working paper · Section 2; Supplemental Material A
• Labour market experience Employment and labour force status
  Statistics Denmark's experience measure based on accumulated mandatory ATP pension payments, which depend on hours worked; part-time employment counts as 2/3 of full time.
  Registers: IDAP
  Codes: ERHVER, ERHVER79
  working paper · Section 2; Supplemental Material A
• Sample exclusions by main activity Sample definition and timing
  Individuals aged 25-60 with an opposite-sex partner; observations dropped if mainly student, self-employed, retired or on disability pension, based on main activity at end of November.
  Registers: IDAP/AKM (socio-economic status)
  Codes: SOCSTIL, SOCSTIL_KODE
  working paper · Section 2; Supplemental Material A
• Mechanical net-of-tax rate change (instrument) Taxes and tax rates
  Four-year mechanical change log(1 - tau_{t+4}(z_t)) - log(1 - tau_t(z_t)) in the marginal net-of-tax rate, holding income and characteristics at year-t values, computed with a Danish tax simulator; used to instrument own and partner net-of-tax rate changes and virtual income. Restricted to couples where both have personal income of 50,000-600,000 DKK in the base year. Outcome is the 4-year difference in number of children born.
  Registers: IND
  working paper · Section 3; Supplemental Material A.1
• Household wealth Wealth, debt and consumption
  Bank deposits, bonds, stocks and property net of loans and mortgages at end of year; registered property values raised by 10% (following Leth-Petersen 2010); deflated by CPI, in 1000 DKK.
  Registers: IND
  Codes: FORM; FORMREST_NY05 (after 1996)
  working paper · Supplemental Material A
• Child linkage and age of children Family links, fertility and household
  Children and their ages are linked to parents through the parent identifiers in BEF; almost all children can be linked to a mother.
  Registers: BEF
  online appendix · Supplemental Material, Data Appendix (p. S2)
• Estimation sample vs tax sample Sample definition and timing
  Two samples: the BFL estimation sample of all individuals aged 25-60 with partners, 2010-2018; and the tax sample of women aged 25-40 with partners, 2004-2018 (used with personal income PERINDKP). Both split by less skilled vs high skilled.
  Registers: BFL (eIndkomst), IND, BEF
  online appendix · Supplemental Material, Table A.1

• Default liquid wealth Wealth, debt and consumption
  Third-party-reported year-end value of current accounts, savings accounts and securities accounts (bank deposits and listed securities); excludes durables, real estate and pension accounts and does not net out debt. Available for the full population; mirrors the 'default wealth' the authorities use for the old-age check asset test.
  Registers: IND, Statistics Denmark administrative wealth registers
  paper · Section I.A and I.C; Table 1 notes; Figure 1 notes
• Reported liquid wealth Wealth, debt and consumption
  Default liquid wealth plus self-reported corrections submitted via the government website (e.g. adding cash, reducing account balances); available only for individuals above the statutory retirement age. The variable available to the authors does not capture corrections made close to year-end.
  Registers: SOCP, IND
  paper · Section I.A and I.C; Table 1 notes
• Total wealth Wealth, debt and consumption
  Default liquid wealth plus pension wealth and housing wealth, net of liabilities (approximates total net wealth for the low-income elderly).
  Registers: IND, PENSFORM
  paper · Figure 1 notes (panel B)
• Old-age check eligibility (age and income) Retirement and pensions
  Age-eligible = aged 65 or older (statutory retirement age in the sample period). Low-income (income-eligible) = income, net of social security and before self-reported corrections, below the threshold at which the old-age check is fully phased out (disregard DKK 42,000, phase-out rate 34% for singles in 2020). Asset test: check lost entirely when liquid wealth at end of t-1 exceeds DKK 90,000 (2020 value; threshold raised 15% in 2010). For the 2001-2007 reform analysis the 2003 income threshold is applied throughout.
  Registers: IND, SOCP, BEF
  paper · Section I.A; Figure 1 and 2 notes; Supplemental Appendix Figure A.4 notes
• Treatment and control groups for bunching Sample definition and timing
  Treated: low-income singles aged 65+. Controls: low-income singles aged 60-64 (age-ineligible) and singles aged 65+ with incomes between 150% and 200% of the full phase-out income threshold. Sample restricted to singles; main cross-section pools 2014-2018 (2014-2016 for bank data).
  Registers: BEF, IND
  paper · Section I.C (Sample); Figure 1 notes (panel C)
• Excess mass below asset-test threshold Wealth, debt and consumption
  Panel version: dummy for default liquid wealth between 50% and 100% of the asset-test threshold regressed on age dummies and individual fixed effects for low-income elderly born 1941-1945, coefficients relative to age 64; bunching jumps by 2.5 points at eligibility age 65 and keeps rising with age. Liquid wealth inflated to 2020 values with the growth rate of the asset-test threshold; DKK 5,000 bins.
  Registers: IND, BEF
  Codes: 50%-100% of threshold window; reference age 64
  online appendix · Supplemental Appendix, Figure A.3 notes
• Cohort-approach counterfactual Sample definition and timing
  For single income-eligible individuals born 1941-1945, compare the indexed liquid wealth distribution at ages 70-74 (five to nine years into eligibility) with ages 60-64 (before eligibility); only those income-eligible in a given year are included (unbalanced panel).
  Registers: IND, BEF
  paper · Section II (Cohort Approach); Table 1 Panel B2 notes
• Bank liquid wealth, spending and cash withdrawals Wealth, debt and consumption
  Bank data 2014-2016 used to compare spending and changes in consumer debt in December, January and other months by bank liquid wealth (DKK 5,000 bins), for age-eligible (65+) low-income singles vs age-ineligible (60-64) low-income singles.
  Registers: Danske Bank customer data
  online appendix · Supplemental Appendix, Figure A.4 notes
• Old-age check parameters and indexation Retirement and pensions
  Old-age check (created 2003) parameters expressed in 2020 values by inflating with the index used to adjust tax and transfer parameters (tied to average wage earnings per worker; $1 = 6.53 DKK in 2020). Maximum check raised above indexation in 2005, 2007, 2009 and 2013; asset-test threshold raised about 15% above indexation in 2010; income test (phase-out start and end thresholds) very strict initially but unchanged in real terms since 2005 except a small increase in 2019.
  Registers: Tax-transfer rules
  Codes: $1 = 6.53 DKK (2020)
  online appendix · Supplemental Appendix, Figure A.1 notes

• ATP-based hours (1980-2018) Hours worked
  Hours inferred from mandatory ATP pension contributions, which employers pay based on a binned measure of hours; for monthly-paid workers annual contributions depend on monthly hours in four bins, giving 37 annual hours bins. Capped at full time in all 12 months.
  Registers: ATP pension contributions
  working paper · Section 3, footnotes 17-18
• Monthly uncapped hours (from 2008) Hours worked
  Uncapped hours worked for all workers at monthly level, available from 2008; used where precise hours variation is needed (aggregated to quarters in the promotion event study).
  Registers: Monthly payroll records
  working paper · Section 3; Section 4 Fact 3
• Job (firm x occupation cell) and job switch Occupation and job hierarchy
  Job is a firm x occupation cell; a switcher moves between firm x occupation cells after the reform. Occupations use DISCO-08 from 2010 (563 occupations), DISCO-88 1991-2009 (372) bridged with the Humlum (2021) crosswalk, and an older Danish classification before 1991 (299). Switches where more than 50% of an individual's coworkers move to the same job cell in the same year are dropped.
  Registers: Employer-employee matched data
  Codes: DISCO-08; DISCO-88; pre-1991 Danish occupation codes
  working paper · Section 3, footnote 15; Section 5.1
• Mass layoff switchers Firm characteristics and productivity
  Mass layoffs are layoffs in which firms with at least 20 employees reduce their workforce by at least 30% in a single year; switchers are workers who move firm following such a layoff. Used in an event study of log earnings vs a linear pre-layoff trend.
  Registers: Employer-employee matched data
  Codes: firm size >= 20 employees; workforce reduction >= 30% within one year
  online appendix · Online Appendix, Figure A.8 notes
• Labour earnings outcome Earnings and income
  Total labour compensation, including contributions to employer-administered retirement savings plans; outcome is the log change between 2008 and 2010 (placebo 2006-2008).
  Registers: Income tax records
  working paper · Section 5.1, footnote 26
• Reform treatment and simulated net-of-tax rate instrument Taxes and tax rates
  Treatment assigned by pre-reform baseline taxable income (labour income, transfers, pensions, alimony and certain capital income) in bins relative to the 2009 reform threshold (around the 70th percentile). The instrument is the simulated log change in the marginal net-of-tax rate from tax code changes, holding income at baseline (Gruber-Saez). Baseline sample: taxpayers with baseline income 250,000-1,000,000 DKK.
  Registers: Income tax records
  working paper · Section 5.1, footnote 27; Table 2 notes
• Promotion Occupation and job hierarchy
  Within-firm switch to an occupation cell whose median earnings are at least 10% higher (robustness 5% and 20%); promoted workers matched to unpromoted co-workers in the same firm.
  Registers: Employer-employee matched data, Monthly payroll records
  working paper · Section 4, Fact 3
• Occupational rank / top vs bottom occupations Occupation and job hierarchy
  Occupation cells (6-digit occupation codes) ranked by mean wage earnings; for top/bottom occupations the ranking is computed among workers aged 45-50 (mean, P10 and P90 of wage earnings reported per cell). Occupational rank used in balance checks orders occupation cells by their mean earnings.
  Registers: Employer-employee matched data (DISCO occupation codes)
  Codes: 6-digit DISCO occupation codes
  online appendix · Online Appendix, Table A.1 and Figure A.6 notes
• AKM firm wage premium Firm characteristics and productivity
  Firm fixed effects from an AKM model log z_it = alpha_i + psi_J(i,t) + X_it beta + nu_it, with controls for year, age and tenure-in-current-firm dummies, estimated on pre-reform data 2002-2005 for firms with at least 10 employees; estimated firm effects merged onto the 2009 tax-reform sample to measure changes in firm premia for firm switchers.
  Registers: Employer-employee matched data, Income tax records
  Codes: firms >= 10 employees; estimation window 2002-2005
  online appendix · Online Appendix, Section C and Figure A.9
• Survey hours (LFS validation) Hours worked
  Validation measure of hours: actual, uncapped hours worked from the Danish component of the EU Labour Force Survey, compared with the administrative ATP pension-based hours measure; the survey measure is much noisier, especially at the top.
  Registers: Danish Labour Force Survey (AKU), ATP pension contributions
  online appendix · Online Appendix, Figure A.1 notes
• Number of job switches (ages 20-50) Occupation and job hierarchy
  Count of job (firm x occupation cell) switches between ages 20 and 50 in a balanced panel of workers observed at every age 20-50, split by earnings percentile at age 50 (top 10%, top 1%, bottom 50%).
  Registers: Employer-employee matched data
  online appendix · Online Appendix, Figure A.3 notes
• Demographic controls (job-switcher regressions) Sample definition and timing
  Robustness controls: dummies for age, gender, marital status and number of children, alone or combined with initial job-cell fixed effects (baseline uses job-cell fixed effects only).
  Registers: Population register (BEF)
  online appendix · Online Appendix, Figure A.7 notes

• Reform treatment (birthdate cutoff) Retirement and pensions
  Those born on or after 1 January 1954 learned in 2011 that their early retirement age (ERA), incentivized retirement age and full retirement age (FRA) rose from 60, 62 and 65 to 60.5, 62.5 and 65.5; those born one day earlier were unaffected. Running variable is birth date relative to the cutoff; RD sample is those born within 56 days (8 weeks).
  Registers: Population register (exact date of birth)
  working paper · Sections 2.3 and 4.2
• VERP participant analysis sample Sample definition and timing
  Four cumulative restrictions: (1) Danes born within six months of the cutoff date January 1, 1954 (71,095 individuals); (2) regular participatory VERP contributors, defined as making contributions in 70% of pre-announcement years 2001-2010 (47,047); (3) balanced panel: present in the data in every year 2006-2018 (43,348); (4) drop the self-employed, defined over the pre-announcement period using Statistics Denmark's definition of an individual's main source of income (final analysis sample 40,042).
  Registers: Population register, VERP contribution records, Statistics Denmark main source of income classification
  Codes: birthdate within +/-6 months of 1954-01-01; VERP contributions in >=70% of years 2001-2010; balanced 2006-2018
  online appendix · Online Appendix Table A.1 and notes
• Contributions to employer-sponsored and personal retirement accounts Retirement and pensions
  Annual contribution flows to employer-sponsored DC accounts in levels (winsorized at the 95th percentile by year) and indicators for positive contributions to personal retirement accounts (also indicators for contributions of various sizes).
  Registers: Labour-market/income registers (tax-reported pension contributions)
  working paper · Section 4.1
• Changes in bank, stock and property values Wealth, debt and consumption
  Year-t minus year-(t-1) year-end values of bank account balances and stock market account balances (reported by financial institutions to tax authorities) and of property cash values as assessed by tax authorities; winsorized at the 5th and 95th percentiles each year (also 10th/90th).
  Registers: Tax-reported asset registers
  working paper · Section 4.1, footnote 20
• Pre-tax earnings Earnings and income
  Income on which individuals pay the 8 percent labor market tax, winsorized by year at the 95th percentile; 2010 prices, 5.56 DKK per USD.
  Registers: Income register
  working paper · Section 4.1
• Monthly retirement age Retirement and pensions
  Absorbing-state measure from the monthly eIncome register (2008-2017): age in the last month with positive earnings before earnings permanently fall to zero.
  Registers: eIncome (eIndkomst) monthly earnings register
  working paper · Section 4.1, footnote 17
• VERP benefit claiming Retirement and pensions
  Annual Voluntary Early Retirement Pension (VERP) benefit payments.
  Registers: VERP benefit records
  working paper · Section 4.1
• Firm default contribution rate Firm characteristics and productivity
  All individuals over 18 are assigned to workplaces in 2010; individual contribution rate = employer-sponsored plan contributions divided by earnings; the firm default rate is the median contribution rate among the firm's workers in 2010. Predicted contributions = current earnings times the 2010 firm default rate.
  Registers: Linked employer-employee data, Pension contribution records
  working paper · Section 6 (employer defaults), footnote 30
• VERP participation indicator (qualified UI-fund contributions) Retirement and pensions
  Participatory VERP contribution in a year is measured as an indicator for making qualified (VERP) contributions to unemployment insurance (UI) funds in that year; VERP eligibility requires contributions in 25 of the previous 30 years.
  Registers: VERP contribution records
  online appendix · Online Appendix Table A.7 notes; Appendix C.2
• RD control variables Other
  Baseline RD controls: indicator for being male, indicator for being married in 2010, and indicators for region of residence in 2010 (five regions: Hovedstaden, Sjaelland, Syddanmark, Midtjylland, Nordjylland). Regressions use separate linear polynomials in birthdate on each side of the cutoff with triangular weights.
  Registers: Population register
  Codes: 5 regions as of 2010
  online appendix · Online Appendix Table A.2 notes
• Retirement at eligibility-age spikes Retirement and pensions
  Control group (born before cutoff): first spike = retirement age between 60y0m and 60y5m; second spike = 62y0m to 62y5m. Treatment group: first spike = 60y6m to 60y11m; second spike = 62y6m to 62y11m. 15.8% retire at the first spike and 10.6% at the second as defined.
  Codes: monthly retirement-age windows 60:00-60:05, 62:00-62:05 (control); 60:06-60:11, 62:06-62:11 (treated)
  online appendix · Online Appendix B
• Pre-reform personal retirement plan users Sample definition and timing
  Subsample of individuals who had used personal retirement plans before the reform announcement: those who made contributions to personal plans in two or three of the years 2008-2010.
  Registers: Tax-reported pension contribution records
  Codes: contributions in >=2 of 2008, 2009, 2010
  online appendix · Online Appendix table notes (personal-plan subsample)
• Outcome winsorization and time windows Sample definition and timing
  Anticipation-period outcomes are averaged over 2011-2013 and pre-announcement placebo outcomes over 2008-2010; robustness uses stricter winsorization at the 10th and 90th percentiles; an additional robustness subsample excludes individuals with real estate transactions.
  online appendix · Online Appendix Tables A.5-A.6, placebo table notes

• Baseline sample (childless, single, low-educated natives) Sample definition and timing
  Individuals aged 20-29 (age measured in months, weekly panel 2000-2006) who are of Danish origin, have only basic education at the start and end of the year, live alone (not in a couple), have no children and are unmarried at both start and end of year; running variable is age relative to the 25th birthday.
  Registers: BEF, UDDA
  Codes: IE_TYPE == 1; edu (AUDD2015_L1L5_K from HFAUDD) == '10'; FM_MARK == 6; d_child == 0; CIVST == 'U'
  replication code · baseline.do, 'Applying selection criteria' block; bef_red.do
• Social assistance receipt (weekly) Unemployment and benefits
  Weekly indicator equal to 1 if the person's DREAM weekly benefit code is a social assistance code.
  Registers: DREAM
  Codes: DREAM y_01-y_53 in 130-139, 140-149, 700-709, 710-719, 720-729, 730-739
  replication code · baseline.do, 'Defining weekly ordinary employment (wage), earnings (wage2), social assistance (sa)'
• Ordinary employment (weekly) Employment and labour force status
  Weekly indicator equal to 1 if the person has any employment in the corresponding month in MIA (monthly A-income employer reports, linked to CONESR/RAS jobs), set to 0 in weeks with a DREAM code for study grant, wage-subsidy programmes or death.
  Registers: MIA, CONESR, RAS, DREAM
  Codes: wage = 0 if DREAM y in 651,652 (study grant); 135,136,145,146,215,216,705,706,715,716,725,726,735,736,745,746,755,756,765,766,815,816,875,876,895,896 (wage subsidy); 997 (deceased)
  replication code · baseline.do, lines ~165-180; mia_con.do
• Monthly earnings Earnings and income
  Annual job earnings (LONBLB from CONESR 2000-2005, LOENBLB from RAS 2006) for each person-firm spread evenly over the months with MIA employment at that firm; weekly earnings = ordinary employment indicator x monthly earnings.
  Registers: MIA, CONESR, RAS
  Codes: lon_md = lonsum_se / ans_mia_md; wage2 = wage * q
  replication code · mia_con.do
• Full-time/part-time hours code Hours worked
  Monthly hours-of-work category from RAS (HELTID_DELTID_KODE) attached to months of MIA employment at the workplace, used for heterogeneity by hours.
  Registers: RAS, MIA
  Codes: HELTID_DELTID_KODE
  replication code · hoursofwork.do, lines ~22-140
• Criminal charges Crime
  Charges from KRSI 2000-2006 (one per person and case number), dated by charge date SIG_GER1DTO and assigned to age in months; outcome is any/number of charges by age month.
  Registers: KRSI
  Codes: SIG_GER1DTO, SIG_GER7, JOURNR
  replication code · crime.do

Replication package

• Weeks of parental leave Family links, fertility and household
  Total number of parental leave weeks computed from start and end dates of benefit spells in the parental leave spell data (children born 2008-2015); includes all full-time leave taken at most three years after childbirth, separately for mother and father.
  Registers: Parental leave benefit spell data, BFL/eIndkomst
  working paper · Section 3.2 (Data)
• Individual benefit replacement rate Unemployment and benefits
  Computed from total parental leave benefits paid out per parent-child spell relative to weeks of leave; statutory benefits replace 100% of pre-birth earnings for a full-time (37 hours/week) employee up to a cap of EUR 580 per week (2019 level), scaled down proportionally for part-time.
  Registers: Parental leave benefit spell data
  working paper · Sections 2 and 3.2, footnote 11
• Weeks with full wage compensation Unemployment and benefits
  Weeks of leave with employer-provided full wage compensation, computed from the amount of parental leave benefits paid to the employer (benefits are paid to the employer as reimbursement when it pays full wage) using the individual replacement rate.
  Registers: Parental leave benefit spell data
  working paper · Section 3.2
• Imputed individual right to wage compensation Unemployment and benefits
  Public sector: imputed from collective-agreement rules (6-6-6 model, Apr 2008-Apr 2015: 6 weeks each parent plus 6 shared weeks if both parents work in the same public sub-sector, otherwise the shared 6 accrue to the public-sector parent) using both parents' employment sector. Private sector: equals observed compensated weeks for parents who also take uncompensated leave (exhausters); otherwise the mode of compensated weeks among exhausting co-workers with weakly higher compensation, with co-worker groups defined as firm-year cells, or 6-digit then 5-digit industry-year cells when fewer than two exhausting co-workers exist.
  Registers: Parental leave benefit spell data, employer-employee links
  working paper · Section 3.3, eq. (11), Table 3
• Birth sample Sample definition and timing
  Births 2008-2015; parents must be continuously employed in the same firm in the 3 months before birth and not on non-birth-related public benefits during those 3 months; births are excluded if parents have a subsequent child within 15 months; parents must appear in the parental leave data (final sample 189,974 births).
  Registers: BEF, BFL/eIndkomst, Parental leave benefit spell data
  working paper · Section 3.2, Table 1 and notes

• Performed hours of work Hours worked
  Contract hours plus paid overtime, minus paid absence, vacation and weekday holiday hours, summed over all employments in a year per individual (including secondary jobs). Contract hours are available from 2002, so the sample covers 2002-2016. LON covers all public firms and private firms with 100+ employees, plus a sample of private firms with 10-99 employees (about 50% of main employments).
  Registers: LON (Lønstatistikken)
  paper · Section 4 (Data), eq. (5); Appendix A.1
• Three-year log change in performed hours Hours worked
  Main outcome: log change in annual performed hours between base year t and t+3. Strictly positive hours are required in both years. Decompositions condition on staying in the same job versus switching jobs between t and t+3.
  Registers: LON
  paper · Section 3, eq. (1); Section 5.1, Tables 3-4
• Marginal tax rate on labour income Taxes and tax rates
  Simulated using tax programs from Kleven and Schultz (2014), extended by the Danish Ministry of Finance (available 1984-2016), as [T_t(z+100) - T_t(z)]/100, with income information mainly from IND, INDK and INPI. The treatment is the three-year log change in the marginal net-of-tax rate.
  Registers: IND, INDK, INPI
  paper · Section 4 (Data)
• Predicted (mechanical) change in net-of-tax rate instrument Taxes and tax rates
  ln(1 - T'_{t+s}(z_{i,t-k})) - ln(1 - T'_t(z_{i,t-k})): marginal rates under the period t+s and t tax systems evaluated at income lagged k periods, following Weber (2014). The main specification uses k = 2, with lags 2-4 used for overidentification tests.
  Registers: IND, INDK, INPI
  paper · Section 3.1.1, eq. (2); Section 3.2
• Lagged income controls Earnings and income
  Ten-piece spline with breakpoints at the deciles of log labour income lagged twice, to control for mean reversion and heterogeneous income trends.
  Registers: IND, INDK
  paper · Section 3.1
• Estimation sample Sample definition and timing
  Individuals aged 25-65 with all information needed to simulate marginal taxes, who are neither jobless nor outside the labour force as primary occupation at end of November in year t, and have strictly positive hours at t and t+3. Base-year labour income must exceed DKK 250,000 (2015 prices, CPI-deflated) and all controls must be non-missing.
  Registers: LON, IDAN, BEF, IND
  paper · Section 4 (Data), Table A.1
• Workplace, socioeconomic and family controls Firm characteristics and productivity
  Workplace controls from IDAN, socioeconomic and municipal controls from BEF and FAIN, children from FAMILIE, and education from UDDA; industry groups follow Statistics Denmark's standard groupings.
  Registers: IDAN, BEF, FAIN, FAMILIE, UDDA
  paper · Section 4 (Data); Section 3.1 footnote 6

• 225-hour work requirement / sanction Unemployment and benefits
  After 12 months on welfare benefits within the last three years, recipients keep the full benefit only if they worked at least 225 hours in regular non-subsidised jobs in the previous 12 months. Otherwise the benefit is reduced by about DKK 500-1,000 per month depending on family status; after a sanction the hour count resets to zero. Caseworkers may grant exemptions to those unable to work at least five hours a week.
  Registers: Public employment service data (sanctions, exemptions, collected hours)
  working paper · Section 2
• Experimental sample and treatment arms Sample definition and timing
  All welfare benefit recipients in stock on 15 August 2018 who were subject to the work requirement and had received the official notification letter (N = 47,294), randomised into three arms. Treated: up to six monthly messages plus access to the personalised online tool on jobnet.dk. Control I: general reminders only. Control II: neither.
  Registers: Experimental data (jobnet.dk)
  working paper · Sections 3.1-3.2, Table 1 notes
• Cumulated job finding rate, hours and earnings Employment and labour force status
  Cumulated job finding rates by month since the intervention (extensive margin), and total working hours and labour earnings accumulated over the first 12 months after the start (intensive margin), from monthly income and employment records.
  Registers: E-income (eIndkomst)
  working paper · Section 4.2, Figure 3, Table 3
• Full-time vs part-time job Hours worked
  Full-time jobs have average weekly working hours at or above 90% of the full-time equivalent (37 hours per week); part-time jobs are below that.
  Registers: E-income (eIndkomst)
  working paper · Section 4.2, Table 3 note (a)
• Entering other transfer program Unemployment and benefits
  Indicator that the individual receives any other benefit not subject to the work requirement, such as educational support, sickness or disability benefits.
  Registers: DREAM
  working paper · Section 4.3
• Active vs inactive job seeker Unemployment and benefits
  Active job seekers have a positive number of collected working hours at the onset of the intervention, or at least one job application registered in the public employment service portal in the last four weeks; all others are inactive.
  Registers: Public employment service data (collected hours, job applications)
  working paper · Section 4.4
• Local counselling intensity Geography and neighbourhood
  Average number of caseworker meetings per recipient in the municipality in the six months before the intervention, split at the median. A residualised version conditions on observable characteristics.
  Registers: Public employment service data
  working paper · Section 4.4, footnote 20, Appendix Table A.2

• Annual hours worked Hours worked
  Annual hours of regular plus overtime work from Lønstatistikken (LON), reported by employers whose pension contributions depend on hours. Only firms where hours are observed for at least 95 percent of the workforce are kept; for spells in week 48 shorter than a year, hours and earnings are annualized.
  Registers: LON, IDA, FIDA
  working paper · Section 3 (The data)
• Hourly wage Earnings and income
  Annual earnings divided by the sum of regular and overtime hours.
  Registers: IDA, LON
  working paper · Section 3
• Employer link and full-time sample Sample definition and timing
  Each employee is matched to the highest-paying employer in the employment spell of week 48 (FIDA). Sample: full-time employees aged 15-65 in 2003-2011, full-time defined (official definition) as working more than an average of 26 weekly hours over a year, in private firms covered by Firmstat.
  Registers: FIDA, Firmstat, IDA
  working paper · Section 3
• Hours coordination (dispersion of hours across skill groups) Hours worked
  Standard deviation across skill groups in a firm-year of the average annual (regular plus overtime) hours of each skill group; firm measure is its average over 2003-2011. Lower dispersion means greater coordination. Skill groups are deciles of the sum of the fixed and time-varying individual components of an AKM hourly-wage model; alternative: education (primary/secondary/tertiary) x occupation (manager/middle manager/blue collar) cells.
  Registers: LON, IDA
  working paper · Section 4.3, equation (9)
• Simulated marginal tax rate Taxes and tax rates
  Marginal tax rate on labour income simulated from the Danish Tax Register with an extension of the Kleven-Schultz (2014) tax simulator to 2006-2011: increase in tax liability from raising labour income by 100 DKK, holding other income components fixed. Changes are taken over the 2008-2011 interval.
  Registers: Danish Tax Register
  working paper · Section 5.2
• Mechanical net-of-tax rate change (instrument) Taxes and tax rates
  Post-reform (2011) marginal tax rate simulated using pre-reform (2008) income adjusted for inflation; the log change in the marginal net-of-tax rate from this mechanical rate instruments the observed change. For spillovers, the firm average of mechanical changes among high-income workers instruments the change in their average hours.
  Registers: Danish Tax Register
  working paper · Section 5.5
• High- vs low-income workers Taxes and tax rates
  Low-income: tax-exempt or in the bottom tax bracket in 2008. High-income: in the middle or top bracket in 2008, excluding a residual group in the top bracket in 2008 predicted (from 2008 real income and the post-reform schedule) to fall into the bottom bracket in 2011.
  Registers: Danish Tax Register
  working paper · Section 5.1
• Firm productivity Firm characteristics and productivity
  Value added per employee and TFP from Firmstat accounting data; in some specifications value added per employee is instrumented with TFP. Construction details are in an online appendix not included in the WP.
  Registers: Firmstat
  working paper · Section 4.1, footnote 15; Figure notes

• Labour market income (LI) Earnings and income
  Salary, wages, honoraria, fees, bonuses, fringe benefits and business earnings, as defined in the Danish income tax system.
  Registers: Danish tax records
  paper · Section 3.1, Table 1
• Broad income (BI) Earnings and income
  Labour market income plus transfers, certain pension payouts and received alimony; it equals personal income before deductions.
  Registers: Danish tax records
  paper · Table 1; Fig. 3 notes
• Personal income (PI) Earnings and income
  Broad income minus the labour market contribution and certain pension contributions (employer-administered annuity pension contributions were fully deductible until 2010, then capped at DKK 100,000 for expiring annuities). PI is the main outcome and initial-income concept, and responses are also shown with and without pension contributions.
  Registers: Danish tax records
  paper · Table 1; Section 3.1; footnote 21
• Taxable income (TI) Earnings and income
  Personal income plus capital income (interest, rental and business capital income minus interest on debt) minus itemized deductions (commuting, union fees, unemployment insurance contributions, other work expenses, charitable donations, paid alimony). Stock income is taxed separately.
  Registers: Danish tax records
  paper · Table 1
• Simulated marginal tax rate / predicted change in log net-of-tax rate Taxes and tax rates
  Marginal tax rates are imputed with an extension of the Kleven and Schultz (2014) Danish tax simulator, as the increase in total household taxes from an incremental increase in labour income. The predicted change compares rates under the period-t and t-1 tax systems, holding income and characteristics fixed at their t-k levels (k = 1 baseline).
  Registers: Danish tax records
  paper · Section 2 (eq. 5); Section 3.3
• Reform and pre-reform period samples Sample definition and timing
  Four-year periods: 2004-08 (pre) and 2008-12 (reform) for the 2009-10 reform; initial years 1999 and 2003 for the 2004 reform. Individuals present in tax data with positive income over the period, and initial personal income above DKK 250,000 (2009-10 reform) or DKK 200,000 (2004 reform) at 2019 levels deflated by tax-code wage indexes.
  Registers: Danish tax records
  paper · Section 3.3, Table 3
• Initial income bin dummies Earnings and income
  Fifty dummies for initial personal income measured at the start of each period (e.g. 2004 or 2008). The DKK 300,000 bin is omitted, so coefficients measure income-growth trend differentials relative to that bin.
  Registers: Danish tax records
  paper · Section 4, eq. (11)

• Car types (green/brown x heavy/light) Other
  Cars are split by whether weight is above or below the median for the car's vintage cohort, then within each half by whether fuel efficiency is above (green) or below (brown) the median, giving four types; attributes are averaged within type. Cars are grouped into 25 age groups from new to 24+ years (cars over 22 years dropped from estimation).
  Registers: Car ownership register (Statistics Denmark)
  working paper · Appendix F.1 (Car types); Section 6.2
• Household types Family links, fertility and household
  Eight household types from couple vs single, high vs low work distance, and high vs low income, where the income split is at the median income within the demographic cell.
  Registers: Demographic and income registers
  working paper · Appendix F.1 (Household types); Table 2
• High work distance Geography and neighbourhood
  Based on the Danish tax deduction for travel distance to work, which applies only to full-time workers living more than 12 km (each way) from their workplace; households with the deduction are classed as high work distance (slightly under half of Danes).
  Registers: Tax records (commuting deduction)
  working paper · Appendix F.1 (Household types)
• Kilometres driven Other
  Driving measured from odometer readings recorded at mandatory vehicle inspections, which occur biannually from car age 4 (ages 4, 6, 8, 10, ...).
  Registers: Vehicle inspection odometer data
  working paper · Appendix F.1 and footnote 32
• Scrappage Other
  Scrappage is not observed directly: a vehicle is defined as scrapped if an ownership spell ends and no other ownership spell ever begins afterwards (ownership register extract from September 2011, last sample year 2008, so at least about 3 years without an owner). Accidents are not observed.
  Registers: Car ownership register
  working paper · Appendix F.1
• New car registration tax Taxes and tax rates
  T(P) = 1.25 x [tau1 x min(P, K) + tau2 x max(P - K, 0)], with 25% VAT on the price including tax; in 2008 K = DKK 81,000, tau1 = 105% and tau2 = 180%. Fuel prices are country averages from eof.dk.
  Registers: Fuel prices (eof.dk)
  working paper · Appendix F.1
• Estimation sample Sample definition and timing
  Universe of Danish households and all privately owned cars, 1996-2008 (estimation on holdings 1997-2008, about 39 million household-year observations of car ownership transitions).
  Registers: Demographic registers, Car ownership register
  working paper · Appendix F.1; Section 6

• Homeownership (household-based) Wealth, debt and consumption
  Baseline measure from housing records (living in an owner-occupied home) validated against the tax-record measure based on housing wealth available from 1987, for the population above age 20; housing-register measure is slightly lower in level because it misses owners not residing in their owned property. A variant uses only 'clean' cases (defined in main text).
  Registers: Housing register (BBR); tax records (housing wealth, from 1987)
  online appendix · Figure A.I notes
• Home size Wealth, debt and consumption
  Area of the house in square metres used for habitation, from the housing register.
  Registers: Housing register
  working paper · Section 3.2
• Home value Wealth, debt and consumption
  Appraisal of the property by the Danish Tax Authorities made for assessing imputed rental income and taxable wealth.
  Registers: Tax authority property appraisals
  working paper · Section 3.2
• Total deductible interest expenses Wealth, debt and consumption
  Interest paid on mortgage debt, consumer debt and other loans (all deductible), for homeowners; mortgage interest cannot be separated over the whole period.
  Registers: Income tax records
  working paper · Section 3.2
• Household tax bracket / treatment status Taxes and tax rates
  A taxpayer is assigned to the top bracket if she or her spouse is in the top bracket, to the middle bracket if either is in the middle and neither in the top, and to the bottom bracket if both are in the bottom. Treated are those in the top bracket in every year 1984-1986; controls are in the middle or bottom bracket in every year 1984-1986.
  Registers: Income tax records, Population register spouse links
  working paper · Section 4
• Net-of-tax interest rate Taxes and tax rates
  (1 - tau) R, where tau is the marginal tax rate on negative capital income; the 1987 reform set this rate to 51 percent in all brackets (pre-reform top about 73, middle 62, bottom 48 percent in the average municipality), raising the net-of-tax rate by 81 percent for the top group, 29 percent for the middle and reducing it 6 percent for the bottom.
  working paper · Section 3.1, Figure 1
• Semi-balanced panel 1980-1996 Sample definition and timing
  Individuals observed in every year 1980-1996 while aged 20-80 (entering at 20, leaving after 80), giving 7 pre-reform and 10 post-reform years; about 40 million person-years.
  Registers: Population register
  working paper · Sections 3.2 and 4
• Long-run balanced panel 1980-2011 Sample definition and timing
  Very-long-run home size analysis uses a balanced panel of individuals observed in all years 1980-2011 (vs. main semi-balanced panel 1980-1996). Pre-trend robustness: group-specific linear trends estimated on 1983-1986 and residualized out.
  Registers: Housing register; population register
  online appendix · Figures A.II-A.IV notes

• Household (tax unit) and married status Family links, fertility and household
  Household id is the family id when the person's spouse id equals the cohabitation partner id and civil status is in the married range (up to 2007, from FAIN); from 2008 from BEF using spouse id = partner id and civil status G or P. Persons not so linked form their own household. Only persons aged 20+ are kept and wealth is summed within household.
  Registers: FAIN, BEF
  Codes: FAIN (<=2007): aegtefaelle_id == c_faelle_id, c_status 2-5, hhid = c_familie_id; BEF (>=2008): aegte_id == efalle, civst in {G, P}; alder >= 20
  replication code · Empirical Analysis/DD_make_hh_ids.do
• Household taxable wealth (outcome) Wealth, debt and consumption
  Household net taxable wealth = sum over household members aged 20+ of the wealth variable FORM from the income-tax register (INDH); outcome is log(FORM). Portfolio shares use INDK components: equity share = listed stocks / total assets, housing share = (mortgage deeds + real estate value) / total assets.
  Registers: INDH, INDK
  Codes: FORM (INDH); share_tx_equities = KURSAKT/QAKTIVF; share_tx_housing = (KURSPANT+KOEJD)/QAKTIVF; other INDK fields: FORMSK, LOFTFORM, QPASSIV, PRIGALD, KONTANT, KURSOBL, INDESTPI
  replication code · Empirical Analysis/DD_base_data.do, section 1
• Wealth tax exemption threshold and marginal wealth tax Taxes and tax rates
  Annual exemption threshold (allowance) by year; for couples it equals 2 x single allowance, except 1989-1992 when couples' allowance uses a separate married allowance x 2. Household marginal wealth tax rate tau_it = statutory rate if FORM exceeds the household allowance, else 0. Bunching analysis uses distance b = FORM - allowance, deflated, in DKK 20,000 bins from -540,000 to 540,000.
  Registers: INDH
  Codes: Allowance (DKK): 1980-81 1,000,000; 1982 1,083,000; 1983 1,158,900; 1984 1,204,900; 1985 1,229,200; 1986 1,253,500; 1987 1,278,700; 1988 1,329,900; 1989 1,383,600; 1990 1,424,500; 1991 1,452,700; 1992 1,482,100; 1993 1,534,500; 1994 1,580,500; 1995 1,592,000; 1996 2,400,000; married: 1989-90 811,500x2, 1991 1,249,600x2, 1992-96 2x single
  replication code · Empirical Analysis/DD_base_data.do section 2.2; bunching_base_data.do
• Ceiling DD treatment (unbound by tax ceiling) Taxes and tax rates
  X = wealth tax actually paid (FORMSK) / theoretical wealth tax ((FORM - allowance) x tax rate). Treatment: household in top 1% of the wealth distribution with X > 0.99 in every pre-reform year of the window (e.g. 1982-1988), i.e. not bound by the tax ceiling; control: top 1% with X < 0.80 (bound by ceiling). Households above the 99.99th percentile excluded. In the panel, status is re-coded post-reform using X, LOFTFORM > 0 and wealth relative to allowance.
  Registers: INDK, INDH, FAIN
  Codes: treatmt_2e_X80: X > .99 and top1 in 1982-1988; control_2e_X80: X < .80 and top1; drop form_pctile > .9999
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 3.4; DD_base_data.do section 2.3
• Couples DD treatment Taxes and tax rates
  Treatment: married households with wealth between the (single) allowance and 2 x allowance in each pre-reform year (1988 and earlier years depending on window), which became exempt when the couples' threshold doubled in 1989; controls: (c1) singles with wealth in the same band, (c4) married couples in the top 5% with wealth below the allowance, or (c5) singles with wealth between allowance/2 and allowance.
  Registers: INDH, FAIN
  Codes: treatmt_c1*: married & allowance <= FORM <= 2*allowance; control_c1*: unmarried, same band; control_c4*: married & FORM < allowance & top5; control_c5*: unmarried & allowance/2 <= FORM <= allowance
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 3.3
• Balanced panel 1980-1996 Sample definition and timing
  Households (persons 20+ aggregated by household id) observed with strictly positive aggregate wealth FORM in all 17 years 1980-1996.
  Registers: INDH, FAIN
  Codes: drop if form <= 0 | missing; keep if _N == 17 (balanced_1980_96)
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 2
• Self-employment / business income indicator (bunching) Earnings and income
  Household flagged as self-employed if business profit (NETOVSKUD_GL) is nonzero; a broader 'ext' flag also uses nonzero values of EGNKAPVK, UDLFKURS, UDLIND2 or UDLIND3 (foreign/business-related components).
  Registers: INDH, INDK
  Codes: selv = netovskud_gl != 0; ext = netovskud_gl!=0 | egnkapvk!=0 | udlfkurs!=0 | udlind2!=0 | udlind3!=0
  replication code · Empirical Analysis/bunching_base_data.do

Replication package

• Treated (non-EU) immigrants Immigration and origin
  First-generation immigrants from outside the EU and the four EFTA countries (Iceland, Liechtenstein, Norway, Switzerland), who are subject to the immigrant welfare scheme. 'Inside the EU' is defined by member countries in the pre-scheme year 2001, to avoid confounding from later EU enlargements.
  Registers: CPR, BEF
  Codes: EU membership as of 2001 + EFTA
  working paper · Section 2.1 and footnote 6
• Net immigration flow Immigration and origin
  Year-to-year change in the stock of first-generation immigrants residing in Denmark according to the Central Person Register, by group, scaled by the group's pre-scheme (2001) immigrant stock and residualized on a linear group-specific trend estimated on 1980-2001.
  Registers: CPR
  working paper · Section 3, footnote 11 and eq. (1)
• Inflow by residence permit type Immigration and origin
  Immigration flows by type of residence permit granted (asylum, family, work, study; available from 1997). Asylum and family permit holders are treated as potentially welfare-eligible; work/study permit holders as untreated. Asylum applications are used as a supplementary measure.
  Registers: Residence permit data (Statistics Denmark)
  working paper · Section 3, Figure 3, footnote 12
• Welfare benefit groups and maximum monthly benefit Unemployment and benefits
  Maximum monthly cash benefits (2018 USD) for non-employed households at age 40 out of work for three months; with children, the first child is aged 10 and the second aged 8. Danish scheme benefits from STAR (Danish Agency for Labour Market and Recruitment) website; other benefits computed with the OECD tax-benefit calculator. Only cash welfare, excluding UI and in-kind benefits such as housing support.
  Registers: STAR scheme rules, OECD tax-benefit calculator
  online appendix · Online Appendix, Figure A.II notes
• Age restriction (24-Year Rule) Sample definition and timing
  Main within-country specification restricts to immigrants at least 30 years old, to avoid confounding from the 2002 '24-Year Rule' requiring both spouses to be at least 24 for marriage-based residency. The cross-country analysis has no age restriction.
  Registers: CPR, BEF
  working paper · Section 2.1 and footnote 7
• Synthetic Denmark (cross-country control) Immigration and origin
  Weighted average of Finland (0.468), Norway (0.361) and Sweden (0.171), chosen to match 1991-2001 non-EU/EFTA net immigration flows normalized by each country's 2001 non-EU/EFTA immigrant stock. Immigrants from the former Yugoslavia are dropped throughout to remove the Bosnia shock.
  Registers: Nordic national statistics bureaus, OECD International Migration Database
  working paper · Section 3 (cross-country strategy), footnote 18; Appendix Figure A.IV notes
• First- and second-generation immigrant Immigration and origin
  Official Statistics Denmark definitions: a first-generation immigrant is born outside Denmark with neither parent being both a Danish citizen and born in Denmark; a second-generation immigrant (descendant) is born in Denmark with neither parent being both a Danish citizen and born in Denmark.
  Registers: CPR, BEF
  online appendix · Online Appendix, Figure A.III notes
• Asylum applications Immigration and origin
  Asylum applications to Denmark from outside the EU at annual and quarterly frequency: applications submitted (available from 2000), applications registered for administrative review, and residence permits granted to asylum seekers; asylum seekers from Bosnia dropped. An inflow measure without a control group, used to validate timing of responses.
  Registers: Statistikbanken tables VAN5 and VAN77
  Codes: Statistikbanken VAN5, VAN77
  online appendix · Online Appendix, Section B and Figures A.IV-A.V

• Household (unit of analysis) Family links, fertility and household
  A household is either a single person aged 20 or above or a married couple, in both cases with dependent children if any; fractiles are defined relative to the total number of households, including those not liable to wealth tax.
  Registers: Administrative wealth and income records
  online appendix · Online Appendix A.1.2
• Household net wealth (ranking variable) Wealth, debt and consumption
  100% of national-accounts household wealth at market value distributed to households: all non-financial and financial assets minus debts, including funded pensions, excluding durables and valuables. Pension wealth imputed (40% to wage earners pro rata wage income, 60% to retirees pro rata pension income); unlisted equity imputed by capitalizing unlisted dividends and non-corporate business assets by capitalizing business income. Danish distribution taken from Jakobsen et al. (2018), based on individual wealth records.
  Registers: Danish individual wealth records, National accounts
  working paper · Section 3.2; Online Appendix A and D
• HSBC undeclared account holder Taxes and tax rates
  Households who owned at least one account at HSBC Switzerland (including via shell companies with identified beneficial owners), declared themselves taxable in Scandinavia in 2006 and could be matched to a tax return (2006/2007 records); non-residents excluded; for Denmark all matched households are treated as evaders (about 95% of matched Danish accounts were undeclared).
  Registers: Leaked HSBC list (Danish tax authority version), Tax returns
  working paper · Section 3.1.1, footnotes 13-16; Online Appendix E
• Random-audit detected evasion Taxes and tax rates
  SKAT stratified random audits for tax years 2008, 2010 and 2012 (18,985 taxpayers; self-employed and complex returns oversampled), every line item examined; main results include all errors; deliberate evaders are those rated 0-2 on SKAT's 0-6 compliance scale. Evasion rate = taxes evaded / (taxes paid + taxes evaded) by wealth group, weighted by audit weights.
  Registers: SKAT random audit data
  Codes: SKAT compliance rating 0-2 = deliberate evasion
  online appendix · Section 4 (Random Audit Data); Online Appendix H.1-H.2

Replication package

• Household marginal interest rate (2008) Wealth, debt and consumption
  Interest on marginal liquidity, computed from register data on all loan and deposit accounts of the household obtained from the Danish Tax Agency (SKAT), measured in 2008, in percent.
  Registers: Danish Tax Agency account-level loan and deposit data
  online appendix · Online Appendix Table EA.1 notes
• Propensity to spend the SP payout (survey) Wealth, debt and consumption
  Survey (Epinion) question: considering the increase in disposable funds from the SP payout, how did you allocate it: 1) increase spending, 2) increase free savings, 3) reduce debt, 4) increase pension savings; spending share in [0,1] (96% of answers 0 or 1). Main sample includes only respondents who withdrew the SP balance (N = 5,037); extended sample adds 215 non-withdrawers coded as zero spending. Probit robustness codes share >= 0.5 as 1.
  Registers: Telephone survey linked to registers on CPR
  Codes: Probit: 0 if share in [0,0.5], 1 if in [0.5,1]
  online appendix · Online Appendix sections 4, 7, 8 (survey instruments)
• SP payout / SP account balance Retirement and pensions
  SP payout is the (potential) payout, i.e. the Special Pension account balance less taxes; enters as SP payout / 2008 disposable income (percent).
  Registers: ATP Special Pension (SP) account register
  online appendix · Online Appendix Table EA.1 notes
• Register net savings Wealth, debt and consumption
  Change in net bank assets (bank deposits minus bank debt, including credit card debt, recorded on the last day of the year) between t and t-1 from the income-tax/wealth register; used to validate survey answers for 2005-2011.
  Registers: Income tax register (SKAT wealth data)
  working paper · Section 5.1
• Imputed spending Wealth, debt and consumption
  Disposable income minus net savings (change in net bank assets) minus private pension contributions, c = y - dW - p, following Browning and Leth-Petersen (2003).
  Registers: Income tax register
  working paper · Section 5.1
• Low liquid assets dummy Wealth, debt and consumption
  Indicator equal to one if the individual holds liquid assets worth less than one month's disposable income (threshold chosen for highest predictive power); a continuous liquid-assets-to-disposable-income ratio is also used.
  Registers: Income tax register
  working paper · Section 5.3.2
• High debt service dummy Wealth, debt and consumption
  Indicator equal to one if debt service is 15 percent or more of disposable income; also used as a continuous debt-service-to-disposable-income ratio.
  Registers: Income tax register, SKAT account-level loan records
  working paper · Section 5.3.2
• Holding a deposit balance Wealth, debt and consumption
  Individuals with total deposit-account balances exceeding one month's disposable income are labelled as having a balance (to net out salary accounts credited just before 31 December); used to define the group holding both assets and debt.
  Registers: SKAT account-level deposit records
  working paper · Section 5.3.1
• Liquid assets / disposable income Wealth, debt and consumption
  Liquid assets (deposits, shares and bonds) in 2008 as percent of 2008 disposable income; underlies the low-liquid-assets indicator.
  Registers: Statistics Denmark income and wealth registers
  online appendix · Online Appendix Table EA.1 notes
• Debt service / disposable income Wealth, debt and consumption
  Debt service, defined as total interest payments on loans, in 2008 as percent of 2008 disposable income; underlies the high-debt-service indicator.
  Registers: Statistics Denmark income registers
  online appendix · Online Appendix Table EA.1 notes
• Permanent income Earnings and income
  Average of (register) income over 1998-2008, entered in logs; income controls also include ln(2008 register income) and the standard deviation of ln(income).
  Registers: Statistics Denmark income registers
  Codes: Years 1998-2008
  online appendix · Online Appendix Table EA.1 notes
• Education level dummies Education and test scores
  Short, medium and long education correspond to vocational training, college BA level, and college Master and PhD level, respectively; reference group is high school or less.
  Registers: Statistics Denmark education register
  Codes: Short = vocational; medium = BA; long = Master/PhD; ref = high school or less
  online appendix · Online Appendix Table EA.1 notes
• Income and credit expectations (survey) Other
  Survey questions on whether total 2009 income evolved better/worse than expected (5-point scale plus don't know), changes in expected 2010 income, and changes in expectations about the possibility of borrowing money in 2010 (5-point scale from much more optimistic to much more pessimistic).
  Registers: Telephone survey
  Codes: 5-point Likert scales
  online appendix · Online Appendix section 8 (survey instruments)

• Core student sample Sample definition and timing
  Individuals aged 18-35 enrolled in tertiary education, fully eligible for student benefits (SU), and with positive labour earnings, defined as earned income above DKK 8,000 per year. This gives about 75,500 observations per year, 2006-2011.
  Registers: Danish administrative registers (education, tax returns), Student Benefit Administration records
  working paper · Section 3.2, Table 2
• Income counted against the SU income limit Earnings and income
  Labour income (net of labour market contribution), transfers other than student benefits, and capital income, except certain types of stock income. All amounts are scaled to 2018 levels using the official wage indexation of student benefits.
  Registers: Tax returns
  working paper · Section 3.1, footnote 4
• Individual income limit Unemployment and benefits
  The baseline income limit is 12 times the monthly basic amount (DKK 7,700, 2018 levels; about DKK 93,000 annually) and rises by DKK 11,600 for each month of benefits not taken up. It is computed from monthly benefit take-up records of the Student Benefit Administration.
  Registers: Student Benefit Administration records
  working paper · Section 3.1, Figure 1 notes; Appendix C
• Effective marginal tax / phase-out rate Taxes and tax rates
  Marginal income tax of 42% above a personal allowance of DKK 49,900. Benefits are phased out at 50% for the first DKK 11,600 above the individual income limit and at 100% beyond that, so the effective marginal rate jumps from 42% to 72% at the baseline limit under optimal take-up. If the payback exceeds DKK 9,200, the whole payback rises by 7% (a notch). The 2009 reform raised the baseline limit by 24% and the phase-out rate from 52% to 62%.
  working paper · Section 3.1, footnotes 5 and 11, Figure 2
• Excess income and predicted payback Earnings and income
  Total earned income from tax records minus the individual income limit; positive amounts translate into payback of benefits. Predicted end-of-year benefits (initial payout minus predicted payback) match reported benefits within DKK 2 for 97.9% of the core sample.
  Registers: Tax returns, Student Benefit Administration records
  working paper · Section 4.3; Appendix C, Table C.I
• Study-related job Sector and industry
  All jobs not in wholesale (NACE 46), retail (47), postal service (53), hotels and restaurants (55, 56), cleaning (81), elderly, social and childcare (87, 88), or sports and amusement facilities (93).
  Registers: Monthly earnings records (employer/firm link)
  Codes: NACE 46, 47, 53, 55, 56, 81, 87, 88, 93 = not study-related
  working paper · Table 3, note e
• Monthly earnings and hours Hours worked
  Monthly records of earnings and hours worked for all wage earners, available from 2008 and used for average weekly hours and the within-person variability of monthly earnings.
  Registers: Monthly payroll records (from 2008)
  working paper · Section 3.2, Table 3, Appendix Figure A.II
• Student experience Education and test scores
  Number of years enrolled in tertiary education between 2004 and 2007 (that is, over the previous four years).
  Registers: Education registers
  working paper · Table 3 note a; Appendix Figure A.VI

• Monthly wage change relative to 2008 Earnings and income
  Percentage change in individual monthly gross wage income (year t, month m) relative to the individual's average monthly gross earnings in 2008; percentage changes are used instead of logs because monthly earnings may be zero or near zero.
  Registers: eIncome register (SKAT)
  working paper · Section II, eq. (1)
• Treatment and control groups (top-tax incentive) Sample definition and timing
  Robustness variants of the control group by average monthly 2008 earnings: narrow C-group DKK 32,000-35,000, wide C-group DKK 24,000-35,000; doughnut sample excludes individuals with average monthly 2008 earnings DKK 33,000-37,000; alternative baseline year 2009 (groups and income differences relative to 2009); alternative post-reform year 2011. T-group by percentile bins of average monthly earnings: [P80,P90), [P90,P95), [P95,P99), [P99,+].
  Registers: eIncome monthly payroll register (SKAT)
  Codes: Narrow C: 32,000-35,000 DKK/month; wide C: 24,000-35,000 DKK/month; doughnut excl. 33,000-37,000 DKK/month (2008)
  online appendix · Online Appendix Table A2 notes; Figure A1
• Analysis sample Sample definition and timing
  Individuals employed throughout 2008 (a wage record in every month of 2008) with positive income in all years, observed January 2008 - December 2011; includes people who are both wage earners and self-employed. Sensitivity: excluding employees with any exit to unemployment during 2009-2010.
  Registers: eIncome register (SKAT)
  working paper · Section II, footnote 5; footnote 6
• Income-shifting indicator Earnings and income
  Equals one if income in a month is high (more than 50 percent above) relative to its level in the same month of 2008 and income in the preceding month is low (more than 50 percent below) relative to its 2008 level; zero otherwise. Robustness with 25%-25% and 75%-75% cutoffs; a stricter version also requires November 2009 income to be unusually low.
  Registers: eIncome register (SKAT)
  working paper · Section III.A, eq. (3) and footnote 6
• Financial capacity (liquidity) Wealth, debt and consumption
  Financial assets (money in bank accounts and value of shares and bonds) at end of 2008 relative to 2008 annual disposable income; groups below one month, between one and two months, and above two months of disposable income.
  Registers: Background registers at Statistics Denmark
  working paper · Section III.B, Figure 8
• Firm size groups and top-five paid Firm characteristics and productivity
  Firm size groups: fewer than 25, 25-99, 100-499 and 500+ employees; large firms defined as more than 100 employees (top decile by number of employees); top-five paid employees within each firm used to isolate top management.
  Registers: eIncome register (SKAT), Firm registers
  working paper · Section III.B, Figures 6-7
• Awareness of shifting opportunity Other
  Survey respondents are classified as informed about the opportunity to shift income if they answered Q1 = 'After New Year' and Q2 = 'Legal' in the Capacent Epinion telephone survey (588 T-group and 290 C-group respondents).
  Registers: eIncome monthly payroll register (SKAT) linked to telephone survey
  Codes: Q1 = 'After New Year' & Q2 = 'Legal'
  online appendix · Online Appendix Figure A2 notes
• Covariate set for ETI regressions Sample definition and timing
  Controls: age dummies, gender dummy, marriage dummy, 5 regional dummies, 10 sector dummies (as in paper Table 2) and 4 company-size dummies (as in paper Figure 7).
  Registers: Statistics Denmark administrative registers
  Codes: 5 regions; 10 sectors; 4 firm-size groups
  online appendix · Online Appendix Table A1 notes

• Workplace (occupation-firm cell) Firm characteristics and productivity
  Workplace = all employees with the same 2-digit occupation code in the same firm; employer default contribution rate measured by the workplace-level median (robustness: workplace-level mode, using only workplaces with a unique mode).
  Registers: IDA; income tax register
  Codes: 2-digit occupation code x firm ID
  online appendix · Online Appendix A
• Employer contribution rate (default) Retirement and pensions
  Employer-sponsored pension contributions divided by taxable labor income, computed separately for capital (lump-sum) and annuity pension accounts; the workplace-level measure is the median contribution rate within the occupation-firm cell in a year (modes used as robustness). Main outcome is the change from t-1 to t.
  Registers: Income tax register (third-party reported pension contributions)
  working paper · Section 2.2; Section 3 eq. setup; Appendix A
• Share of workplace above top tax threshold Taxes and tax rates
  Fraction of employees in the occupation-firm cell with labor income above the top income tax threshold (about 49 percent of workers on average), interacted with year dummies with 1998 as baseline; the 1999 reform cut the deduction value for capital pension contributions from 59 to 45 cents per DKr for top-bracket taxpayers.
  Registers: Income tax register
  working paper · Section 2.1; Section 2.3; Section 3
• White-collar occupation Occupation and job hierarchy
  White-collar = 2-digit (D)ISCO occupation code with first digit 1-5 (1 leadership; 2 highest-level knowledge work; 3 intermediate knowledge; 4 office jobs; 5 retail sales, service and care work). Blue-collar/other = first digit 6-9. Main sample excludes public-sector workers and blue-collar occupations to avoid workplaces with collectively bargained contribution rates.
  Registers: IDA (occupation code)
  Codes: 2-digit DISCO first digit 1-5 = white-collar; full list 11,12,21-24,31-34,41,42,51,52 (white-collar); 61,71-74,81-83,91-93
  online appendix · Online Appendix E: Occupation Codes
• Analysis sample Sample definition and timing
  Employees in private-sector firms with white-collar occupations; excludes self-employed individuals and workplaces with fewer than five employees; years 1996-2001.
  Registers: Income tax register; IDA
  Codes: workplace size >= 5
  online appendix · Notes to Appendix Figures 1-2
• Individual private (IRA-type) pension contributions Retirement and pensions
  Individual contributions to private capital and annuity pension accounts outside the employer plan, as contribution rates relative to taxable labor income, used as the comparison to employer responses.
  Registers: Income tax register
  working paper · Section 2.3; Section 4
• Robustness subsample: no annuity contributions Sample definition and timing
  Workplaces whose median annuity contribution rate in 1996-1998 was zero and whose median capital (lump-sum) account contribution rate in 1998 exceeded 1.5 percent (4,056 observations).
  Registers: Income tax register (third-party reported pension contributions)
  Codes: median capital contribution rate 1998 > 1.5%
  online appendix · Online Appendix A, item 4
• Highly educated workplace Education and test scores
  Private-sector white-collar workplaces in which more than 20 percent of workers had more than 16 years of education (used as a group less likely covered by collective bargaining).
  Registers: IDA
  Codes: share with >16 years education > 20%
  online appendix · Online Appendix D

• Estimation sample Sample definition and timing
  Workers aged 18-65 employed in private-sector workplaces in 2005, drawn from IDA for 2004-2006; employment at each workplace is observed in a specific week in each year. Workers with hourly wage rates in the top percentile (above DKK 714) or below DKK 90 are excluded.
  Registers: IDA
  working paper · Section 3.1 (Data)
• Workplace Firm characteristics and productivity
  A workplace is an IDA establishment defined by a geographical location; most firms have one workplace but some larger firms have several.
  Registers: IDA
  working paper · Section 3.1, footnote 9
• Hourly wage rate (IDA) Earnings and income
  Annual labour income (including mandatory pension fund payments) divided by annual hours; annual hours are based on the mandatory ATP pension contributions, so overtime is not counted and part-time hours are imprecise.
  Registers: IDA
  working paper · Section 3.1; Section 5 (robustness)
• Workplace average wage Earnings and income
  Each worker is assigned the average hourly wage rate of employees at his/her workplace in 2005 (following Christensen et al. 2005), after trimming wages below DKK 90 and above DKK 714.
  Registers: IDA
  working paper · Section 3.1
• Workplace size and stayers Firm characteristics and productivity
  Size is the number of employees at the workplace in 2005; stayers are workers employed at the workplace in 2005 who are still employed there in 2006.
  Registers: IDA
  working paper · Section 3.1, Table 1
• New hires / new hires from non-employment Employment and labour force status
  New hires are workers employed at the workplace in 2005 who were not employed there in 2004; new hires from non-employment are workers employed at the workplace in 2005 who were not employed at any workplace in 2004 (non-employment or non-participation).
  Registers: IDA
  working paper · Section 3.1
• Wage offer distribution Earnings and income
  Distribution of firm-specific average wages for persons entering firms in 2005 from non-employment in 2004, weighting each firm's wage by its share of all hires from non-employment.
  Registers: IDA
  working paper · Section 3.1
• Effective marginal tax rate Taxes and tax rates
  Effective marginal tax rates by annual labour income from the Kleven and Kreiner (2006) empirical tax function for 2006 (income taxes, indirect taxes and benefit claw-back), deflated to 2005 by average labour income growth; hourly wages are converted to annual income using median annual hours of 1,660. The participation tax rate m0 is 0.59 at baseline (0.82 in a robustness check).
  working paper · Section 3.1, Figure 1; Section 5
• Hours-adjusted wage rate (WSR imputation) Earnings and income
  Robustness check: the employer-reported hourly wage from the Wage Statistics Register (Lønstatistikken) is regressed on the IDA wage rate after dropping the top and bottom 1% and IDA wages flagged as low quality by Statistics Denmark. The fitted relationship is then used to impute an overtime- and part-time-adjusted wage for all workers.
  Registers: IDA, Lønstatistikken (WSR)
  working paper · Section 5 (robustness), Figure 5

• Labor income (LI) Earnings and income
  Tax-code labor income: salary, wages, honoraria, fees, bonuses, fringe benefits and business earnings. Main outcome is the three-year growth rate (s to s+3) in real labor/wage earnings; regressions weighted by labor income and restricted to positive labor income.
  Registers: Income Tax Register, IDA
  paper · Table 1; Section II; Table notes for labor income elasticity tables
• Capital income (CI), positive and negative Earnings and income
  Interest income, rental income and business capital income minus interest on debt (mortgage, bank loans, credit cards, student loans), excluding stock income; analysed separately as positive and negative capital income because they are taxed very differently.
  Registers: Income Tax Register
  paper · Table 1; Section I; footnote 6
• Personal, stock and taxable income (PI, SI, TI) Earnings and income
  Personal income PI = LI + transfers, grants, awards, gifts, received alimony minus labour market contribution and certain pension contributions; stock income SI = dividends and realized capital gains on shares; deductions D = commuting, union fees, UI contributions, other work expenses, charity, paid alimony; taxable income TI = PI + CI + SI - D (differs from the tax-code label, which excludes SI).
  Registers: Income Tax Register
  paper · Table 1
• Simulated marginal tax rate Taxes and tax rates
  Marginal tax rate on income component j computed with the authors' own Danish tax simulator (1984-2005) as [T(z_j + 100) - T(z)]/100, i.e. by raising that component by DKK 100; accounts for spousal jointness of exemptions and the marginal tax ceiling.
  Registers: Income Tax Register
  paper · Section II; footnote 9
• Mechanical change in net-of-tax rate (instrument) Taxes and tax rates
  log(1 - tau_{s+3}(z_s)) - log(1 - tau_s(z_s)): change in the marginal net-of-tax rate simulated by applying year s+3 tax law to base-year s incomes; analogous mechanical change in virtual income instruments for the change in log virtual income.
  Registers: Income Tax Register
  paper · Section III.B
• Stacked three-year difference sample Sample definition and timing
  Stacked three-year intervals 1984-1987 to 2002-2005, keeping taxpayers also observed in year s-1 (used for pre-reform income controls: splines of log s-1 income and the log deviation between s-1 and s incomes); restricted to ages 15-70, fully tax liable in Denmark, excluding those whose base-year income comes primarily from welfare benefits (about 37 million observations).
  Registers: Income Tax Register, IDA
  paper · Section II; Section V table notes
• Taxpayers close to kink points Taxes and tax rates
  Taxpayers with income within DKK 5,000 of the top kink, DKK 3,000 of the middle kink or DKK 2,000 of the bottom kink (excluded in robustness checks).
  Registers: Income Tax Register
  paper · Notes to labor income elasticity table (Section V)
• 1987 reform DD treatment/control groups Sample definition and timing
  Treatment group: taxpayers experiencing an increase in the marginal net-of-tax rate on labor income (or on positive capital income) due to the 1987 reform over 1986-1989; control group: those experiencing a reduction; large-treatment group has net-of-tax rate increases of at least 15 percent.
  Registers: Income Tax Register
  paper · Section IV, Figure 4 discussion

• Scheme earnings eligibility threshold Taxes and tax rates
  Foreigners' scheme requires annual wage earnings with the requesting employer above a threshold that grows roughly with average earnings and lies between the 99.2th and 99.4th percentile of the Danish wage earnings distribution (DKK 765,600 in 2009); it must be met on an annualized basis (e.g. a 6-month contract needs half the annual threshold). For pre-1991 years the threshold is imputed by holding the threshold-to-average-earnings ratio at its 1991 value, with average earnings computed over all workers with positive earnings.
  Registers: Administrative tax and payroll records
  working paper · Section 2.1 and footnote 14
• Scheme eligibility (prior tax liability) Taxes and tax rates
  To be eligible the taxpayer must have been recruited abroad and not have been tax liable in Denmark in the 3 years before going on the scheme (applied to each scheme spell separately; prior scheme time not counted); citizenship plays no formal role, so Danish citizens tax-resident abroad for 3+ years qualify. Scheme lasts at most 36 months.
  Registers: Foreigners' tax scheme beneficiary records
  working paper · Section 2.1 and footnote 12
• Annualized earnings Earnings and income
  Earnings of part-year residents are annualized based on duration of stay in the year, with duration of stay measured from the migration database (daily dates of entry and exit from CPR records).
  Registers: Administrative tax records, CPR migration records
  working paper · Section 4.1, footnote 23
• Treatment and control groups of foreigners (migration DiD) Immigration and origin
  Treatment: foreigners with (annualized) earnings above the eligibility threshold; controls: foreigners with earnings between 80% and 90% (control 1) and between 90% and 99.5% (control 2) of the threshold, series normalized to 1990. In the elasticity regressions controls are foreigners between 80% and 99% of the threshold; an alternative uses the 99.5-100th percentile vs the 95-97th and 97-99th percentiles of the full earnings distribution.
  Registers: Administrative tax records, CPR citizenship records
  working paper · Section 4.1, Figures 4-5, Table 2
• Citizenship groups and expatriates Immigration and origin
  Nordic foreigners: citizens of Finland, Iceland, Norway, Sweden; English-speaking: Australia, Canada, Ireland, New Zealand, South Africa, UK, US. Expatriates are Danish citizens returning after a spell of at least three years abroad.
  Registers: CPR citizenship records
  working paper · Section 4.1, Table 2 Panels A5-A7
• Wage-effect repeated cross-section sample Sample definition and timing
  For year t, foreigners who arrived in Denmark in year t-1 and stayed the full calendar year t; earnings between 70-95% (below) and 105-400% (above) of the threshold, excluding 95-105% (bunchers) and above 400%; years 1991-1994 excluded because scheme earnings are not computerized.
  Registers: Administrative tax records, CPR migration records, Scheme beneficiary records
  working paper · Section 4.2, Figure 8, Table 3, footnote 30
• Top natives and top foreign employees (spillovers) Earnings and income
  Top natives: native workers with earnings above 75% of the scheme eligibility threshold (about the top 3% of native earnings). Top foreign employees: foreign employees with less than 3 years of presence in Denmark and annualized earnings above the threshold; industry scheme use = growth in top foreign employees between 1980-1990 and 1991-2000 divided by top native employees in 1980-1990.
  Registers: Administrative tax and payroll records, Employer links with 2-digit industry
  working paper · Section 4.3 (industry-level spillovers), footnote 35

• Primary analysis sample Sample definition and timing
  Danish population 1995-2009 from merged income tax register, population register and IDA, excluding observations below age 18 or above 60 and observations with self-employment income (about 41 million observations on 4 million individuals).
  Registers: IND (income tax register), BEF, IDA
  working paper · Section III (Sample and Variable Definitions)
• Taxable saving (gross and net) Wealth, debt and consumption
  Gross taxable saving = change in third-party reported end-of-year assets in taxable accounts; net saving = gross saving minus change in liabilities, where liabilities cover secured and unsecured debt except home mortgages. Saving measures and rates below the 1st or above the 99th percentile in the estimation sample are set to missing.
  Registers: Third-party reported wealth data (tax authority)
  working paper · Section III
• Pension contributions by account Retirement and pensions
  Annual employer and individual contributions to capital pension (lump-sum) and annuity pension accounts as reported by employers and pension funds; contributions above the 99th percentile are top-coded at the 99th percentile in each account and year.
  Registers: IND (income tax register)
  working paper · Section III
• Income concepts Earnings and income
  Labor income = labor earnings before taxes excluding contributions to employer-administered pensions; total compensation = labor earnings plus employer and employee contributions to employer-administered accounts; taxable income = labor income plus other income (e.g., UI) affecting tax liability. Measured at the individual level, with joint-account balances split equally.
  Registers: IND (income tax register)
  working paper · Section III
• Firm switch Employment and labour force status
  Individual switches firms in year t if he has earnings from two distinct firms in t and t-1 (primary job for multiple jobholders); switches where earnings fell by more than half or more than doubled are excluded; set missing in 2003, 2005 and 2007 when firm identifiers changed. Event studies require at least 4 years of data before and after.
  Registers: IDA, IND
  working paper · Section IV.A, footnotes 9-10
• Mandatory Savings Plan (MSP) contribution Retirement and pensions
  From 1998 individuals with labor income at or above DKr 34,500 were required to contribute 1% of earnings (DKr 345 at the cutoff); RD on changes in pension contributions/saving from 1997 to 1998 for individuals within DKr 25,000 of the cutoff.
  Registers: IND (income tax register)
  Codes: cutoff Y >= DKr 34,500; bandwidth DKr 25,000
  working paper · Section IV.B, equation (6)
• 1999 capital pension subsidy reduction Taxes and tax rates
  From 1999 the deduction for capital pension contributions fell from 59 to 45 cents per DKr for individuals in the top income tax bracket (cutoff DKr 251,200 in 1998); treatment group: taxable income DKr 25,000-75,000 above the top-tax cutoff; control: DKr 25,000-75,000 below.
  Registers: IND (income tax register)
  working paper · Section V, V.A (Estimator 1)

• Self-employed (main occupation) Employment and labour force status
  Persons whose primary occupation is self-employment, classified with the Statistics Denmark occupation variables pstill2 (primary) and sstill2 (secondary); narrower than Chetty et al. (2011), excluding wage earners with self-employment as secondary occupation.
  Registers: IDA, Income Tax Register
  Codes: pstill2, sstill2
  working paper · Section 2.1, Table 1 notes
• Analysis sample Sample definition and timing
  Entire population panel 1994-2009, aged 25-59, main occupation self-employment; restricted to self-employed whose spouse's primary or secondary occupation is not self-employment (because transfers to assisting spouses are only observed 1994-2001).
  Registers: Income Tax Register, IDA
  working paper · Section 2 'Data and Institutional Background'
• Personal income Earnings and income
  Sum of wage earnings, declared firm profits (net of retained earnings), remunerations, provisions, alimony etc. minus deductions; interest income is capital income and not part of personal income.
  Registers: Income Tax Register
  working paper · Figure 1 notes; Section 2.1
• Taxable income relative to the effective kink Taxes and tax rates
  For each individual, the difference between actual taxable income and the taxable income needed to reach the middle or top tax bracket, computed with the authors' own tax simulator accounting for special deductions and joint taxation of couples (simulated tax within +/-5 DKK of actual for 95% of the sample); distributions plotted in 1,000 DKK bins.
  Registers: Income Tax Register
  working paper · Section 2 and Figure 2
• Tax schedule kinks Taxes and tax rates
  Piecewise-linear personal income tax with five brackets before 1996 and four from 1996; e.g. 2001: about 8% below 33,400 DKK, about 44% up to 179,900 DKK (bottom tax), 49% above (middle tax) and 63% above 279,900 DKK (top tax). Top-tax base is mainly individual; middle-tax base depends on household income; middle cutoff raised to the top cutoff in 2009.
  Registers: Income Tax Register
  working paper · Section 2
• Business tax scheme and retained earnings Firm characteristics and productivity
  Self-employed choose the personal income tax scheme, the capital returns scheme (from 1993) or the firm tax scheme (from 1987); the latter two are treated as one and allow retaining profits in the firm, taxed provisionally at the firm tax rate (34% in 1994 falling to 25% in 2009), with withdrawals taxed as personal income. Retained earnings = profits retained under these schemes.
  Registers: Income Tax Register
  working paper · Section 2.1 'Three Tax Schemes for the Self-employed'; Table 2
• Earned income Earnings and income
  Income before shifting: personal income plus retained earnings plus income shifted along other margins such as pension contributions and transfers to assisting spouses; used (adjusted for income shocks) to estimate the real bunching response.
  Registers: Income Tax Register
  working paper · Section 4 discussion of Figures 11-12
• Excess mass at kink (bunching) Taxes and tax rates
  Excess mass at the tax cutoff estimated following Saez (2010) within a bandwidth around the kink, applied to taxable income and to adjusted earned income; the dynamic extension separates bunching from intertemporal shifting via retained earnings.
  Registers: Income Tax Register
  working paper · Section 3-4, equation (3)

• Audit treatment (stage 1) Taxes and tax rates
  Stratified random sample of 25,020 employees and 17,764 self-employed (strata: employee/self-employed, light/heavy return complexity, 30 regional tax centres) randomly split into a 100% group subjected to unannounced comprehensive audits of returns filed in 2007 (2006 income) and a 0% group never audited; effects are intent-to-treat (actual audit rates 98.7% employees, 92% self-employed).
  Registers: SKAT Business Object Database
  working paper · Section 4.2
• Threat-of-audit letters (stage 2) Taxes and tax rates
  Employees in both audit groups randomly assigned (thirds) to receive a letter one month before the 1 May 2008 filing deadline stating that the return would certainly be audited (100% letter), that half of the group would be audited (50% letter), or no letter. Outcome is the probability of self-reported adjustments (upward/downward) relative to the latest return at the time the letter was received.
  Registers: SKAT Business Object Database
  working paper · Section 4.2; Section 7 and footnote 38; Table 8
• Detected tax evasion (audit adjustments) Taxes and tax rates
  Net audit adjustment = after-audit return (A) minus filed return (F), split into under-reporting (upward adjustments raising tax liability) and over-reporting (downward adjustments), for net income, total tax and income components.
  Registers: SKAT Business Object Database
  Codes: Returns: information report (I), pre-populated (P), filed (F), after-audit (A)
  working paper · Section 4.3; Section 5.1; Table 2 notes
• Third-party vs self-reported income and evasion Taxes and tax rates
  Third-party income comes from information reports; self-reported income = total reported income minus third-party income. For always-positive line items, third-party under-reporting = A-F if F<A<I, = I-F if F<I<=A, zero otherwise; self-reported under-reporting is total upward adjustment net of third-party under-reporting. Third-party under-reporting is defined only for positive items.
  Registers: SKAT Business Object Database
  Codes: F<A<I: A-F; F<I<=A: I-F; else 0
  working paper · Section 5.2; Table 2 notes
• Net income and components Earnings and income
  Net income = personal income + capital income + stock income + self-employment income + foreign income - deductions. Personal income: earnings, pensions, alimony minus some retirement contributions; capital income: interest income, bond returns, net rents minus interest payments; stock income: dividends and realised capital gains on stocks; self-employment income: net profits of unincorporated businesses.
  Registers: SKAT Business Object Database
  working paper · Table 2 notes; Table 1
• Audit flags Taxes and tax rates
  Presence and number of computer-generated audit flags, a deterministic function of the return characteristics that SKAT normally uses to select returns for examiner review.
  Registers: SKAT audit flag system
  working paper · Section 4.1; Section 4.3

• Wage earner sample Employment and labour force status
  Individuals aged 15-70 in the tax register 1994-2001 classified as wage earners by IDA employment status, excluding self-employed, pensioners, wage earners with self-employment income and wage earners with unemployment; much of the analysis uses wage earners within DKr 50,000 of the top tax cutoff. Self-employed analysed separately.
  Registers: IND (tax register), IDA
  Codes: BESKST=4 (robustness: BESKST in 4, 5, 7, 8)
  working paper · Section 4; Appendix B
• Tax simulator and marginal tax rates Taxes and tax rates
  Authors' tax simulator computing tax liabilities and marginal tax rates from own and spouse's tax variables and previous-year municipality of residence; it matches actual municipal, regional, bottom, middle and top taxes within DKr 5 for 95% of observations.
  Registers: IND (tax register)
  Codes: pnr, henv (spouse), glskkmnr (municipality), pfrdst, berfrdst (personal exemption), perindkp, berpi (personal income), kapindkp, berkap (capital income), lignfrdp (special deductions), kappens, fosfufrd (individual pension contributions), arbpen14, arbpen15 (employer capital pension contributions), underhol (alimony)
  working paper · Appendix B
• Wage earnings and net deductions Earnings and income
  Wage earnings from the tax register; net deductions = deductions minus non-wage income (accounting for spousal deductions), equivalently wage earnings minus taxable income; in the top-tax base, net deductions = wage earnings needed to start paying top tax minus the statutory top tax cutoff. Small-deduction group: net deductions below DKr 7,500 in magnitude.
  Registers: IND (tax register)
  Codes: wage earnings qlontmp2; self-employment profits qovskvir, virkordind
  working paper · Section 4; Appendix B
• Bunching (excess mass) at kinks Taxes and tax rates
  Excess mass b = number of individuals in a window around the kink in excess of a counterfactual density, relative to the average counterfactual density in the window; the counterfactual is a seventh-degree polynomial fitted to binned counts of the earnings distribution excluding a DKr 15,000 window centred on the kink (q=7, R=7).
  Registers: IND (tax register)
  Codes: q=7 (polynomial degree), R=7 (bins on each side), excluded window DKr 15,000
  working paper · Section 4, equations (15)-(16)
• Firm bunching and occupation earnings modes Occupation and job hierarchy
  Occupation-year modes = DKr 5,000 wage-earnings bin with most workers in each 4-digit ISCO occupation-year (occupation-years with under 7,000 workers excluded). Firm bunching b_F = excess mass at the statutory top-tax cutoff for workers with more than DKr 20,000 in deductions; individual bunching b_I = excess mass at the pension kink for the same group; compared across 2-digit ISCO occupations.
  Registers: IDA, IND
  Codes: occupation discok (ISCO 4-digit; 2-digit for Figure 18); deductions > DKr 20,000
  working paper · Section 6 (Predictions 2 and 3); Appendix B; Figures 14, 18
• Pension-driven kink Retirement and pensions
  Cap on tax-deductible pension contributions (about DKr 33,000 on average) means workers contributing up to the cap enter the top bracket at wage earnings about DKr 33,000 above the statutory cutoff; bunching there is used to measure individual (job-search) bunching.
  Registers: IND (tax register)
  working paper · Introduction; Section 6

• Main monthly employer (BFL) Employment and labour force status
  From monthly eIndkomst job records (BFL), each person-month is assigned the job with the most paid hours (AJO_LOENTIMER) as the primary employer; its CVR number, wage and hours are kept. Month and year come from the job start date field AJO_JOB_START_DATO.
  Registers: BFL
  Codes: AJO_CVR_NR_FRA_PROD_JOB, AJO_SMALT_LOENBELOEB, AJO_LOENTIMER, AJO_JOB_START_DATO; primary job = max(AJO_LOENTIMER) within pnr-year-month
  replication code · script/cleaning/1_c_monthly_data.R
• Stayer (same employer Dec 2020 to Dec 2021) Employment and labour force status
  Among workers with an employer and monthly wage above 2,000 DKK in December 2020, stay=1 if the main employer in December 2021 is the same as in December 2020, else 0. Sample restricted to ages 20-65 and 2020 annual BFL earnings of at least 24,000 DKK.
  Registers: BFL
  Codes: wage>2000 in Dec 2020; employer_g == F12.employer_g; sum_earnings>=24000
  replication code · script/analysis/fig4/preprocess_fig4a.do; paper Figure 4 note
• Months out of work after separation Unemployment and benefits
  A separation is a month in which the worker has a main employer with monthly wage above 2,000 DKK and next month's employer differs; only the first separation per person is used. Time out of work = months until a different non-missing employer appears (capped at 24) minus one, so job-to-job moves count as 0 and returns to the same employer are not reemployment. The paper describes separations in 2020; the posted do-file's window (date<=24 from Dec 2018) covers Dec 2018-Dec 2019.
  Registers: BFL
  Codes: jan_sep = employer_g != F1employer_g; reemp = first k with F`k'employer_g != employer_g & non-missing; gap = reemp-1; cap 24
  replication code · script/analysis/fig4/preprocess_fig4b.do; paper Section 3.2 and Figure 4 note
• Annual earnings (register) Earnings and income
  Annual earnings are wage income LOENMV_13 from the income register (IND); self-employment income is NETOVSKUD_13 kept separately. Realized earnings growth 2020-2021 is ln(earnings_2021*100/102) - ln(earnings_2020), dropping observations with earnings below 24,000 DKK in either year and trimming at the 1st/99th percentile.
  Registers: IND
  Codes: LOENMV_13 (earnings), NETOVSKUD_13 (self-employment income); threshold 24,000 DKK; deflator 100/102; winsor2 trim 1/99
  replication code · script/cleaning/1_b_annual_data.R; script/analysis/fig7/preprocess_fig7.do; paper Figure 6 note
• Net wealth and liquid wealth Wealth, debt and consumption
  Net wealth = (KOEJD + BANKAKT + KURSAKT + OBLAKT + PANTAKT) - (BANKGAELD + GAELD_TIL_OFF + OBLGAELD + PANTGAELD), missing public debt set to 0. Liquid wealth = BANKAKT + KURSAKT + OBLAKT + PANTAKT. Pension wealth is the sum of PF_FORMUE across contracts in PENSFORM.
  Registers: IND, PENSFORM
  Codes: KOEJD, BANKAKT, KURSAKT, OBLAKT, PANTAKT, BANKGAELD, GAELD_TIL_OFF, OBLGAELD, PANTGAELD; PF_FORMUE
  replication code · script/cleaning/1_b_annual_data.R
• Monthly UI, cash benefits and retirement transfers Unemployment and benefits
  From ILME monthly non-employee e-income records, amounts across A/B income fields are summed per person-month and classified by the income-type code of the largest payment: UI if VMO_INDK_ART codes 2,5,8,16,104; cash benefits if type code 6 or art codes 40,41,43,84,87; retirement if art codes 6,9,10,13,20,24,25,44,45,74,83,86,113,115.
  Registers: ILME
  Codes: VMO_INDKOMST_ART_KODE UI: 2,5,8,16,104; CB: VMO_INDKOMST_TYPE_KODE 6 or ART 40,41,43,84,87; Retirement ART: 6,9,10,13,20,24,25,44,45,74,83,86,113,115
  replication code · script/cleaning/1_c_monthly_data.R
• Public vs private sector Sector and industry
  Main job (highest LOENTIMER in IDAN per year) is public if ARB_SEKTORKODE is non-missing and not 99, and recoded private if ARB_SEKTORKODE is in 18, 28, 89, 81, 58, 48, 42, 52. Industry groups come from DB07 ranges of ARB_HOVED_BRA_DB07 (e.g. >=840000 public services, 850000 teaching, 860000 health).
  Registers: IDAN
  Codes: ARB_SEKTORKODE private: 18,28,89,81,58,48,42,52; ARB_HOVED_BRA_DB07 6-digit DB07 ranges
  replication code · script/cleaning/1_b_annual_data.R
• Occupation group Occupation and job hierarchy
  Main job's DISCO_KODE first digit mapped to 9 groups (1 management, 2 high knowledge, 3 medium knowledge, 4 office, 5 sales/services, 6 farming, 7 craft, 8 skilled, 9 unskilled); military (DISCO below 100000) dropped. Tenure = year - ANSAAR.
  Registers: IDAN
  Codes: DISCO_KODE first digit 1-9; DISCO_KODE<100000 excluded; ANSAAR
  replication code · script/cleaning/1_b_annual_data.R
• Education level Education and test scores
  Highest completed education HFAUDD (latest available year) mapped through the audd format's H1 level: 10 elementary; 20, 25 high school; 30, 35 vocational; 40, 50, 60 bachelor; 65, 70 master.
  Registers: UDDA
  Codes: HFAUDD via uddan_2014_audd H1: 10; 20,25; 30,35; 40,50,60; 65,70
  replication code · script/cleaning/1_b_annual_data.R
• Population survey weights Sample definition and timing
  Inverse predicted propensity scores from a probability model of survey participation estimated on 2020 administrative data on the demographics of Danes active in the labour market.
  paper · Footnote 10 (details in Online Appendix B)

Replication package

• Household total income Earnings and income
  Yearly total income before taxes and labour market contributions, including labour income, public transfers, property income and most other taxable income attributable to the individual (excludes imputed rental value, pension contributions to employer schemes and lottery winnings). Household value is the average of the two spouses' values if a spouse is present in CPR, otherwise the individual value.
  Registers: IND, BEF
  Codes: PERINDKIALT_13 (income); spouse via E_FAELLE_ID
  replication code · code_dstserver/code_epinion/2_build_registries.do; NBER WP w32664 Section 1.2 and footnote 6
• Household labour income Earnings and income
  Total taxable wage income, benefits, bonuses, severance pay and value of stock options, averaged across spouses as for total income.
  Registers: IND
  Codes: LOENMV_13 (income_work)
  replication code · code_dstserver/code_epinion/2_build_registries.do; NBER WP w32664 Section 1.2
• Recent and future household income change Earnings and income
  For a respondent interviewed in any month of year t, recent change = log nominal household income in t-1 minus t-2; future change = log income in t+1 minus t-1. Changes are trimmed at the 2.5 and 97.5 percentiles.
  Registers: IND
  Codes: Log_Income_HH_L1L2, Log_Income_HH_F1L1; winsor2 cut(2.5 97.5) trim
  replication code · code_dstserver/analysis/A1c_sample_restrictions.do; NBER WP w32664 Section 1.3
• Liquid assets and net wealth Wealth, debt and consumption
  Liquid assets follow Andersen et al. (2020): total value of bank deposits, stocks and bonds reported by financial institutions to SKAT. Net wealth is SKAT's tax-purpose measure of net household assets excluding non-deposited cash and foreign assets, with real estate at tax-assessed values; both averaged across spouses and used as asinh changes.
  Registers: IND
  working paper · Section 1.2 and footnote 7
• Self-employment exclusion Sample definition and timing
  Exclude respondents if self-employment income exceeds one fourth of household income (in absolute value) in any year from t-4 to t+1 around the interview.
  Registers: IND
  Codes: abs(Income_selfemp_HH)/abs(Income_HH) > .25 for timestamps L4..F1
  replication code · code_dstserver/analysis/A1c_sample_restrictions.do; NBER WP footnote 11
• Main survey sample Sample definition and timing
  Monthly survey respondents 2012-2019 (backcast analyses 2013-2019) aged 25-60, excluding high self-employment income, missing key survey answers (Do not know, refused or forecast >100 pp) and missing registry variables, after trimming income changes and inflation answers at 2.5/97.5 percentiles.
  Registers: BEF, IND, UDDA
  Codes: Age<25 | Age>60 dropped
  replication code · code_dstserver/analysis/A1c_sample_restrictions.do; NBER WP Section 1.3
• Family emergency room visit in survey month Health care use
  Indicator equal to one if any household member (household identifier from BEF) visited the emergency room in the calendar month of the interview. An ER visit is an acute outpatient hospital contact in the NPR (recorded as patient type ER before 2014, acute outpatient from 2014). Respondents with more than 7 family ER visits over 2008-2018 are dropped.
  Registers: LPR, BEF
  Codes: has_now_visit_fam; ERvisit_count_surveysample_fam <= 7
  replication code · code_dstserver/analysis/A9_healthregs.do; NBER WP Online Appendix A.1 (Emergency Room Visits Data) and ER table notes
• Highest education (ISCED) Education and test scores
  HFAUDD from UDDA mapped via an ISCED crosswalk to single-digit ISCED 2011 levels 1-6; code 9 (not known) set to missing.
  Registers: UDDA
  Codes: HFAUDD -> ISCED_AUDD_AFSP4L_KT first digit
  replication code · code_dstserver/code_epinion/2_build_registries.do

Replication package

• Fatal heart attack or stroke (death event) Health outcomes, diagnoses and mortality
  Robustness version requires the spouse to die from the same cause as the index health event within 12 months.
  Registers: LPR, Death Registry
  online appendix · Supplemental Appendix, Figures B.2-B.3 notes
• Mental health medication take-up Health care use
  Indicator for redeeming a prescription for psychotropic drugs classified as psycholeptics (ATC N05) or psychoanaleptics (ATC N06) by the surviving spouse, measured at monthly and annual frequency (extensive margin).
  Registers: LMDB
  Codes: ATC N05, N06
  working paper · Section I (Variable definitions)
• Household income loss Earnings and income
  Household income decomposed into seven sources for each spouse: earnings, sick pay, DI + old-age pension (OAP), other government benefits, private pensions, capital income and other income, compared between the year before (period -1) and the year after (period 1) the index event. In 2SLS the household income loss is normalized at the treatment-group mean and winsorized at the top and bottom 1 percent.
  Registers: Income Registry (IND)
  online appendix · Supplemental Appendix, Tables B.4-B.7 notes
• Deceased's income share (instrument) Earnings and income
  Household income loss is instrumented with an indicator for whether the deceased spouse had earnings and a quadratic in the share of household income earned by the deceased spouse in the year before the event; 2SLS uses observations for years -1 and 0 with household FE, calendar-year FE and a quadratic in the survivor's age; SEs clustered by household.
  Registers: Income Registry (IND)
  online appendix · Supplemental Appendix, Table B.7 notes
• Treatment/control by event timing Sample definition and timing
  Households where one spouse has a health event 1999-2018 and dies within 12 months; treatment = event in 1999-2013, control = similar event in 2004-2018 with placebo event assigned five years earlier; both spouses aged 45-80 at baseline; balanced from four years before to four years after. Specifications include household FE, calendar-year FE and a quadratic in the surviving spouse's age. For Delta = 3-7 robustness the calendar interval of treated spousal deaths is held fixed so only the control group changes. Variables are measured in the year before the actual or placebo event.
  Registers: BEF, LPR, Death Registry, IND
  online appendix · Supplemental Appendix, Table B.1-B.3 and Figure B.8 notes
• Baseline psychiatric treatment Health care use
  Indicator for whether the surviving spouse had psychiatric treatment (visits to psychiatrists recorded in the Health Insurance Registry from 1990), used as a heterogeneity/control variable.
  Registers: SSSY, SYSI
  working paper · Section I (Data); Section III.B robustness
• Deceased earner status Earnings and income
  Indicator for whether the deceased spouse had labour market earnings in the year before the health event; used to split households (earner vs non-earner) and as an instrument.
  Registers: Income Registry (IND)
  online appendix · Supplemental Appendix, Figures B.5-B.7 notes
• Disability insurance and old-age pension rules Retirement and pensions
  DI for those below OAP age requires a loss of work capacity of at least 50 percent (assessed by municipality) and can also be granted for social reasons; paid until OAP age. OAP for all residents from full retirement age, rising from 65 to 67 for cohorts born after 1954; both means-tested against own and partner income and independent of earnings history.
  Registers: Social pension rules
  Codes: work-capacity loss >= 50%
  online appendix · Supplemental Appendix A: Institutional Background

• Default liquid wealth Wealth, debt and consumption
  Third-party-reported year-end value of current accounts, savings accounts and securities accounts (bank deposits and listed securities); excludes durables, real estate and pension accounts and does not net out debt. Available for the full population; mirrors the 'default wealth' the authorities use for the old-age check asset test.
  Registers: IND, Statistics Denmark administrative wealth registers
  paper · Section I.A and I.C; Table 1 notes; Figure 1 notes
• Reported liquid wealth Wealth, debt and consumption
  Default liquid wealth plus self-reported corrections submitted via the government website (e.g. adding cash, reducing account balances); available only for individuals above the statutory retirement age. The variable available to the authors does not capture corrections made close to year-end.
  Registers: SOCP, IND
  paper · Section I.A and I.C; Table 1 notes
• Total wealth Wealth, debt and consumption
  Default liquid wealth plus pension wealth and housing wealth, net of liabilities (approximates total net wealth for the low-income elderly).
  Registers: IND, PENSFORM
  paper · Figure 1 notes (panel B)
• Old-age check eligibility (age and income) Retirement and pensions
  Age-eligible = aged 65 or older (statutory retirement age in the sample period). Low-income (income-eligible) = income, net of social security and before self-reported corrections, below the threshold at which the old-age check is fully phased out (disregard DKK 42,000, phase-out rate 34% for singles in 2020). Asset test: check lost entirely when liquid wealth at end of t-1 exceeds DKK 90,000 (2020 value; threshold raised 15% in 2010). For the 2001-2007 reform analysis the 2003 income threshold is applied throughout.
  Registers: IND, SOCP, BEF
  paper · Section I.A; Figure 1 and 2 notes; Supplemental Appendix Figure A.4 notes
• Treatment and control groups for bunching Sample definition and timing
  Treated: low-income singles aged 65+. Controls: low-income singles aged 60-64 (age-ineligible) and singles aged 65+ with incomes between 150% and 200% of the full phase-out income threshold. Sample restricted to singles; main cross-section pools 2014-2018 (2014-2016 for bank data).
  Registers: BEF, IND
  paper · Section I.C (Sample); Figure 1 notes (panel C)
• Excess mass below asset-test threshold Wealth, debt and consumption
  Panel version: dummy for default liquid wealth between 50% and 100% of the asset-test threshold regressed on age dummies and individual fixed effects for low-income elderly born 1941-1945, coefficients relative to age 64; bunching jumps by 2.5 points at eligibility age 65 and keeps rising with age. Liquid wealth inflated to 2020 values with the growth rate of the asset-test threshold; DKK 5,000 bins.
  Registers: IND, BEF
  Codes: 50%-100% of threshold window; reference age 64
  online appendix · Supplemental Appendix, Figure A.3 notes
• Cohort-approach counterfactual Sample definition and timing
  For single income-eligible individuals born 1941-1945, compare the indexed liquid wealth distribution at ages 70-74 (five to nine years into eligibility) with ages 60-64 (before eligibility); only those income-eligible in a given year are included (unbalanced panel).
  Registers: IND, BEF
  paper · Section II (Cohort Approach); Table 1 Panel B2 notes
• Bank liquid wealth, spending and cash withdrawals Wealth, debt and consumption
  Bank data 2014-2016 used to compare spending and changes in consumer debt in December, January and other months by bank liquid wealth (DKK 5,000 bins), for age-eligible (65+) low-income singles vs age-ineligible (60-64) low-income singles.
  Registers: Danske Bank customer data
  online appendix · Supplemental Appendix, Figure A.4 notes
• Old-age check parameters and indexation Retirement and pensions
  Old-age check (created 2003) parameters expressed in 2020 values by inflating with the index used to adjust tax and transfer parameters (tied to average wage earnings per worker; $1 = 6.53 DKK in 2020). Maximum check raised above indexation in 2005, 2007, 2009 and 2013; asset-test threshold raised about 15% above indexation in 2010; income test (phase-out start and end thresholds) very strict initially but unchanged in real terms since 2005 except a small increase in 2019.
  Registers: Tax-transfer rules
  Codes: $1 = 6.53 DKK (2020)
  online appendix · Supplemental Appendix, Figure A.1 notes

• Crime (conviction for offence committed in year t) Crime
  Indicator equal to one if the individual is convicted of a crime committed in year t, dated by offence year and scaled to 100 in event-study figures. Offence types are classified as economic/non-economic and as property/sexual/violent (the latter following Statistics Denmark's system), covering 4,732,529 crimes 1980-2018.
  Registers: Statistics Denmark crime registers
  Codes: Offence classification in Online Appendix Table I.I
  online appendix · Online Appendix H eq. (15)-(16); Table I.I; Figure I.6 notes
• Time from offence to apprehension Crime
  Log of one plus the days between when a crime is committed and when the perpetrator is apprehended, using the median time if the person commits multiple crimes in a year; only observations in the offence year are used.
  Registers: Statistics Denmark crime registers
  online appendix · Online Appendix Table I.IX notes
• Decline in 5-year survival probability due to cancer Health outcomes, diagnoses and mortality
  Logit of dying within 5 years, estimated on people aged 18-62 in 1980-2013 (diagnosed plus up to ten exactly matched controls per case) with cancer-type indicators (14 types plus multiple), age polynomial interactions, post-cancer, gender and marriage controls, separately for 1980-1992, 1993-2003 and 2004-2013. The decline is predicted probability minus the counterfactual with no diagnosis. Low survival means an above-median decline within gender (also split by quintile).
  Registers: Cancer register, Death register, Population registers
  online appendix · Online Appendix E
• Exact-matched controls for survival estimation Sample definition and timing
  For each diagnosed individual, controls who in the year before diagnosis share age, marital status and gender and are in the same income ventile and years-of-education quintile. Up to ten controls with the smallest difference in total income are kept; unmatched observations are dropped.
  Registers: Population, income and education registers
  online appendix · Online Appendix E
• Treatment/control (Fadlon-Nielsen DiD) Sample definition and timing
  Alternative design comparing individuals diagnosed in a given year with those diagnosed Delta years later (Delta from 3 to 8), with person, year-by-age, in-prison and cancer-recurrence fixed effects. The baseline sample keeps only the 10 years before cancer for pre-diagnosis observations.
  Registers: Cancer register, Crime registers
  online appendix · Online Appendix H; Appendix A
• Change in municipal welfare generosity (2007 reform) Unemployment and benefits
  Percentage change in labour income plus transfers in post-diagnosis years relative to the five-year pre-diagnosis average, regressed on municipality fixed effects, age fixed effects, a cubic in pre-diagnosis income rank and year effects, separately pre-reform (1987-2006) and post-reform (2007-2018). People diagnosed 2004-2006 are excluded. The generosity change is the difference between post- and pre-reform municipality effects.
  Registers: Income registers (IND), Population register (municipality)
  online appendix · Online Appendix G, eq. (14)
• Psychological distress Health care use
  Indicator for having received treatment by a psychologist or psychiatrist following the cancer diagnosis, from data on medical visits.
  Registers: Health insurance / medical visits data (name not given)
  online appendix · Online Appendix F; Table I.VI

Replication package

• Monthly household consumer spending Wealth, debt and consumption
  Sum of three types of outgoing transactions in Danske Bank accounts: card payments (including payments initiated via mobile phone apps), bill payments and cash withdrawals. Excluded: transactions unlikely to reflect consumption, such as fee payments to the bank, payments for securities purchases and uncategorized bank transfers with unobservable purpose; Merchant Category Codes are used to exclude tax payments and debt repayments. Outgoing transactions from joint accounts are split evenly between account owners before aggregating to the household. Validated against Statistics Denmark's Household Budget Survey (table FU08, total spending minus imputed owner-occupied housing) by age of oldest household member, 2009-2016.
  Registers: Danske Bank transaction data, Household Budget Survey (Statistikbanken FU08)
  Codes: Merchant Category Codes (MCC) used to drop tax payments and debt repayments
  online appendix · Online Appendix B and Figure A1 notes
• Stock market gains Wealth, debt and consumption
  Monthly gain G = sum over days in the month and securities of the payout-adjusted daily price change times the number of units held (adjusted for splits, mergers and dividends, so it includes dividends). The portfolio covers direct equity and mutual fund shares in all liquid securities accounts, excluding bonds, deposits and retirement accounts.
  Registers: Danske Bank portfolio and price data
  working paper · Section 2.1, eq. (1)
• Passive gains (instrument) and portfolio controls Wealth, debt and consumption
  Passive gains are the gains earned in the month by the portfolio held at the beginning of the month, used to instrument actual gains. Controls are full interactions of portfolio size (stock value on the last day of the previous month), risk (SD of portfolio returns over the past 12 months) and expected return (mean monthly return over the past 12 months).
  Registers: Danske Bank portfolio and price data
  working paper · Sections 2.1 and 3
• Net investment in stocks Wealth, debt and consumption
  Gross purchases net of gross sales, computed residually as the total change in portfolio value minus the change due to capital gains.
  Registers: Danske Bank portfolio data
  working paper · Section 2.1
• Liquidity Wealth, debt and consumption
  Liquid assets (deposits, stocks and other securities) at the beginning of the month relative to average monthly income; household-months are ranked within month and split into terciles.
  Registers: Danske Bank data, Tax register (income)
  working paper · Section 4.3 (Heterogeneity)
• Financial attention (logins) Other
  Number of days in a month on which a household member logs into any Danske Bank account (not only brokerage), averaged over months, and split into within-month terciles.
  Registers: Danske Bank online login data
  working paper · Sections 2.1 and 4.3
• Estimation sample Sample definition and timing
  Robustness variants: Full sample (keeps households invested in a firm that is the main employer of a household member); Exclusive (drops households with accounts in banks other than Danske Bank); Excl. own industry (ignores investments in industries where household members are employed when building the portfolio); Excl. unlisted (drops households with unlisted shares); Future returns (adds interactions of average passive return since month 2 in 20 bins, portfolio size at start of month 1 in 100 bins, and calendar month, for positive horizons). Spending validation sample requires at least one spending transaction per month for each adult and at least five summed across adults.
  Registers: Danske Bank data, Employment register, Tax register (bank account reports)
  online appendix · Online Appendix A, Figure A5 and Figure A1 notes
• Household aggregation and controls Family links, fertility and household
  Households are formed from population-register identifiers linking cohabiting couples and dependants; spending, income and stock variables are summed within household. Controls: age of the oldest member, number of children, and highest education in the household (primary, high school, college, graduate).
  Registers: Population register, Education register
  working paper · Section 2.2; Section 3.3
• Household balance sheet (tax records) Wealth, debt and consumption
  End-of-year measures from tax records: total assets = financial assets plus housing; financial assets = securities plus bank deposits; risky share = risky assets / total financial assets. DKK amounts converted to USD at the 2009-16 average rate of 5.85 DKK/USD.
  Registers: Tax register (IND / wealth records)
  Codes: 5.85 DKK/USD
  online appendix · Online Appendix A, Table A1 notes

• Reform treatment (birthdate cutoff) Retirement and pensions
  Those born on or after 1 January 1954 learned in 2011 that their early retirement age (ERA), incentivized retirement age and full retirement age (FRA) rose from 60, 62 and 65 to 60.5, 62.5 and 65.5; those born one day earlier were unaffected. Running variable is birth date relative to the cutoff; RD sample is those born within 56 days (8 weeks).
  Registers: Population register (exact date of birth)
  working paper · Sections 2.3 and 4.2
• VERP participant analysis sample Sample definition and timing
  Four cumulative restrictions: (1) Danes born within six months of the cutoff date January 1, 1954 (71,095 individuals); (2) regular participatory VERP contributors, defined as making contributions in 70% of pre-announcement years 2001-2010 (47,047); (3) balanced panel: present in the data in every year 2006-2018 (43,348); (4) drop the self-employed, defined over the pre-announcement period using Statistics Denmark's definition of an individual's main source of income (final analysis sample 40,042).
  Registers: Population register, VERP contribution records, Statistics Denmark main source of income classification
  Codes: birthdate within +/-6 months of 1954-01-01; VERP contributions in >=70% of years 2001-2010; balanced 2006-2018
  online appendix · Online Appendix Table A.1 and notes
• Contributions to employer-sponsored and personal retirement accounts Retirement and pensions
  Annual contribution flows to employer-sponsored DC accounts in levels (winsorized at the 95th percentile by year) and indicators for positive contributions to personal retirement accounts (also indicators for contributions of various sizes).
  Registers: Labour-market/income registers (tax-reported pension contributions)
  working paper · Section 4.1
• Changes in bank, stock and property values Wealth, debt and consumption
  Year-t minus year-(t-1) year-end values of bank account balances and stock market account balances (reported by financial institutions to tax authorities) and of property cash values as assessed by tax authorities; winsorized at the 5th and 95th percentiles each year (also 10th/90th).
  Registers: Tax-reported asset registers
  working paper · Section 4.1, footnote 20
• Pre-tax earnings Earnings and income
  Income on which individuals pay the 8 percent labor market tax, winsorized by year at the 95th percentile; 2010 prices, 5.56 DKK per USD.
  Registers: Income register
  working paper · Section 4.1
• Monthly retirement age Retirement and pensions
  Absorbing-state measure from the monthly eIncome register (2008-2017): age in the last month with positive earnings before earnings permanently fall to zero.
  Registers: eIncome (eIndkomst) monthly earnings register
  working paper · Section 4.1, footnote 17
• VERP benefit claiming Retirement and pensions
  Annual Voluntary Early Retirement Pension (VERP) benefit payments.
  Registers: VERP benefit records
  working paper · Section 4.1
• Firm default contribution rate Firm characteristics and productivity
  All individuals over 18 are assigned to workplaces in 2010; individual contribution rate = employer-sponsored plan contributions divided by earnings; the firm default rate is the median contribution rate among the firm's workers in 2010. Predicted contributions = current earnings times the 2010 firm default rate.
  Registers: Linked employer-employee data, Pension contribution records
  working paper · Section 6 (employer defaults), footnote 30
• VERP participation indicator (qualified UI-fund contributions) Retirement and pensions
  Participatory VERP contribution in a year is measured as an indicator for making qualified (VERP) contributions to unemployment insurance (UI) funds in that year; VERP eligibility requires contributions in 25 of the previous 30 years.
  Registers: VERP contribution records
  online appendix · Online Appendix Table A.7 notes; Appendix C.2
• RD control variables Other
  Baseline RD controls: indicator for being male, indicator for being married in 2010, and indicators for region of residence in 2010 (five regions: Hovedstaden, Sjaelland, Syddanmark, Midtjylland, Nordjylland). Regressions use separate linear polynomials in birthdate on each side of the cutoff with triangular weights.
  Registers: Population register
  Codes: 5 regions as of 2010
  online appendix · Online Appendix Table A.2 notes
• Retirement at eligibility-age spikes Retirement and pensions
  Control group (born before cutoff): first spike = retirement age between 60y0m and 60y5m; second spike = 62y0m to 62y5m. Treatment group: first spike = 60y6m to 60y11m; second spike = 62y6m to 62y11m. 15.8% retire at the first spike and 10.6% at the second as defined.
  Codes: monthly retirement-age windows 60:00-60:05, 62:00-62:05 (control); 60:06-60:11, 62:06-62:11 (treated)
  online appendix · Online Appendix B
• Pre-reform personal retirement plan users Sample definition and timing
  Subsample of individuals who had used personal retirement plans before the reform announcement: those who made contributions to personal plans in two or three of the years 2008-2010.
  Registers: Tax-reported pension contribution records
  Codes: contributions in >=2 of 2008, 2009, 2010
  online appendix · Online Appendix table notes (personal-plan subsample)
• Outcome winsorization and time windows Sample definition and timing
  Anticipation-period outcomes are averaged over 2011-2013 and pre-announcement placebo outcomes over 2008-2010; robustness uses stricter winsorization at the 10th and 90th percentiles; an additional robustness subsample excludes individuals with real estate transactions.
  online appendix · Online Appendix Tables A.5-A.6, placebo table notes

• Granted debt relief (treatment) Other
  Applicants identified from Statstidende announcements (decision types P = proklama, K = kendelse/ruling, T = tilbagekaldelse/revocation, I); granted = the person's sequence of announcements contains a Kendelse (court ruling granting debt relief); alternative definition requires no subsequent revocation.
  Registers: Statstidende (hand-collected), CPR
  Codes: granted = 1 if any decision 'K'; granted_not_revoked = granted & no 'T' after 'K'
  replication code · 01classify_rep.sas, section 'Classify individuals into those who receive debt relief and those who don't'
• Trustee leniency instrument Other
  Leave-one-out mean grant rate of the applicant's court-appointed trustee (lawyer ID unique within court), net of the leave-one-out court grant rate; sample restricted to trustees with at least 20 cases; regressions include court x year fixed effects, SEs clustered by trustee.
  Registers: Statstidende (hand-collected)
  Codes: iv_lawyer2 = (mean_lawyer_leniency*n_lawyer - granted)/(n_lawyer-1) - (iv_court*obs_court - granted)/(obs_court-1); keep if n_lawyer >= 20
  replication code · 05IV_regressions_rep.do, lines ~19-35
• Employment status Employment and labour force status
  Through 2012: employed if IDA occupational status PSTILL < 38 or in 71-77; unemployed if PSTILL = 40; out of labour force if PSTILL 41-57 or 90-98. From 2013: employed if JOB_P_SOCIO_KODE 110, 120, 131-136; unemployed if 200; out of labour force if 311-517. Alternative: employed if wage income LOENMV_13 > 0.
  Registers: IDAP, IDA
  Codes: PSTILL <38 or 71-77 (employed), 40 (unemployed), 41-57/90-98 (OLF); SOCIO 110,120,131-136 / 200 / 311-517
  replication code · 03panel_rep.sas, 'Employment' block
• Earned income and hourly wage Earnings and income
  Earned income ERHVERVSINDK_13 from the income register and hourly wage TIMELON from IDA, CPI-deflated and winsorized at the 1st and 99th percentiles by year.
  Registers: IND, IDA
  Codes: ERHVERVSINDK_13; TIMELON; JOB_TIME_LOEN_SMAL (IDAN, JOB_TYPE = 'H')
  replication code · 04winsorize_rep.sas; 03panel_rep.sas
• Assets, debt, wealth, secured debt, home ownership Wealth, debt and consumption
  Assets = QAKTIVF (before 1997) / QAKTIVF_NY05 (from 1997); debt = QPASSIV / QPASSIVN; wealth = assets - debt; secured (mortgage) debt = PRIGALD before 1995 and OBLGAELD from 1995; home owner if property value KOEJD > 0. Bank debt BANKGAELD and mortgage deed debt PANTGAELD also used.
  Registers: IND, INDK
  Codes: QAKTIVF, QAKTIVF_NY05, QPASSIV, QPASSIVN, PRIGALD, OBLGAELD, KOEJD, BANKGAELD, PANTGAELD, GALDIO
  replication code · 03panel_rep.sas, 'Assets, Debt, and Wealth' block
• Benefit receipt indicators Unemployment and benefits
  Disability insurance if TILBTOT > 0; social assistance if cash assistance (KONT_GL, KONTANTHJ_13 from 1994) > 0; unemployment benefits if ARBLHUMV > 0 or ANDAKAS > 0.
  Registers: IND
  Codes: TILBTOT; KONT_GL / KONTANTHJ_13; ARBLHUMV, ANDAKAS
  replication code · 03panel_rep.sas, 'Disability insurance', 'Social assistance', 'Unemployment benefits' blocks

Replication package

• Expiring interest-only loan sample / treatment year Sample definition and timing
  Households with a mortgage loan taken in year t-10 that started with an interest-only (IO) period lasting at least 9 years (IO length = years between IO start and end dates in REAL) and that survives from t-1 to t; treatment year is when the IO period expires; event time = year - treatment year. Households with any self-employed member, non-tax-liable members, or a property purchase/sale/ownership change in event years -1 to 1 are dropped.
  Registers: REAL, BEF, EJER
  Codes: y_laanoptag = t-10 & l_afdragsfrihed_aar >= 9; drop if selvst, ikkeskattepligtig, or koebt/solgt/ejerskift in t_count -1..1
  replication code · 1_Extractdata.sas macro 'sample' lines ~119-135; 4_VarDefinition.do section 1
• Imputed household consumption Wealth, debt and consumption
  Family disposable income (excluding imputed rent of own home) plus one-off pension payouts (ATP, SP, LD, withdrawn pensions) net of their tax, minus savings in bank deposits, minus stock savings corrected for price changes (change in stock holdings less lagged holdings x stock index return), minus foreign asset savings, minus repayments (afdrag) on mortgage, bank, mortgage-deed and foreign debt, minus private pension contributions. From 2010 the mortgage repayment uses loan-level REAL repayments (version 3).
  Registers: IND, FORMUE, REAL
  Codes: famforbrug1a / famforbrug3 = famdispindk + (famatp_engang+famsp_pension+famld_udbetal+famhaevpen) - taxes on these - op_famindestpi - (op_famkursakt - famkursakt_lag*makro_aktie/100) - op_famudlakt - af_famoblgaeld (or af_famrealgaeld) - af_fambankgaeld - af_fampantgaeld - af_famudprigael - fampripensindbetal
  replication code · 0c_Familydata.sas lines ~1586-1695
• Outcomes scaled by lagged disposable income Wealth, debt and consumption
  Annual changes in household consumption, bank deposits, financial assets, bank debt and mortgage debt, and levels of installments and interest payments, divided by previous-year family disposable income (DISPON_13); all censored at the 1st and 99th percentile by year.
  Registers: IND, FORMUE, REAL
  Codes: rdiffconsump = (famforbrug - L.famforbrug) / L.famdispon_13; rfamafdrag_for = famafdrag_for / L.famdispon_13
  replication code · 4_VarDefinition.do section 4
• Liquid assets, debt-to-income, mortgage payment Wealth, debt and consumption
  Liquid assets = bank deposits + bonds + stocks (market value); total assets add housing value; liabilities = bank debt + mortgage outstanding; debt-to-income = total liabilities (QPASSIVN) / gross income (PERINDKIALT_13); total mortgage payment (ydelse) = installments + interest paid + administration fee (bidrag).
  Registers: FORMUE, IND, REAL
  Codes: liq_assets = fambankakt + famoblakt + famkursakt; dti = famqpassivn/famperindkialt_13; ydelse = famafdrag_for + famrente_bet_nu + fambidrag
  replication code · 4_VarDefinition.do section 2
• Refinancing behaviour at IO expiry Wealth, debt and consumption
  At the expiry year: loan stays (survives, no new loan in t-1, t, t+1); new IO loan taken in t-1, t or t+1; new amortising loan (no IO); or pays out the loan (none of the above).
  Registers: REAL
  Codes: out_loan_stays, out_newloan_io, out_newloan_noio, out_paysout
  replication code · 4_VarDefinition.do section 3

Replication package

• Job loss event Unemployment and benefits
  Month in which wage payments from the individual's main employer cease and total gross wage income drops below DKr 1,000; the individual must receive unemployment benefits or social insurance in months -1 to 3 and no early retirement, sickness or parental leave benefits then, must have had gross wage income of at least DKr 10,000 for at least 18 consecutive months before, and must not return to the same employer within three months.
  Registers: Monthly payroll and income transfer records (Danish Tax Agency, via Statistics Denmark)
  paper · Section III
• Analysis sample and event window Sample definition and timing
  Baseline sample of active customers: at least five outgoing spending transactions in each month of the event window, with the partner (if any) meeting the same criterion. Exclusive-customer subsample: active customers (and partners) with no deposits or loans at other retail banks. Characteristics measured in month -6 relative to job loss (annual income over the calendar year containing month -6; shares held at other banks and mortgage debt at end of previous calendar year).
  Registers: Danske Bank; tax records; mortgage data
  Codes: >= 5 spending transactions per month
  online appendix · Table A1 notes
• Mass layoff subsample Sample definition and timing
  Mass layoff notifications from the Ministry of Employment (firms with >20 employees must report planned large-scale layoffs; 'large scale' ranges from 10 workers for firms with 20-100 employees to 30 workers for firms with >300 employees), linked by firm ID to Tax Agency payroll data; individuals whose employer's report date falls in months -7 to -1 relative to the observed layoff month; 1,156 individuals.
  Registers: Ministry of Employment mass layoff notifications; BFL/eIndkomst
  Codes: report in months -7 to -1
  online appendix · Appendix E
• Household disposable income (bank inflows) and components Earnings and income
  Disposable income = sum of monthly incoming direct deposits, person-to-person transfers from outside the household and cash deposits into Danske Bank accounts owned by the affected person and/or spouse (excluding transfers between own accounts). Wage income: inflow coded as wage if sender's bank registration number (4-digit Danish bank code) is the one mapped to the recipient's employer in that month (mapping = registration number from which the highest share of the employer's payroll employees, per Tax Agency payroll data, received payments; second-ranked used if top is the central-government number and employer is not central government), or if transaction type code is e.g. 'salary transfers'. Joint-account wage payments go to the spouse if from the spouse's employer, otherwise to the affected person. Government transfers: sender registration number linked to a paying agency per payroll data, or the central-government registration number when person does not work for central government. Private transfers and other income = residual.
  Registers: Danske Bank transactions; BFL/eIndkomst monthly payroll and transfer records
  online appendix · Appendix A (Income variables) and Appendix C
• Total spending (transaction based) Wealth, debt and consumption
  Sum of outgoing card payments (incl. mobile-phone payments), bill payments and cash withdrawals from the household's bank accounts, excluding outflows identified via recipient MCC as tax or debt payments; excludes bank fees, securities purchases and uncategorized transfers. Bank aggregates raw transactions to daily totals per account x transaction type x recipient category. Household = affected person + spouse; joint-account outflows split evenly between owners.
  Registers: Danske Bank transactions
  Codes: MCC-based exclusion of tax/debt payments
  online appendix · Appendix A (Spending) and Appendix B
• Spending subcategories (MCC-based) Wealth, debt and consumption
  Utilities: MCC 4900, 4812, 4814, 4821, 4899 or bill-payment label utilities/elec/gas/water/heating/internet/cable TV/telephone/TV license. Restaurants and bars: MCC 5813, 5462, 5811, 5812, 5814. Groceries: MCC 5411, 5422, 5441, 5499, 5921 or bill-payment label groceries.
  Registers: Danske Bank transactions
  Codes: Utilities MCC 4900,4812,4814,4821,4899; Restaurants MCC 5813,5462,5811,5812,5814; Groceries MCC 5411,5422,5441,5499,5921
  online appendix · Appendix A (Spending)
• Net saving in liquid assets Wealth, debt and consumption
  Change in end-of-month balances on Danske Bank deposit accounts owned by the affected person or spouse plus outflows minus inflows of transactions with type code 'securities trade'.
  Registers: Danske Bank account records
  online appendix · Appendix A
• Net repayments on non-mortgage loans Wealth, debt and consumption
  Change in end-of-month balances on loan accounts (amounts owed coded as negative balances) owned by the affected person or spouse.
  Registers: Danske Bank account records
  online appendix · Appendix A
• Mortgage equity extraction and loan types (monthly) Wealth, debt and consumption
  Monthly mortgage portfolio reconstructed from annual end-of-year loan-level snapshots: loans tracked by unique loan ID; loans appearing/disappearing dated by origination of new loans, and, if ambiguous, termination month inferred by comparing interest implied by rate and balance with mortgage interest reported in annual tax returns. Adjustable-rate and interest-only indicators from loan type; average monthly payment imputed from remaining balance, time to maturity, interest rate and amortization profile; mortgage repayments = sum of average monthly payments over loans held at end of month.
  Registers: Mortgage bank loan-level data (end-of-year); tax returns (interest)
  online appendix · Appendix A and Appendix D

• Household definition and head Family links, fertility and household
  Households have one or two adults; two cohabiting adults are one household if (i) married or in a registered partnership, (ii) have at least one common child in the Civil Registration System, or (iii) are of opposite sex, age difference of 15 years or less, not closely related and with no other adults present. Children under 25, never married and childless belong to the parents' family. The head is the highest earner over the 13-year sample period.
  Registers: Civil Registration System (CPR) family links
  working paper · Section 4.1, footnotes 26-27
• Imputed household expenditure Wealth, debt and consumption
  C = Y - P - dNW: after-tax income (all labor and capital income, excluding capital gains) minus contributions to privately administered pension schemes minus the change in non-pension, non-housing net worth between 31 December of t-1 and t. Households in housing-transaction years and with more than trivial business income are excluded; equity holders are assumed to hold a diversified stock index; household-years with negative imputed expenditure (about 3 percent) are discarded.
  Registers: Tax-authority income records, Tax-reported wealth data, Private pension contribution records
  working paper · Section 4.2
• Unexpected income change Earnings and income
  After dropping the top and bottom 1 percent of household after-tax-and-transfer income, income is normalized by average household income over the period and regressed on dummies for age, year, highest education, marital status, homeowner status and number of children plus age interactions with education, marital and homeowner status; the year-to-year change in the residual is the unexpected income change, trimming the top and bottom 1 percent. Expenditure residuals are constructed the same way.
  Registers: Tax-authority income records, Education register, Population register
  working paper · Sections 4.1 and 4.2
• Estimation sample Sample definition and timing
  Households whose head is aged 30-55 in 2008 (to exclude young/old whose income growth is likely anticipated and whose income variance is non-constant). Out-of-sample exposures allocated to seven bins: young (<30) heads, old (>55) heads, pension funds (incl. children's savings accounts), government, non-financial corporates, financials, rest of the world.
  Registers: CPR household links
  Codes: head age 30-55 in 2008
  online appendix · Online Appendix E and G
• Liquid wealth quintiles Wealth, debt and consumption
  Households are assigned to quantiles by their mean level of liquid wealth (or income, etc.) over the sample period (long-term quintiles); liquid wealth is the end-of-year balance, averaged in real terms over 2009-2015 in descriptive comparisons.
  Registers: Tax-reported wealth data
  online appendix · Online Appendix E.1 and Table F.1 notes
• Unhedged Interest Rate Exposure (URE) Wealth, debt and consumption
  Household URE = annual savings (after-tax income minus imputed expenditure) plus maturing assets minus maturing liabilities. Maturing assets: all bank deposits (assumed floating) plus 20% of securities (5-year maturity assumed). Maturing liabilities: all bank debt (assumed floating), mortgage debt contractually due to reset its interest rate in the coming 12 months (counted at half weight since the new rate applies half the year; voluntary refinancing excluded), and other liabilities with an assumed 5-year maturity. Non-household sectors' maturity calibrated to 3.65 years so economy-wide URE sums to zero.
  Registers: Tax-reported wealth data, Mortgage reset data (Danmarks Nationalbank)
  Codes: securities 20%; mortgage resets within 12 months
  online appendix · Online Appendix H.1-H.2
• Net Nominal Position (NNP) Wealth, debt and consumption
  Household NNP = financial assets minus liabilities, averaged over 2009-2015 and deflated by CPI. Assets: bank deposits plus market value of securities excluding shares. Liabilities: all debt to financial institutions (including credit card debt), publicly administered student debt, tax debt and other debt to government bodies, as reported to the tax authorities by financial institutions. Other sectors from national financial accounts: net assets in currency and deposits, securities other than shares, loans, and trade credits/other accounts receivable-payable.
  Registers: Tax-reported wealth data, Danmarks Nationalbank financial accounts
  online appendix · Online Appendix H.1-H.2
• Nondurable consumption proxy Wealth, debt and consumption
  Imputed expenditure adjusted using registry data on household car purchases to proxy nondurable consumption (details in online appendix G, not in this file).
  Registers: Car purchase registry data
  working paper · Section 7.2

• Net worth Wealth, debt and consumption
  Sum of housing equity, liquid assets and financial wealth minus uncollateralized (bank) debt, at year end, CPI-deflated (2010=100) and divided by permanent income; pension funds and durables (cars, boats) excluded. Each component floored at 0 in code; individuals with net worth outside the 1st-99th percentile in any year are dropped.
  Registers: IND/FORMUE (GRUND files)
  Codes: totassets = stocks(KURSANP+KURSAKT) + bonds(PANTAKT+OBLAKT) + KOEJD + BANKAKT; totdebts = mortgage(OBLGAELD+PANTGAELD) + BANKGAELD; networth = totassets - totdebts
  replication code · data/server/rawbuild.do ('Wealth' block, 'SAMPLE SELECTION'); CEBI WP 02-17 Table D.1
• Liquid assets Wealth, debt and consumption
  Sum of cash and savings accounts held in Danish banks (BANKAKT), floored at 0, normalized by permanent income.
  Registers: IND/FORMUE
  Codes: BANKAKT
  replication code · rawbuild.do; WP Table D.1
• Housing equity Wealth, debt and consumption
  Value of owned real estate (KOEJD) minus collateralized debt (mortgage bonds OBLGAELD + other mortgage debt PANTGAELD, at market value of associated bonds at year end), normalized by permanent income.
  Registers: IND/FORMUE
  Codes: hequity = KOEJD - (OBLGAELD + PANTGAELD)
  replication code · rawbuild.do; WP Table D.1
• Financial wealth and uncollateralized debt Wealth, debt and consumption
  Financial wealth = market value of directly held stocks, bonds and mutual funds (KURSANP+KURSAKT plus PANTAKT+OBLAKT in code; WP lists KURSANP+KURSAKT+OBLAKT); uncollateralized debt = bank debts not associated with a bond (BANKGAELD). Both normalized by permanent income.
  Registers: IND/FORMUE
  Codes: finw = max(0, KURSANP+KURSAKT + PANTAKT+OBLAKT); liqdebts = BANKGAELD
  replication code · rawbuild.do; WP Table D.1
• Permanent income Earnings and income
  Weighted average of current and four lagged years of disposable income: 0.45, 0.25, 0.15, 0.10, 0.05 for t to t-4; disposable income = BRUTTO + SKATFRIYD + AKTIEINDK - SKATMVIALT_NY, CPI-deflated; floored at DKK 60,000 in code. Inheritance is normalized by permanent income in the year before parental death.
  Registers: IND
  Codes: dispinc = brutto + skatfriyd + aktieindk - skatmvialt_ny; perminc = filter(dispinc), coef(45 25 15 10 5) lags(0/4) normalise
  replication code · rawbuild.do ('Income'); WP Appendix D
• Sudden parental death Health outcomes, diagnoses and mortality
  Parental death (1996-2012) is sudden if the primary cause of death has the listed ICD-10 codes. The WP states I21*-I22*, V*, X*, Y*, R96*; the replication code additionally includes I23, I24 and I46.
  Registers: Register of Causes of Death (doedsaars1), FTDB (parent-child links)
  Codes: WP: ICD-10 I21-I22, V*, X*, Y*, R96; code: substr(doedsaars1,1,3) in I21,I22,I23,I24,I46,R96 or first letter V, X, Y
  replication code · rawbuild.do ('DEATHS AND INHERITANCE'); CEBI WP 02-17 fn. 7 and Appendix D
• Potential inheritance Wealth, debt and consumption
  Deceased parent's gross bequest (max(0,housing equity) + financial wealth + liquid assets - uncollateralized debt, floored at 0, in the last observed year) split equally among all children (from FTDB), net of inheritance tax: if estate exceeds the exemption bundfr, net = (0.85*(estate - bundfr) + bundfr)/n_heirs. Where both parents die, the largest inheritance is kept. Normalized by heir's permanent income; main sample has potential inheritance > 1 year of permanent income.
  Registers: IND/FORMUE, FTDB2012, Register of Causes of Death
  Codes: gross_bequest; n_heirs; bundfr (yearly exemption, e.g. 264,100 DKK in 2010); inhgroup==2 if inheritance>1
  replication code · rawbuild.do ('Remove taxes from inheritance'); main_analysis.do line ~96; WP Appendix D, Table D.3
• Heir sample Sample definition and timing
  Heirs aged 25-59 in observed years, aged under 50 at the parental death (alder_atshock<50), with non-missing inheritance; net-worth outliers (outside p1-p99 in any year) dropped. WP: parents die unmarried between 1995 and 2012.
  Registers: BEF/GRUND, FTDB
  Codes: inrange(alder,25,59); alder_atshock<50
  replication code · rawbuild.do ('SAMPLE SELECTION'); main_analysis.do lines 14-29; WP Section I
• Labour income, salary and pension contributions Earnings and income
  Labour income = ERHVERVSINDK (labour-market income incl. bonuses, compensations and self-employment income); salary = LOENMV (salary only); pension contributions = private (qpripen) plus labour-market (qarbpen) pension contributions.
  Registers: IND
  Codes: erhv (labor_income), loenmv (salary), pencontr = qarbpen + qpripen
  replication code · rawbuild.do ('INCOME UPDATE'); WP Table D.2
• Number of housing units owned Wealth, debt and consumption
  Count of distinct properties (ejendom number x municipality) owned by the individual in the ownership register (EJER), plus total ownership share.
  Registers: EJER
  Codes: bnr = group(ejdnr komnr); n_owned; pct_owned = total(ejerpct)
  replication code · rawbuild.do ('HOUSING DATASET')

Replication package

• Car types (green/brown x heavy/light) Other
  Cars are split by whether weight is above or below the median for the car's vintage cohort, then within each half by whether fuel efficiency is above (green) or below (brown) the median, giving four types; attributes are averaged within type. Cars are grouped into 25 age groups from new to 24+ years (cars over 22 years dropped from estimation).
  Registers: Car ownership register (Statistics Denmark)
  working paper · Appendix F.1 (Car types); Section 6.2
• Household types Family links, fertility and household
  Eight household types from couple vs single, high vs low work distance, and high vs low income, where the income split is at the median income within the demographic cell.
  Registers: Demographic and income registers
  working paper · Appendix F.1 (Household types); Table 2
• High work distance Geography and neighbourhood
  Based on the Danish tax deduction for travel distance to work, which applies only to full-time workers living more than 12 km (each way) from their workplace; households with the deduction are classed as high work distance (slightly under half of Danes).
  Registers: Tax records (commuting deduction)
  working paper · Appendix F.1 (Household types)
• Kilometres driven Other
  Driving measured from odometer readings recorded at mandatory vehicle inspections, which occur biannually from car age 4 (ages 4, 6, 8, 10, ...).
  Registers: Vehicle inspection odometer data
  working paper · Appendix F.1 and footnote 32
• Scrappage Other
  Scrappage is not observed directly: a vehicle is defined as scrapped if an ownership spell ends and no other ownership spell ever begins afterwards (ownership register extract from September 2011, last sample year 2008, so at least about 3 years without an owner). Accidents are not observed.
  Registers: Car ownership register
  working paper · Appendix F.1
• New car registration tax Taxes and tax rates
  T(P) = 1.25 x [tau1 x min(P, K) + tau2 x max(P - K, 0)], with 25% VAT on the price including tax; in 2008 K = DKK 81,000, tau1 = 105% and tau2 = 180%. Fuel prices are country averages from eof.dk.
  Registers: Fuel prices (eof.dk)
  working paper · Appendix F.1
• Estimation sample Sample definition and timing
  Universe of Danish households and all privately owned cars, 1996-2008 (estimation on holdings 1997-2008, about 39 million household-year observations of car ownership transitions).
  Registers: Demographic registers, Car ownership register
  working paper · Appendix F.1; Section 6

• Hedonic expected house value Wealth, debt and consumption
  ln(P_it) regressed on time x municipality effects, apartment indicator, and property characteristics: ln(lot size), ln(interior size), number of rooms, bathrooms, showers, unoccupied-at-sale dummy, ln(building age), rural dummy, historic-building dummy, ln(distance to nearest major city), all interacted with the apartment indicator, plus a third-order polynomial of the previous-year (biennial) SKAT tax-assessed value. R2 = 0.88 on sold listings (N = 114,303). Variants add repeat-sales residuals and lagged renovation deductions.
  Registers: BBR/BOL, EJVK, EJSA, EJMT
  online appendix · Online Appendix C.1; Figure L.4 notes; E
• Listing premium Wealth, debt and consumption
  l = ln(L) - predicted ln(P), where L is the initial listing price and predicted ln(P) is from the hedonic model.
  Registers: RealView, SKAT, BBR
  paper · Section 4.1
• Potential gain (vs reference price) Wealth, debt and consumption
  G = predicted ln(P) - ln(R), where R is the nominal original purchase price of the property; measurable only for properties bought in 1992 or later.
  Registers: SKAT property transactions, BBR
  paper · Sections 3.5 and 4.1
• Potential home equity Wealth, debt and consumption
  H = predicted ln(P) - ln(M), where M is the outstanding mortgage balance reported by the household's mortgage bank that year (multiple mortgages on a property summed; LTV and rate weighted); sellers with H < 20% are classed as potentially facing a binding down-payment constraint.
  Registers: Danmarks Nationalbank mortgage data
  paper · Sections 3.3 and 4.1
• Listings sample Sample definition and timing
  From 615,040 RealView listings of owner-occupied housing (excluding cooperative housing): drop unmatched in registers (-107,679), no reference price i.e. purchased before 1992 (-144,962), owner ID not uniquely determined (-71,876), non-household buyer i.e. government/company/organization (-10,382), foreclosures (-6,416), extreme prices (<100,000 DKK or >20,000,000 DKK in 2015 prices, or flagged extreme by Statistics Denmark; -5,499), owner ID not found, missing lot size or hedonic characteristics, listing date before previous purchase date, intra-family/special-circumstance sales, missing listing price; then drop summer houses (-24,098) and professional investors owning more than 3 properties (-18,312). Final: 214,103 listings 2009-2016 (172,225 with mortgage, 41,878 without), 191,507 households, 178,933 properties; 70.4% sold, 29.6% retracted. Extensive-margin stock sample: 5,538,052 observations of 807,345 properties.
  Registers: RealView listings, EJSA, EJER, EJVK, BBR/BOL, REAL, BEF
  Codes: price bounds 100,000-20,000,000 DKK (2015 prices); >3 properties = investor
  online appendix · Online Appendix C.5 and Table L.1
• Household Family links, fertility and household
  Households consist of one or two adults and any children below age 25 living at the same address, linked via the CPR family identification number; individual income and wealth are aggregated to the household. Household education = average years of education across all adults in the household; household age = average age in the household. The seller is the owner (CPR) at the beginning of the year of the transaction.
  Registers: CPR/BEF, UDDA, IND, EJER
  Codes: children <25
  online appendix · Online Appendix C.1, C.4 (fn 5), Figure L.29 notes
• Mortgage balance and loan-to-value Wealth, debt and consumption
  Danmarks Nationalbank loan-level mortgage data (via Finance Denmark), annual 2009-2016, covering all mortgage banks: principal, outstanding balance, LTV and interest rate, with borrower CPR and property number. Multiple mortgages on a property are summed, with balance-weighted average interest rate and LTV.
  Registers: REAL (Boligfinansiering)
  online appendix · Online Appendix C.3
• Renovation tax deductions (BoligJob) Wealth, debt and consumption
  Proxy for maintenance/renovation: claimed BoligJob ('Boligjobordningen') tax deductions for labour costs of service work in the home, available from 2011, aggregated by household and linked to the seller; maximum 15,000 DKK per adult 2011-2015, then 12,000 DKK for maintenance/renovation and 6,000 DKK for other services (2016-2018). Used as 1-year lag (2012-2016) and 3- and 5-year cumulative lags.
  Registers: INDSLUT (Servicebidragsregister)
  online appendix · Online Appendix E.3.1
• Liquid and net financial assets Wealth, debt and consumption
  Liquid financial assets = cash, stocks and bonds; net financial assets = liquid financial assets minus short-term non-mortgage liabilities (mainly unsecured debt). Used as household controls with age and education length.
  Registers: IND (SKAT income and wealth records)
  online appendix · Online Appendix K.1
• Share of apartments and row houses (IV) Geography and neighbourhood
  Instrument: share of apartments and row houses (as defined in the BBR) listed in a given sub-market (municipality), capturing homogeneity of the housing stock.
  Registers: BBR
  online appendix · Online Appendix (IV section; Figure L.14)

• Homeownership (household-based) Wealth, debt and consumption
  Baseline measure from housing records (living in an owner-occupied home) validated against the tax-record measure based on housing wealth available from 1987, for the population above age 20; housing-register measure is slightly lower in level because it misses owners not residing in their owned property. A variant uses only 'clean' cases (defined in main text).
  Registers: Housing register (BBR); tax records (housing wealth, from 1987)
  online appendix · Figure A.I notes
• Home size Wealth, debt and consumption
  Area of the house in square metres used for habitation, from the housing register.
  Registers: Housing register
  working paper · Section 3.2
• Home value Wealth, debt and consumption
  Appraisal of the property by the Danish Tax Authorities made for assessing imputed rental income and taxable wealth.
  Registers: Tax authority property appraisals
  working paper · Section 3.2
• Total deductible interest expenses Wealth, debt and consumption
  Interest paid on mortgage debt, consumer debt and other loans (all deductible), for homeowners; mortgage interest cannot be separated over the whole period.
  Registers: Income tax records
  working paper · Section 3.2
• Household tax bracket / treatment status Taxes and tax rates
  A taxpayer is assigned to the top bracket if she or her spouse is in the top bracket, to the middle bracket if either is in the middle and neither in the top, and to the bottom bracket if both are in the bottom. Treated are those in the top bracket in every year 1984-1986; controls are in the middle or bottom bracket in every year 1984-1986.
  Registers: Income tax records, Population register spouse links
  working paper · Section 4
• Net-of-tax interest rate Taxes and tax rates
  (1 - tau) R, where tau is the marginal tax rate on negative capital income; the 1987 reform set this rate to 51 percent in all brackets (pre-reform top about 73, middle 62, bottom 48 percent in the average municipality), raising the net-of-tax rate by 81 percent for the top group, 29 percent for the middle and reducing it 6 percent for the bottom.
  working paper · Section 3.1, Figure 1
• Semi-balanced panel 1980-1996 Sample definition and timing
  Individuals observed in every year 1980-1996 while aged 20-80 (entering at 20, leaving after 80), giving 7 pre-reform and 10 post-reform years; about 40 million person-years.
  Registers: Population register
  working paper · Sections 3.2 and 4
• Long-run balanced panel 1980-2011 Sample definition and timing
  Very-long-run home size analysis uses a balanced panel of individuals observed in all years 1980-2011 (vs. main semi-balanced panel 1980-1996). Pre-trend robustness: group-specific linear trends estimated on 1983-1986 and residualized out.
  Registers: Housing register; population register
  online appendix · Figures A.II-A.IV notes

• Prescription drug spending and purchases Health care use
  The full history of prescription drug purchases from the Danish Prescription Drug Register (Laegemiddelstatistikregisteret), covering 1995-2014: date, shelf price, government coverage, type of coverage and type of drug. It is aggregated to daily (descriptive) or monthly (estimation) spending within coverage years. Outcomes include spending, number of pharmacy visits, and days until next purchase truncated at 365.
  Registers: Laegemiddelstatistikregisteret (Danish Prescription Drug Register)
  paper · Section 3; Section 4 (outcomes before reset)
• Coverage year and synthetic pre-reform coverage year Health care use
  Post-reform coverage years start with the first purchase after the March 2000 reform and reset 365 days after the first purchase, with the next year starting at the next purchase. Synthetic pre-reform coverage years apply the same rule from the first observed purchase, keeping only years that end by December 1999.
  Registers: Laegemiddelstatistikregisteret
  paper · Section 2; Section 3
• Spot price Health care use
  Coinsurance rate at the start of coverage month m given year-to-date spending, taking values 1, 0.5, 0.25, 0.15 or 0. Under the 2000 contract: a DKK 500 deductible, 50% coinsurance to DKK 1,200, 25% to DKK 2,800, 15% beyond, and a stop-loss at DKK 18,500 capping out-of-pocket costs at DKK 3,600. The spot price is 1 for the purchase that starts a coverage year.
  Registers: Laegemiddelstatistikregisteret
  paper · Section 2, Figure 1; Section 5 (price definition a)
• Expected end-of-year (shadow) price Health care use
  Average realised end-of-year price within cells of 5 age groups x 6 sextiles of predicted health spending x 100 percentiles of year-to-date spending, for each coverage month. Pre-reform counterfactual prices for 1998 are imputed by matching percentiles to pooled coverage years starting in 2000-2001.
  Registers: Laegemiddelstatistikregisteret
  paper · Section 5 (price definitions b-c, steps i-iii)
• Predicted pharmaceutical spending (health index) Health care use
  Following Handel (2013), year-t drug spending is regressed on year t-1 gender, age and age squared, number of GP office visits, fee-for-service payments to the physician, drug spending and its square, ATC level-3 drug-type indicators (88 categories), and hospital ICD-10 level-2 diagnosis indicators (215 categories). The model is then used to predict next year's spending; for coverage-reset analyses, covariates are measured in the 365 days before the coverage year starts.
  Registers: Laegemiddelstatistikregisteret, National Patient Register (hospital diagnoses), Health insurance (GP) records
  Codes: ATC level 3 (88 categories); ICD-10 level 2 (215 categories)
  paper · Appendix D
• Chronic drug Health outcomes, diagnoses and mortality
  Following Alpert (2016), a drug at ATC level 3 (pharmacological subgroup) is chronic if its median consumer buys it in at least three calendar months of a year (more than twice).
  Registers: Laegemiddelstatistikregisteret
  Codes: ATC level 3
  paper · Section 3 footnote 5; Appendix C
• Self-reported liquidity constraint Wealth, debt and consumption
  From the 2000 Danish National Health Interview Survey (DNHIS) question 'During the last 12 months, how often have you had trouble paying your bills'. Respondents answering at least 'some months' are coded as liquidity constrained (12%).
  Registers: DNHIS 2000 (survey linked by personal ID)
  paper · Section 6 (mechanisms), Table 5
• Plan literacy groups Education and test scores
  'Economic literates' are individuals with a master-level degree in economics or business (ISCED level 7); medical professionals hold a university degree in medicine or pharmacology (ISCED levels 7-8). Cognitive decline is proxied by age 65 and above versus below 65.
  Registers: Education register
  Codes: ISCED 7 (economics/business); ISCED 7-8 (medicine/pharmacology)
  paper · Section 6, Table 5 notes

• Loan default (financial trouble) Wealth, debt and consumption
  An individual is in default if at least one interest-bearing personal loan is at least 60 days late with payments at the end of the year, as reported by banks and other financial intermediaries to the Danish Tax Agency (SKAT); annual data 2004-2011 covering all loan accounts. Mortgage loans are supplied in separate files; loan type is otherwise not observed.
  Registers: SKAT loan register
  working paper · Section 3.1 and footnote 2; Figure 1 notes
• Default by loan size Wealth, debt and consumption
  Alternative severity measures defining default at defaulted loan levels exceeding DKK 0, 10,000, 50,000 and 100,000 (the defaulted amount may reflect several smaller loans).
  Registers: SKAT loan register
  working paper · Figure notes (Panel B, defaulting at loan levels)
• Bad-payer file registration Wealth, debt and consumption
  Recorded in the bad-payer files of credit bureaus Experian or Debitor Registret (snapshot of persons registered in 2009); people can be recorded after at least three reminders and a warning letter, for at most five years per missing payment.
  Registers: Experian bad-payer file, Debitor Registret
  working paper · Section 3.1 and footnote 3
• Parent-child sample Sample definition and timing
  All individuals aged 18-45 in 2004-2011 with at least one living parent, linked via CPR to parents in the population register; unit of observation is the parent-child-year cell (separate child-mother and child-father observations; new partners of divorced parents ignored).
  Registers: Population register, SKAT loan register
  working paper · Section 3.2
• Unemployment shock (event study) Unemployment and benefits
  Unemployment spell of at least three months in event year 2007, 2008 or 2009 for individuals employed with no unemployment spells in the five preceding years, aged 18-38 at t-5; controls are similar individuals without a shock in t. Amounts index-adjusted to 2009-DKK within cohort.
  Registers: Income register, Unemployment histories (Statistics Denmark)
  working paper · Section 4 (Unemployment event analysis)
• Liquidity constraint indicator / financial wealth Wealth, debt and consumption
  Financial wealth = stocks + bonds + bank deposits; indicator for having financial assets at end of year below one month's worth of average disposable income over the five years preceding the shock year.
  Registers: Income and wealth registers (Statistics Denmark)
  working paper · Figure 8 notes
• Historical parental asset path (instrument) Wealth, debt and consumption
  Ratio of parental financial assets to average disposable income for each year 1987-1996; parents are assigned to behavioural types by within-cohort decile of the average asset-income ratio over the ten years, used to instrument parental default in 2011.
  Registers: Income and wealth registers (Statistics Denmark)
  working paper · Section 4 (Historical financial behavior of parents)
• Middle-school GPA decile Education and test scores
  Within-cohort deciles of grade point average from the 9th grade graduation exam in Danish and Mathematics, available for those completing 9th grade in 2001/2002 or later.
  Registers: Education registers (grades)
  working paper · Section 3.2 and Figure on GPA deciles
• Survey preference measures Other
  January 2014 telephone survey of 1,748 individuals asking respondents to place themselves and both parents on a 1-10 scale for risk willingness, patience and impulsivity; linked to register data.
  Registers: 2014 survey linked to registers
  working paper · Section 4 (survey), Figure 11

• Household (tax unit) and married status Family links, fertility and household
  Household id is the family id when the person's spouse id equals the cohabitation partner id and civil status is in the married range (up to 2007, from FAIN); from 2008 from BEF using spouse id = partner id and civil status G or P. Persons not so linked form their own household. Only persons aged 20+ are kept and wealth is summed within household.
  Registers: FAIN, BEF
  Codes: FAIN (<=2007): aegtefaelle_id == c_faelle_id, c_status 2-5, hhid = c_familie_id; BEF (>=2008): aegte_id == efalle, civst in {G, P}; alder >= 20
  replication code · Empirical Analysis/DD_make_hh_ids.do
• Household taxable wealth (outcome) Wealth, debt and consumption
  Household net taxable wealth = sum over household members aged 20+ of the wealth variable FORM from the income-tax register (INDH); outcome is log(FORM). Portfolio shares use INDK components: equity share = listed stocks / total assets, housing share = (mortgage deeds + real estate value) / total assets.
  Registers: INDH, INDK
  Codes: FORM (INDH); share_tx_equities = KURSAKT/QAKTIVF; share_tx_housing = (KURSPANT+KOEJD)/QAKTIVF; other INDK fields: FORMSK, LOFTFORM, QPASSIV, PRIGALD, KONTANT, KURSOBL, INDESTPI
  replication code · Empirical Analysis/DD_base_data.do, section 1
• Wealth tax exemption threshold and marginal wealth tax Taxes and tax rates
  Annual exemption threshold (allowance) by year; for couples it equals 2 x single allowance, except 1989-1992 when couples' allowance uses a separate married allowance x 2. Household marginal wealth tax rate tau_it = statutory rate if FORM exceeds the household allowance, else 0. Bunching analysis uses distance b = FORM - allowance, deflated, in DKK 20,000 bins from -540,000 to 540,000.
  Registers: INDH
  Codes: Allowance (DKK): 1980-81 1,000,000; 1982 1,083,000; 1983 1,158,900; 1984 1,204,900; 1985 1,229,200; 1986 1,253,500; 1987 1,278,700; 1988 1,329,900; 1989 1,383,600; 1990 1,424,500; 1991 1,452,700; 1992 1,482,100; 1993 1,534,500; 1994 1,580,500; 1995 1,592,000; 1996 2,400,000; married: 1989-90 811,500x2, 1991 1,249,600x2, 1992-96 2x single
  replication code · Empirical Analysis/DD_base_data.do section 2.2; bunching_base_data.do
• Ceiling DD treatment (unbound by tax ceiling) Taxes and tax rates
  X = wealth tax actually paid (FORMSK) / theoretical wealth tax ((FORM - allowance) x tax rate). Treatment: household in top 1% of the wealth distribution with X > 0.99 in every pre-reform year of the window (e.g. 1982-1988), i.e. not bound by the tax ceiling; control: top 1% with X < 0.80 (bound by ceiling). Households above the 99.99th percentile excluded. In the panel, status is re-coded post-reform using X, LOFTFORM > 0 and wealth relative to allowance.
  Registers: INDK, INDH, FAIN
  Codes: treatmt_2e_X80: X > .99 and top1 in 1982-1988; control_2e_X80: X < .80 and top1; drop form_pctile > .9999
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 3.4; DD_base_data.do section 2.3
• Couples DD treatment Taxes and tax rates
  Treatment: married households with wealth between the (single) allowance and 2 x allowance in each pre-reform year (1988 and earlier years depending on window), which became exempt when the couples' threshold doubled in 1989; controls: (c1) singles with wealth in the same band, (c4) married couples in the top 5% with wealth below the allowance, or (c5) singles with wealth between allowance/2 and allowance.
  Registers: INDH, FAIN
  Codes: treatmt_c1*: married & allowance <= FORM <= 2*allowance; control_c1*: unmarried, same band; control_c4*: married & FORM < allowance & top5; control_c5*: unmarried & allowance/2 <= FORM <= allowance
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 3.3
• Balanced panel 1980-1996 Sample definition and timing
  Households (persons 20+ aggregated by household id) observed with strictly positive aggregate wealth FORM in all 17 years 1980-1996.
  Registers: INDH, FAIN
  Codes: drop if form <= 0 | missing; keep if _N == 17 (balanced_1980_96)
  replication code · Empirical Analysis/DD_balance_and_treatment_status.do section 2
• Self-employment / business income indicator (bunching) Earnings and income
  Household flagged as self-employed if business profit (NETOVSKUD_GL) is nonzero; a broader 'ext' flag also uses nonzero values of EGNKAPVK, UDLFKURS, UDLIND2 or UDLIND3 (foreign/business-related components).
  Registers: INDH, INDK
  Codes: selv = netovskud_gl != 0; ext = netovskud_gl!=0 | egnkapvk!=0 | udlfkurs!=0 | udlind2!=0 | udlind3!=0
  replication code · Empirical Analysis/bunching_base_data.do

Replication package

• Refinancing event Wealth, debt and consumption
  A refinancing occurs when a household changes from one fixed-rate mortgage to another mortgage (fixed- or adjustable-rate) on the same property; terminations from moves, death or divorce are handled separately as mortgage termination. Annual loan data are expanded to quarters using mortgage issue dates.
  Registers: Danmarks Nationalbank mortgage data, CPR
  working paper · Section 2.2.B Sample selection
• Cash-out refinancing Wealth, debt and consumption
  A refinancing associated with an increase in mortgage principal of 10% or more.
  Registers: Danmarks Nationalbank mortgage data
  working paper · Section 2.1
• Refinancing incentive (ADL threshold) Wealth, debt and consumption
  Interest rate spread = coupon rate on old mortgage minus yield on a newly available mortgage of roughly the same maturity. Threshold from the Agarwal, Driscoll and Laibson (2013) closed form with baseline interest-rate volatility 0.0074 and discount rate 5%; incentive = spread minus threshold. Household-specific mortgage termination probability (from a probit on moves/death/divorce terminations) enters the threshold. Robustness: volatility 0.0037, discount rate 2.5%, constant 10% termination probability, Chen and Ling (1989) thresholds.
  Registers: Danmarks Nationalbank loan-level mortgage data
  Codes: sigma=0.0074 (baseline); discount rate 0.05 (baseline)
  online appendix · Table B2; Appendices F-I, K-L
• Household Family links, fertility and household
  'Single' households have only one adult living at the address (~13% of sample); 'Married' households have two legally bound adults, including registered same-sex partnerships. 'Children in family' = children in household; 'Immigrant' = an immigrant in the household; 'Getting married' = change in marital status over the sample; 'Having children' = child born within the last 12 months.
  Registers: Population register (CPR/BEF)
  online appendix · Table B3 notes
• Single fixed-rate mortgage sample Sample definition and timing
  All Danish households with an unchanging number of household members and a single fixed-rate mortgage at the beginning of each year 2010-2017; refinancing measured as a quarterly dummy. Robustness samples exclude cash-out and maturity-extension refinancings, exclude FRM-to-ARM refinancers, or restrict to mortgages over DKK 250,000 with horizon of at least 20 years.
  Registers: Danmarks Nationalbank loan-level mortgage data
  Codes: mortgage > DKK 250,000 and horizon >= 20 (robustness)
  online appendix · Appendices C-E
• Financial literacy Education and test scores
  'Financially literate' = a household member has a degree in finance or professional financial-industry training. 'Family financially literate' = non-resident parents, siblings, in-laws or children of the household are financially literate.
  Registers: Education register (UDDA), Population register
  online appendix · Table B3 notes
• Ranked household characteristics Wealth, debt and consumption
  Ranks computed each year across all sample households, normalized to [-0.5, 0.5]: age = age of oldest household member; education = years needed to reach the highest education of any household member (12 = high school, 16 = Bachelor's, 18 = Master's, 20 = PhD); income = total taxable household income; financial wealth = cash, bonds, stocks and mutual funds less non-mortgage debt; housing assets = value of properties. Specifications add a nonlinear transform f(x) of each rank.
  Registers: Tax register (income, wealth), Education register (UDDA), Population register
  Codes: education years: 12/16/18/20
  online appendix · Tables B3, B5 notes
• Mortgage termination probability Wealth, debt and consumption
  Probit of a monthly dummy for full prepayment/termination (moving, death, divorce, etc.) on household demographic dummies and rank variables (with nonlinear transforms), fitted on all households with unchanging membership and a fixed-rate mortgage 2010-2017; fitted values feed the ADL threshold.
  Registers: Danmarks Nationalbank loan-level mortgage data, Population register
  online appendix · Table B1, Figure B1

Replication package

• Net wealth Wealth, debt and consumption
  In the respondent vs non-respondent comparison (2015 values), wealth = real estate, deposits, stocks, bonds, mortgage deeds in deposit, cars and pension accounts minus all debt except debt to private persons; tax-assessed housing values adjusted by the average ratio of market prices to tax-assessed values among traded houses of the same property class, location and price range. Robustness variants: market-adjusted housing; adding car stock (available from 2014); adding pension wealth; household-level wealth; liquid financial wealth only (stocks, bonds, deposits).
  Registers: Third-party reported wealth registers (SKAT/Statistics Denmark)
  Codes: Car and pension wealth only available from 2014
  online appendix · Online Appendix Table A5 notes; Table A10
• Wealth percentile rank Wealth, debt and consumption
  Within cohort x year percentile rank in the sample; robustness versions use averages over 2013-2015, 2011-2015 and 2009-2015, and household-level wealth.
  Registers: Wealth registers
  online appendix · Online Appendix Figure A10 notes; Table A11
• Credit-constrained indicator Wealth, debt and consumption
  Holding liquid assets worth less than one month's disposable income (measured in 2014 and over 2001-2015).
  Registers: Third-party reported wealth and income registers
  Codes: Liquid assets < 1/12 of annual disposable income
  online appendix · Online Appendix Figure A8 notes
• Marginal interest rate (interest on marginal liquidity) Wealth, debt and consumption
  From tax-authority loan-level data: interest rate per loan r = interest paid in 2014 / (0.5 x (balance end-2013 + balance end-2014)); only non-mortgage loans, minimum denominator DKK 1,000, rates censored at the 5th and 95th percentiles. For respondents with loan accounts the marginal rate is the highest loan-specific rate; for respondents with only deposit accounts it is the smallest account-specific deposit rate. Distribution: p5 0.00, p25 0.97, p50 6.25, p75 12.73 percent.
  Registers: Danish Tax Agency (SKAT) account-level loan and deposit data, 2013-2014
  Codes: Non-mortgage loans; denominator >= DKK 1,000; censored p5/p95
  online appendix · Online Appendix C.4; Table A8
• Stock market return Wealth, debt and consumption
  (Dividend income + realised capital gains/losses during the year) / market value of stocks, averaged over 2008-2014; plus an indicator for owning stocks during the period.
  Registers: Tax-based income and wealth registers
  working paper · Section 3 (Table 2 col. 7 discussion)
• Parental wealth / initial wealth deciles Family links, fertility and household
  Parental wealth decile = parents' position in the parental wealth distribution measured within the respondent's cohort when the respondent was 18; initial wealth decile = own position in the within-cohort wealth distribution at age 18.
  Registers: Wealth register, Population register (parent links)
  working paper · Table 2 notes; Figure 3 notes
• Income controls Earnings and income
  Gross income = annual income excluding capital income (DKK).
  Registers: Income register
  online appendix · Online Appendix Table A5 notes
• Experiment sample Sample definition and timing
  3,620 respondents (3,552 with controls) with full register variables; robustness subsamples: both parents alive (2,367); in labour force every year and never unemployed 2011-2015 (2,265); stable relationship status 2011-2015 (2,704); socioeconomic status never indicating poor health 2008-2015 (3,030). Inverse probability weights based on respondents vs invited non-respondents or vs population, using education-year dummies, income deciles, income growth, parental wealth, wealth at age 18, age, gender, single and dependent-children dummies.
  Registers: Population register (BEF); income, wealth and education registers
  online appendix · Online Appendix Tables A5, A10, A11
• Patience index Other
  Patience groups by tertiles of the non-parametric patience measure with cut-offs Low [0.0,0.5], Medium [0.5,0.8], High [0.8,1.0]. Survey alternative (follow-up of 2,096 respondents) uses the 1973 Danish Longitudinal Survey of Youth job-profile question: average salary from start (low), low salary first two years then high (medium), very low salary first four years then very high (high).
  Registers: Online experiment linked on CPR
  Codes: Cut-offs 0.5 and 0.8
  online appendix · Online Appendix Figures A8-A10
• Years of education Education and test scores
  Robustness replaces years-of-schooling indicators with indicators for 59 general educational subject groups (e.g. comparative literature studies, economics, physics).
  Registers: Education register
  Codes: 59 educational groups
  online appendix · Online Appendix Table A11 notes

• Household marginal interest rate (2008) Wealth, debt and consumption
  Interest on marginal liquidity, computed from register data on all loan and deposit accounts of the household obtained from the Danish Tax Agency (SKAT), measured in 2008, in percent.
  Registers: Danish Tax Agency account-level loan and deposit data
  online appendix · Online Appendix Table EA.1 notes
• Propensity to spend the SP payout (survey) Wealth, debt and consumption
  Survey (Epinion) question: considering the increase in disposable funds from the SP payout, how did you allocate it: 1) increase spending, 2) increase free savings, 3) reduce debt, 4) increase pension savings; spending share in [0,1] (96% of answers 0 or 1). Main sample includes only respondents who withdrew the SP balance (N = 5,037); extended sample adds 215 non-withdrawers coded as zero spending. Probit robustness codes share >= 0.5 as 1.
  Registers: Telephone survey linked to registers on CPR
  Codes: Probit: 0 if share in [0,0.5], 1 if in [0.5,1]
  online appendix · Online Appendix sections 4, 7, 8 (survey instruments)
• SP payout / SP account balance Retirement and pensions
  SP payout is the (potential) payout, i.e. the Special Pension account balance less taxes; enters as SP payout / 2008 disposable income (percent).
  Registers: ATP Special Pension (SP) account register
  online appendix · Online Appendix Table EA.1 notes
• Register net savings Wealth, debt and consumption
  Change in net bank assets (bank deposits minus bank debt, including credit card debt, recorded on the last day of the year) between t and t-1 from the income-tax/wealth register; used to validate survey answers for 2005-2011.
  Registers: Income tax register (SKAT wealth data)
  working paper · Section 5.1
• Imputed spending Wealth, debt and consumption
  Disposable income minus net savings (change in net bank assets) minus private pension contributions, c = y - dW - p, following Browning and Leth-Petersen (2003).
  Registers: Income tax register
  working paper · Section 5.1
• Low liquid assets dummy Wealth, debt and consumption
  Indicator equal to one if the individual holds liquid assets worth less than one month's disposable income (threshold chosen for highest predictive power); a continuous liquid-assets-to-disposable-income ratio is also used.
  Registers: Income tax register
  working paper · Section 5.3.2
• High debt service dummy Wealth, debt and consumption
  Indicator equal to one if debt service is 15 percent or more of disposable income; also used as a continuous debt-service-to-disposable-income ratio.
  Registers: Income tax register, SKAT account-level loan records
  working paper · Section 5.3.2
• Holding a deposit balance Wealth, debt and consumption
  Individuals with total deposit-account balances exceeding one month's disposable income are labelled as having a balance (to net out salary accounts credited just before 31 December); used to define the group holding both assets and debt.
  Registers: SKAT account-level deposit records
  working paper · Section 5.3.1
• Liquid assets / disposable income Wealth, debt and consumption
  Liquid assets (deposits, shares and bonds) in 2008 as percent of 2008 disposable income; underlies the low-liquid-assets indicator.
  Registers: Statistics Denmark income and wealth registers
  online appendix · Online Appendix Table EA.1 notes
• Debt service / disposable income Wealth, debt and consumption
  Debt service, defined as total interest payments on loans, in 2008 as percent of 2008 disposable income; underlies the high-debt-service indicator.
  Registers: Statistics Denmark income registers
  online appendix · Online Appendix Table EA.1 notes
• Permanent income Earnings and income
  Average of (register) income over 1998-2008, entered in logs; income controls also include ln(2008 register income) and the standard deviation of ln(income).
  Registers: Statistics Denmark income registers
  Codes: Years 1998-2008
  online appendix · Online Appendix Table EA.1 notes
• Education level dummies Education and test scores
  Short, medium and long education correspond to vocational training, college BA level, and college Master and PhD level, respectively; reference group is high school or less.
  Registers: Statistics Denmark education register
  Codes: Short = vocational; medium = BA; long = Master/PhD; ref = high school or less
  online appendix · Online Appendix Table EA.1 notes
• Income and credit expectations (survey) Other
  Survey questions on whether total 2009 income evolved better/worse than expected (5-point scale plus don't know), changes in expected 2010 income, and changes in expectations about the possibility of borrowing money in 2010 (5-point scale from much more optimistic to much more pessimistic).
  Registers: Telephone survey
  Codes: 5-point Likert scales
  online appendix · Online Appendix section 8 (survey instruments)

• Imputed household consumption Wealth, debt and consumption
  Annual household consumption (spending) imputed from the budget constraint as C = Y - T - dA, where Y = gross income from all sources (labour and capital) + housing support + child support - implicit tax on the consumption value of owned housing, T = total taxes paid, and dA = change in assets (cash, bank deposits, stocks and shares, property value, cars and other vehicles, minus liabilities). Values in 2000 DKr.
  Registers: Tax records (income and wealth tax data)
  working paper · Section 2.1, eq. (1)
• Co-worker network (weighted) Firm characteristics and productivity
  Co-workers are individuals working in the same plant (for public employees, the physical address of the workplace) from IDA. Baseline weights favour similar workers: with five education groups and three occupation groups (blue collar, white collar, manager), degree of separation d = |E_i - E_j| + |O_i - O_j| and weight (d+1)^-2; peers' consumption is the weighted average of co-workers' household log consumption.
  Registers: IDA
  working paper · Section 2.1; Section 5.1, eq. (11)
• Distance-3 peers Family links, fertility and household
  Starting from households where both spouses work, distance-1 peers are co-workers, distance-2 peers are the working spouses of co-workers, and distance-3 peers are the co-workers of the spouses of co-workers. Couples working in the same plant are excluded and feedback cases (where own household would serve as instrument for the peer) are discarded.
  Registers: IDA, Tax records
  working paper · Sections 4.1.2 and 5.1
• Firm instruments (distance-3 firm shocks) Firm characteristics and productivity
  Weighted averages over distance-3 peers of their firm's size, firm-specific change in employment, firm type dummies (publicly traded, limited liability, other) and a public-sector dummy, all expressed in first differences. A separate set uses distance-3 peers' demographics (age, age squared, years of schooling, share women, occupation shares, kids 0-6 and 7-13).
  Registers: IDA
  working paper · Section 5.2 (text before Table 4); Table 1
• Estimation sample Sample definition and timing
  Households 1980-1996 whose head is aged 18-65 and where both spouses work as employees (not self-employed); households are dropped from the year one or both members stop working or become self-employed. Peers' consumption is computed using all workers, including singles and one-earner households.
  Registers: Tax records, IDA
  working paper · Section 2.1; Online Appendix Table A1 (selection steps)
• Visible / neutral / non-visible spending (DES) Wealth, debt and consumption
  Spending in the Danish Expenditure Survey is split using the Heffetz (2011) visibility index: visible = tobacco and alcohol, food away from home, clothing, furniture and home goods, electrics/appliances, vehicles, entertainment, books, education, personal care; neutral = food at home; everything else (insurance, rent, etc.) non-visible.
  Registers: DES (Danish Expenditure Survey)
  working paper · Section 2.2; Table A6

• Net wealth Wealth, debt and consumption
  Value of assets minus liabilities measured on the last day of the year from administrative tax return records (SKAT), 1983-2011; mostly third-party reported by banks and financial institutions. Assets include deposits, stocks, bonds, real estate and deposited mortgages; liabilities include bank debt, mortgage credit debt, card debt, student debt, debt to municipalities, unpaid taxes, etc. Pension funds are not included; self-reported items (cars, boats, etc.) are excluded after the 1996 abolition of the wealth tax.
  Registers: SKAT wealth records (via Statistics Denmark)
  working paper · Section 2 (Data)
• Real estate valuation Wealth, debt and consumption
  Public (tax-assessed) values of real estate are adjusted upwards each year using the average relative difference between selling price and public value of real estate traded that year, following Leth-Petersen (2010).
  Registers: SKAT wealth records
  working paper · Section 2, footnote 6
• Portfolio components Wealth, debt and consumption
  From 1995 onwards complete portfolio information on bonds, stocks, cash in banks, real estate, mortgage loans and the sum of other loans; childhood holdings summarised as bank deposits, securities and real estate.
  Registers: SKAT wealth records
  working paper · Sections 2 and 3.1
• Wealth percentile rank Wealth, debt and consumption
  Percentile rank computed within child age-year specific wealth distributions (parents ranked on total parental wealth observed at the same time); ties, notably zero wealth among young children, are randomly broken. Adult outcome is the child's rank at age 45 (27 years after age 18).
  Registers: SKAT wealth records, Population register (parent links)
  working paper · Section 3.3, footnote 11; Section 3.4
• Inferred wealth transfers (residual) Wealth, debt and consumption
  For each child, wealth at a given age minus the present value of earnings up to that age (real interest rates 0%, 3%, 6%), censored at zero, as a lower bound on transfers received.
  Registers: SKAT wealth records, Income registers
  working paper · Section 3.2, Table 1
• Gift-tax bunching measure Taxes and tax rates
  First time an individual (ages 1-12) has positive wealth, wealth at that point treated as the transfer; distance to the gift-tax basic allowance (DKK 58,700 in 2011) and to twice that amount (two donors) is plotted to look for bunching.
  Registers: SKAT wealth records
  working paper · Section 3.2, Figure 6
• PCD and PD samples Sample definition and timing
  PCD: pooled cross-section of all individuals aged 1-40 in 2000-2011 with parents alive (parent IDs registered for Danes born from 1960). PD: panel of the 1965 birth cohort with both parents alive in 2011, followed 1983-2011 (ages 18-46). Monetary values deflated by CPI (2011 prices).
  Registers: Population register (parent links), SKAT wealth records
  working paper · Section 2; Tables A1-A2

• Bank-account link (bank identity) Wealth, debt and consumption
  Year-end loan and deposit account balances reported by financial institutions to the tax authorities; the first four digits of each account number identify the bank branch, mapped to bank identity using hand-collected Nets lists of branch id-numbers for each year 2003-2011.
  Registers: Tax authority third-party account reports, Nets branch lists
  Codes: First four digits of account number = branch id
  working paper · Section 3.1, footnote 5
• Primary bank (2007) Wealth, debt and consumption
  If one bank relation in 2007, that bank; if multiple relations but a loan in only one, that bank; if loans in multiple banks, the bank with the largest loan balance; if no loans, the bank with the largest deposit balance. Always a bank, never a mortgage institution.
  Registers: Tax authority third-party account reports
  working paper · Section 3.3 and footnote 6
• Exposed bank Firm characteristics and productivity
  Individuals ordered by the 2007 loan-to-deposit ratio of their primary bank and split at the median individual: above-median primary bank = exposed (approximately equal numbers in each group).
  Registers: Danish Central Bank bank balance sheets
  working paper · Section 3.3
• Imputed spending Wealth, debt and consumption
  Spending = disposable income minus change in net wealth. Disposable income = total realized income - taxes - interest payments + payments from private pension accounts (net of taxes) + imputed capital gains on stocks (start-of-period stock value times growth in stock price index). Net wealth = market value of securities + savings and checking accounts - mortgage loans - bank debt - other loans; real estate excluded (so spending includes real estate purchases). Main regressions exclude stock owners.
  Registers: Income registers, Wealth registers
  online appendix · Online Appendix Table A1; WP Section 3.2
• Pension withdrawals (borrowing-constraint indicator) Wealth, debt and consumption
  Withdrawals (liquidations) from tax-favoured pension savings accounts, which carry a 60% penalty before pension age, used as an indicator of borrowing constraints.
  Registers: Income and wealth registers
  working paper · Section 5 (Table 4, col. 3)
• Car ownership Wealth, debt and consumption
  Number of cars owned, indicators for owning at least one and at least two cars, and average car weight, from the auto register (weight and production year of each registered automobile).
  Registers: Motor vehicle register
  working paper · Sections 3.1 and 6
• Baseline sample Sample definition and timing
  Residents 2003-2011 aged 20-50 in 2007, excluding self-employed and individuals whose 2007 primary bank failed in 2008-2011 (analysed separately), 25% random sample (about 440,000 individuals); DKK variables winsorized at 1st and 99th percentiles.
  Registers: Tax records, Population registers
  working paper · Sections 3.1 and 3.3
• Unemployment controls/outcome Unemployment and benefits
  Control: dummy for unemployment spells during 2006-2007; outcome: fraction of the year spent in unemployment.
  Registers: Labour market registers
  working paper · Section 4 controls; Table A-notes

Replication package

• Workplace (occupation-firm cell) Firm characteristics and productivity
  Workplace = all employees with the same 2-digit occupation code in the same firm; employer default contribution rate measured by the workplace-level median (robustness: workplace-level mode, using only workplaces with a unique mode).
  Registers: IDA; income tax register
  Codes: 2-digit occupation code x firm ID
  online appendix · Online Appendix A
• Employer contribution rate (default) Retirement and pensions
  Employer-sponsored pension contributions divided by taxable labor income, computed separately for capital (lump-sum) and annuity pension accounts; the workplace-level measure is the median contribution rate within the occupation-firm cell in a year (modes used as robustness). Main outcome is the change from t-1 to t.
  Registers: Income tax register (third-party reported pension contributions)
  working paper · Section 2.2; Section 3 eq. setup; Appendix A
• Share of workplace above top tax threshold Taxes and tax rates
  Fraction of employees in the occupation-firm cell with labor income above the top income tax threshold (about 49 percent of workers on average), interacted with year dummies with 1998 as baseline; the 1999 reform cut the deduction value for capital pension contributions from 59 to 45 cents per DKr for top-bracket taxpayers.
  Registers: Income tax register
  working paper · Section 2.1; Section 2.3; Section 3
• White-collar occupation Occupation and job hierarchy
  White-collar = 2-digit (D)ISCO occupation code with first digit 1-5 (1 leadership; 2 highest-level knowledge work; 3 intermediate knowledge; 4 office jobs; 5 retail sales, service and care work). Blue-collar/other = first digit 6-9. Main sample excludes public-sector workers and blue-collar occupations to avoid workplaces with collectively bargained contribution rates.
  Registers: IDA (occupation code)
  Codes: 2-digit DISCO first digit 1-5 = white-collar; full list 11,12,21-24,31-34,41,42,51,52 (white-collar); 61,71-74,81-83,91-93
  online appendix · Online Appendix E: Occupation Codes
• Analysis sample Sample definition and timing
  Employees in private-sector firms with white-collar occupations; excludes self-employed individuals and workplaces with fewer than five employees; years 1996-2001.
  Registers: Income tax register; IDA
  Codes: workplace size >= 5
  online appendix · Notes to Appendix Figures 1-2
• Individual private (IRA-type) pension contributions Retirement and pensions
  Individual contributions to private capital and annuity pension accounts outside the employer plan, as contribution rates relative to taxable labor income, used as the comparison to employer responses.
  Registers: Income tax register
  working paper · Section 2.3; Section 4
• Robustness subsample: no annuity contributions Sample definition and timing
  Workplaces whose median annuity contribution rate in 1996-1998 was zero and whose median capital (lump-sum) account contribution rate in 1998 exceeded 1.5 percent (4,056 observations).
  Registers: Income tax register (third-party reported pension contributions)
  Codes: median capital contribution rate 1998 > 1.5%
  online appendix · Online Appendix A, item 4
• Highly educated workplace Education and test scores
  Private-sector white-collar workplaces in which more than 20 percent of workers had more than 16 years of education (used as a group less likely covered by collective bargaining).
  Registers: IDA
  Codes: share with >16 years education > 20%
  online appendix · Online Appendix D

• Primary analysis sample Sample definition and timing
  Danish population 1995-2009 from merged income tax register, population register and IDA, excluding observations below age 18 or above 60 and observations with self-employment income (about 41 million observations on 4 million individuals).
  Registers: IND (income tax register), BEF, IDA
  working paper · Section III (Sample and Variable Definitions)
• Taxable saving (gross and net) Wealth, debt and consumption
  Gross taxable saving = change in third-party reported end-of-year assets in taxable accounts; net saving = gross saving minus change in liabilities, where liabilities cover secured and unsecured debt except home mortgages. Saving measures and rates below the 1st or above the 99th percentile in the estimation sample are set to missing.
  Registers: Third-party reported wealth data (tax authority)
  working paper · Section III
• Pension contributions by account Retirement and pensions
  Annual employer and individual contributions to capital pension (lump-sum) and annuity pension accounts as reported by employers and pension funds; contributions above the 99th percentile are top-coded at the 99th percentile in each account and year.
  Registers: IND (income tax register)
  working paper · Section III
• Income concepts Earnings and income
  Labor income = labor earnings before taxes excluding contributions to employer-administered pensions; total compensation = labor earnings plus employer and employee contributions to employer-administered accounts; taxable income = labor income plus other income (e.g., UI) affecting tax liability. Measured at the individual level, with joint-account balances split equally.
  Registers: IND (income tax register)
  working paper · Section III
• Firm switch Employment and labour force status
  Individual switches firms in year t if he has earnings from two distinct firms in t and t-1 (primary job for multiple jobholders); switches where earnings fell by more than half or more than doubled are excluded; set missing in 2003, 2005 and 2007 when firm identifiers changed. Event studies require at least 4 years of data before and after.
  Registers: IDA, IND
  working paper · Section IV.A, footnotes 9-10
• Mandatory Savings Plan (MSP) contribution Retirement and pensions
  From 1998 individuals with labor income at or above DKr 34,500 were required to contribute 1% of earnings (DKr 345 at the cutoff); RD on changes in pension contributions/saving from 1997 to 1998 for individuals within DKr 25,000 of the cutoff.
  Registers: IND (income tax register)
  Codes: cutoff Y >= DKr 34,500; bandwidth DKr 25,000
  working paper · Section IV.B, equation (6)
• 1999 capital pension subsidy reduction Taxes and tax rates
  From 1999 the deduction for capital pension contributions fell from 59 to 45 cents per DKr for individuals in the top income tax bracket (cutoff DKr 251,200 in 1998); treatment group: taxable income DKr 25,000-75,000 above the top-tax cutoff; control: DKr 25,000-75,000 below.
  Registers: IND (income tax register)
  working paper · Section V, V.A (Estimator 1)

• Imputed household total expenditure Wealth, debt and consumption
  Household disposable after-tax income minus the change in net financial wealth from t-1 to t, following Browning and Leth-Petersen (2003); deflated to 1990 prices. Possible only up to 1996 because wealth-tax register data end after the wealth tax was abolished in 1997.
  Registers: Income and wealth tax registers (10% sample)
  working paper · Section 4 (Expenditure imputation), eq. (1), footnote 1
• Household disposable income Earnings and income
  Gross income including interest income, housing transfers and dividends, minus taxes on income, taxes on shares, surplus on owner-occupied housing and tax-exempted interest expenses (labour, transfer and capital income net of personal taxes).
  Registers: Income tax registers
  working paper · Section 4 (Expenditure imputation)
• Net wealth Wealth, debt and consumption
  Financial assets (market value of shares, bank deposits, market value of bonds and securities) minus debt (mortgage debt, bank debt etc.).
  Registers: Wealth tax registers
  working paper · Section 4 (Expenditure imputation)
• Municipal house prices and house price innovations Geography and neighbourhood
  Average sales prices of traded single-family houses by municipality (275 municipalities), 1985-2001, collected by the tax authorities. Log municipal price modelled as an AR process with municipality effects, average house size (square metres, rooms) and year effects, estimated by system GMM; the predicted change is the expected price change and the residual is the house price innovation.
  Registers: Tax authority house sales statistics (Statistics Denmark)
  working paper · Section 4 (House prices); Section 5 (house price process); Appendix 1
• Analysis sample Sample definition and timing
  10% random sample of the Danish population aged 16+, 1987-1996; married or cohabiting stable couples who were either home owners or renters throughout; split before and after the 1992-93 mortgage/collateral and tax reforms.
  Registers: Income and wealth tax registers (10% sample)
  working paper · Section 4 (Data); Abstract
• Liquidity constraint Wealth, debt and consumption
  Liquidity-constrained households have liquid funds (cash and bonds) below 1.5 months of disposable income; high-liquidity households have liquid funds above half a year's disposable income. Mortgage sensitivity measured by mortgage debt / house value.
  Registers: Income and wealth tax registers
  working paper · Section 6 (credit constraints) and robustness section

• Imputed total household expenditure Wealth, debt and consumption
  Total expenditure imputed from tax registers as disposable income (taxable income plus some non-taxable components and transfers, minus final taxes) minus the change in net wealth from the end of one tax year to the end of the next, following Browning and Leth-Petersen (2003); capital gains are not netted out except for housing. Households with negative imputed expenditure are dropped.
  Registers: Income tax register, Wealth tax register
  Codes: c_t = y_t - dW_t (equation 14), with allowance for housing capital gains
  working paper · Section 4.1, equation (14); Section 4.2
• Wealth components Wealth, debt and consumption
  Third-party reported assets in six pre-reform categories (housing assets at tax-assessed cash value, shares, deposited mortgage deeds, cash holdings, bonds, other assets incl. self-reported non-deposited bonds, ship shares and high-value objects) and total liabilities; mortgage debt is separately available only up to 1993, so only total liabilities are consistent across years. Employer pension funds are not observed.
  Registers: Wealth tax register
  working paper · Section 4.2; Appendix figure A2
• Credit-constrained (low liquid assets) Wealth, debt and consumption
  Alternative splits used in robustness: low liquid assets = liquid assets of at most two months of disposable income in 1991; at most one month of disposable income in both 1990 and 1991; at most one month of disposable income in 1991 and subsequently experiencing an unemployment shock.
  Registers: Wealth tax register, Income tax register
  online appendix · Section 5 (3.1-3.3), alternative sample splits
• Credit reform exposure Wealth, debt and consumption
  The 1992 mortgage reform allowed mortgage loans to finance non-housing purposes, up to 60% of the house value, and extended the maximum maturity from 20 to 30 years; homeowners are treated and renters serve as a consistency check.
  Registers: Administrative dwelling registers
  working paper · Section 2; Section 4.3
• Household panel sample Sample definition and timing
  Robustness subsamples: renters (separate analysis) and house owners holding no financial assets (no stocks or bonds).
  Registers: Wealth tax register
  online appendix · Sections 2-3
• Propensity-score matching covariates Sample definition and timing
  Probit for low-liquid-asset status in 1991 uses: housing equity/house value and its square, ln(house value) and square (owners), ln(gross income) and square, age and square, single, UI membership, size of house and square, labour supply (male, female), self-employed (male, female), education short/medium/long, number of children dummies (1-4). Monetary values in DKK, 1990 prices. Matching with replacement; SEs per Abadie and Imbens (2006).
  Registers: Income tax register, Wealth tax register, Population registers
  online appendix · Tables A1, A4
• Household education level Education and test scores
  Highest level of education in the household, grouped into no/low vs medium/high for heterogeneity analysis.
  Registers: Education register
  online appendix · Section 4, Figure A6

• Imputed total household expenditure Wealth, debt and consumption
  Total expenditure imputed at the household level as disposable after-tax income minus the change in net wealth between end of year t-1 and end of year t (c = y - dW). Validated against total expenditure reported in the 1994-96 Danish Expenditure Survey linked to the households' register records. Relies on wealth-tax register information, which ends after the wealth tax was abolished in 1997.
  Registers: Income and wealth tax registers (Statistics Denmark), Danish Expenditure Survey 1994-96
  working paper · Section 4 (Expenditure imputation), eq. (1) and footnote 1 - restatement of the Browning and Leth-Petersen (2003) definition by the same authors
• Household disposable income Earnings and income
  Sum of gross income including interest income, housing transfers and dividends from share capital, minus taxes on income, taxes on shares, surplus on owner-occupied housing and tax-exempted interest expenses.
  Registers: Income tax registers
  working paper · Section 4 (Expenditure imputation) - restatement of Browning and Leth-Petersen (2003)
• Household net wealth Wealth, debt and consumption
  Assets minus liabilities measured at year end; assets are the market value of share capital, bank deposits and the market value of bonds and securities; liabilities include mortgage and bank debt.
  Registers: Wealth tax registers
  working paper · Section 4 (Expenditure imputation) - restatement of Browning and Leth-Petersen (2003)

• Hierarchy level (position) Occupation and job hierarchy
  Job level from the RAS socioeconomic job code: 131 = top manager, 132 = (function) manager, 133 = white-collar, 134 or 135 = blue/grey-collar. Only persons with job_socio_kode 131-135 are extracted from RAS (from 1994 on).
  Registers: RAS, IDA
  Codes: RAS job_socio_kode 131 (top manager), 132 (function manager), 133 (white collar), 134-135 (blue/grey collar)
  replication code · 3 Produce regression data.do, section 'INFORMATION ABOUT JOBLEVEL FROM RAS'; 1 Produce IDA project data.sas (RAS extract)
• Top management appointment (promotion) Occupation and job hierarchy
  prom1 = 1 if the person is a function manager (pos = 2, RAS code 132) in year t and a top manager (pos = 1, RAS code 131) in year t+1 (lead of position in the person-year panel). Logit models for prom1 are estimated on managers (pos = 2) only; in the multinomial version, promotions are split into internal (same workplace lbnr in t+1) and external (new workplace).
  Registers: RAS, IDAN
  Codes: pos==2 & F.pos==1 (job_socio_kode 132 -> 131); new_workplace = lbnr != F.lbnr
  replication code · 3 Produce regression data.do ('promotion categories'); 4 Tables and figures.do (Table 6 mlogit)
• Weekly working hours (actual and usual) Hours worked
  Actual and usual weekly hours from the Labour Force Survey (AKU), harmonised across file versions (hwactual/timuf/b7timuf/v51; hwusual/...); values >= 97 set to missing; only the first AKU observation per person (by year and reference week) is kept, years 1996-2018, and observations with zero or missing hours are dropped. Regressions use hours/100 and interval dummies (<37, 37, 38-47, >=48 hours).
  Registers: AKU
  Codes: h_actual, h_usual; missing if >= 97; dummies ha0_36, ha37, ha38_47, hagt48
  replication code · 2 Produce AKU data.do; 3 Produce regression data.do (interval dummies)
• Non-standard working time Hours worked
  Indicators from AKU for evening, night, Saturday, Sunday and shift work (value 9 = missing). d_frequent = 1 if evening, night, Saturday or Sunday work equals 1 (frequently); d_sometimes = 1 if any equals 2 (sometimes).
  Registers: AKU
  Codes: AKU evenwk/omaften/v205, nightwk/omnat/v206, satwk/loerdage/v207, sunwk/soendage/v208, shiftwk/skifteho/v204; 9 = missing
  replication code · 2 Produce AKU data.do; 3 Produce regression data.do ('work flexibility')
• Labour market experience Employment and labour force status
  Accumulated labour market experience from IDAP: erhverv_akk/1000 plus erhverv_1979/1000 (pre-1979 experience, set to 0 if missing), rounded to whole years.
  Registers: IDAP
  Codes: IDAP erhverv_akk, erhverv_1979
  replication code · 3 Produce regression data.do, 'EXPERIENCE FROM IDAP'
• Tenure and primary job Employment and labour force status
  Primary job from IDAN (job_type = 'H', workplace lbnr != 0); tenure = year minus ansaar (hiring year), set to missing if > 50 or < 0. Workplace size = number of IDAN primary-job employees per lbnr and year.
  Registers: IDAN
  Codes: IDAN job_type == "H", ansaar, lbnr
  replication code · 3 Produce regression data.do, 'TENURE AND WORKPLACE ID FROM IDAN'
• Schooling level Education and test scores
  Highest completed education from UDDA HFAUDD grouped: 1000-1999 high school or less; 2000-3999 vocational (IVU/erhvervsfaglig); 4000-4999 short cycle (KVU); 5000-5999 and 8500-8999 medium (MVU, 'Master'); 6000-8499 long (LVU); >=9000 or missing other. HFAUDD = 1 dropped; 3-digit codes multiplied by 10.
  Registers: UDDA
  Codes: HFAUDD ranges 1000-1999, 2000-3999, 4000-4999, 5000-5999 & 8500-8999, 6000-8499, >=9000
  replication code · 3 Produce regression data.do, 'EDUCATION FROM UDDA'
• Private sector Sector and industry
  private = 1 if RAS job function code (1996-2007) is 32, 42, 52, 80 or 90, or RAS sector code (2008-2016) is 18, 28, 32, 38, 42, 48, 52, 58, 62 or 81; 0 otherwise; job_funktion_kode 99 dropped.
  Registers: RAS
  Codes: JOB_FUNKTION_KODE in 32,42,52,80,90; JOB_SEKTOR_KODE in 18,28,32,38,42,48,52,58,62,81; drop JOB_FUNKTION_KODE 99
  replication code · 3 Produce regression data.do, RAS section
• Industry groups Sector and industry
  Workplace industry from IDAS: 6-digit DB77/93 code (DB03 after 2003) and DB07 code, mapped to about 12 groups (primary, manufacturing, utilities, construction, trade, transport, finance, real estate/business services, public administration, education, health/social, other) by code ranges, e.g. DB07 840000-849999 public administration, 850000-859999 education, 860000-899999 health and social.
  Registers: IDAS
  Codes: ARB_BRANCHE_DB7793 / ARB_BRANCHE_DB03 / ARB_BRANCHE_DB07 ranges as in do-file
  replication code · 3 Produce regression data.do, 'WORKPLACE FROM IDAS'
• Exporter status and sales Firm characteristics and productivity
  Firm sales (oms_firm) and total exports (eksport_ialt) from the IDA firm file; exporter = 1 if exports > 0. Merged by CVR number.
  Registers: FIRM
  Codes: oms_firm, eksport_ialt > 0
  replication code · 3 Produce regression data.do, 'EXPORT AND SALES FROM FIRM'
• Partner and children Family links, fertility and household
  Partner = 1 if civil status CIVST is 'G' (married) or 'P' (registered partnership); number of children and age of youngest/oldest child from the IDA family file (c_antboernf, c_aldyngst, c_aldaeldst), with dummies for youngest/oldest child aged 0-2, 3-4, 5+.
  Registers: IDA personer, IDA familie
  Codes: CIVST in G, P; c_antboernf, c_aldyngst, c_aldaeldst
  replication code · 3 Produce regression data.do, PERSONER and FAMILIE sections

Replication package

• Work accident with permanent earning-capacity loss Health outcomes, diagnoses and mortality
  Work accident reports from Labour Market Insurance (AES) with an estimated earning capacity loss (ECL) > 0, excluding psychological shocks; collapsed to person-year. Earning capacity loss is estimated upfront by AES. Milder accidents without permanent loss serve as placebo.
  Registers: AES
  working paper · Section 1.3; footnote 7; Appendix Table A.2
• Access to reskilling Education and test scores
  Indicator that the worker's vocational degree grants direct entry to relevant higher education programmes (e.g. carpenters, electricians, welders) versus comparable trades that require three years of high school first (e.g. blacksmiths, iron and metal workers, manufacturing technicians). Reskilling = participation in higher education by year +10.
  Registers: UDDA
  working paper · Section 2.2; Figure 1 notes
• Antidepressant use Health care use
  Dummy indicating a prescription for drugs in ATC class N06A (Antidepressants) in the year.
  Registers: Prescription Drug Database (LMDB)
  Codes: ATC N06A
  working paper · Appendix Table A.1, row 19
• Opioid use Health care use
  Dummy indicating a prescription for drugs in ATC class N02A (Opioids).
  Registers: Prescription Drug Database (LMDB)
  Codes: ATC N02A
  working paper · Appendix Table A.1, row 25
• Alcohol diagnoses Health outcomes, diagnoses and mortality
  Accumulated number of hospital diagnoses related to alcohol following Browning and Heinesen (2012): alcohol poisoning, addiction syndrome, delirious abstinence, alcohol psychosis, oesophageal varices, alcoholic liver disease, alcoholic pancreatic disease.
  Registers: LPR
  Codes: ICD-10 F100, F102, F104, F105, I85, K70, K860, T500, T510
  working paper · Appendix Table A.1, row 24
• Cardiovascular / respiratory diagnoses; suicide attempts Health outcomes, diagnoses and mortality
  Accumulated number of hospital diagnoses in ICD-10 chapter I (circulatory) or J (respiratory); accumulated hospitalisations where the reason for contact is registered as a suicide attempt (or accident).
  Registers: LPR
  Codes: ICD-10 chapter I; ICD-10 chapter J
  working paper · Appendix Table A.1, rows 21-23, 26
• Full-time and employment Employment and labour force status
  Full-time = full-time work according to RAS, defined by normal working hours above 32 hours/week. Employment = full-time work while not enrolled in a post-secondary programme and not receiving DI (DI = any weeks in the year on disability insurance). Stable pre-injury employment = full-time in all three years before the accident.
  Registers: RAS, IDA
  working paper · Appendix Table A.1, rows 10, 15-17; Section 1.3
• Hours worked (FTE) Hours worked
  2008-2017: yearly number of paid hours; 1995-2007: yearly labour income divided by hourly wage rate (November job). Divided by 1,924 full-time-equivalent yearly hours (37 hours/week).
  Registers: IDA
  working paper · Appendix Table A.1, rows 11-12
• Labour income Earnings and income
  Total labour market income including bonuses, fringe benefits, wages paid during sick and parental leave, and employer pension contributions.
  Registers: IDA
  working paper · Appendix Table A.1, row 9
• Stable partner / partner retention Family links, fertility and household
  Stable partner = cohabiting (including married) with the same partner for 3 consecutive years; partners are the initial partner in the year before the accident, followed regardless of separation; outcome = share still in the relationship.
  Registers: BEF
  working paper · Appendix Table A.1, row 3; Section 2.1; Figure 3 notes
• Physical-cognitive ability gap Occupation and job hierarchy
  ln(Physical Ability Requirements / Cognitive Ability Requirements) of the occupation, using O*NET importance of strength and stamina items vs. reasoning and idea items; both normalised.
  Registers: IDA, O*NET
  working paper · Appendix Table A.1, row 18
• Union membership Employment and labour force status
  Dummy for a positive deductible union-fee amount reported by unions to the tax authorities.
  Registers: IND
  working paper · Appendix Table A.1, row 14
• Education level and years of schooling Education and test scores
  Years of schooling = prescribed years of study of highest completed degree counting from grade 1. Highest completed degree groups: Primary (pre-school, primary, preparatory courses, Danish language courses at language centres); Vocational (VET, qualifying programmes, labour market educations AMU); Secondary (upper secondary); Post-secondary (short-cycle higher, vocational bachelor, bachelor, master, PhD). School = enrolled in a post-secondary programme in the year.
  Registers: UDDA
  online appendix · Supplementary Appendix, Table A.1 Panel B-C
• Hourly wage Earnings and income
  2008-2017: yearly total labour income divided by yearly hours worked; 1995-2007: average hourly wage rate in the November job, reliable only for individuals working more than 20 hours a week.
  Registers: IDA
  online appendix · Supplementary Appendix, Table A.1 Panel C
• Labour market experience and DI receipt Employment and labour force status
  Labour market experience measured in years since 1964. DI = any weeks within the year spent receiving disability insurance.
  Registers: IDA
  online appendix · Supplementary Appendix, Table A.1 Panel C
• Hospital visits, accidents and mortality Health care use
  Hospital visits = number of hospital contacts (admissions, outpatient treatment and ER visits). Accidents = accumulated hospitalisations with reason for contact registered as an accident. Mortality = accumulated, individual counted as deceased from the year of death.
  Registers: LPR, Death register
  online appendix · Supplementary Appendix, Table A.1 Panel E
• Work accident sample and matched controls Sample definition and timing
  Universe of work accidents causing earning-capacity loss 1998-2017, restricted stepwise: ECL > 0; exclude psychological shock; collapse to person-year; person exists in register data; full-time employed before injury; vocational degrees with (or comparable without) access; male craft workers. Each injured worker matched to a random non-injured worker with the same age, gender, education, occupation, industry and access to higher education in the year before the accident (plus stable partner and parent indicators); controls inherit the year of injury of their match.
  Registers: AES, IDA, BEF, UDDA
  online appendix · Supplementary Appendix, Table A.1 Panel F and Table A.2
• Inverse probability weights for access Sample definition and timing
  Logit propensity score for access to higher education estimated on injured workers at year e-1 using age, gender, stable partner, years of schooling, labour income, hours worked, hourly wage, experience, union membership, physical-cognitive ability gap, antidepressant prescription dummy and year of injury; No-Access workers weighted by p/(1-p) (Abadie 2005), and the matched control receives the same weight.
  Registers: IDA, UDDA, LMDB, BEF
  online appendix · Supplementary Appendix B.1

• Main monthly employer (BFL) Employment and labour force status
  From monthly eIndkomst job records (BFL), each person-month is assigned the job with the most paid hours (AJO_LOENTIMER) as the primary employer; its CVR number, wage and hours are kept. Month and year come from the job start date field AJO_JOB_START_DATO.
  Registers: BFL
  Codes: AJO_CVR_NR_FRA_PROD_JOB, AJO_SMALT_LOENBELOEB, AJO_LOENTIMER, AJO_JOB_START_DATO; primary job = max(AJO_LOENTIMER) within pnr-year-month
  replication code · script/cleaning/1_c_monthly_data.R
• Stayer (same employer Dec 2020 to Dec 2021) Employment and labour force status
  Among workers with an employer and monthly wage above 2,000 DKK in December 2020, stay=1 if the main employer in December 2021 is the same as in December 2020, else 0. Sample restricted to ages 20-65 and 2020 annual BFL earnings of at least 24,000 DKK.
  Registers: BFL
  Codes: wage>2000 in Dec 2020; employer_g == F12.employer_g; sum_earnings>=24000
  replication code · script/analysis/fig4/preprocess_fig4a.do; paper Figure 4 note
• Months out of work after separation Unemployment and benefits
  A separation is a month in which the worker has a main employer with monthly wage above 2,000 DKK and next month's employer differs; only the first separation per person is used. Time out of work = months until a different non-missing employer appears (capped at 24) minus one, so job-to-job moves count as 0 and returns to the same employer are not reemployment. The paper describes separations in 2020; the posted do-file's window (date<=24 from Dec 2018) covers Dec 2018-Dec 2019.
  Registers: BFL
  Codes: jan_sep = employer_g != F1employer_g; reemp = first k with F`k'employer_g != employer_g & non-missing; gap = reemp-1; cap 24
  replication code · script/analysis/fig4/preprocess_fig4b.do; paper Section 3.2 and Figure 4 note
• Annual earnings (register) Earnings and income
  Annual earnings are wage income LOENMV_13 from the income register (IND); self-employment income is NETOVSKUD_13 kept separately. Realized earnings growth 2020-2021 is ln(earnings_2021*100/102) - ln(earnings_2020), dropping observations with earnings below 24,000 DKK in either year and trimming at the 1st/99th percentile.
  Registers: IND
  Codes: LOENMV_13 (earnings), NETOVSKUD_13 (self-employment income); threshold 24,000 DKK; deflator 100/102; winsor2 trim 1/99
  replication code · script/cleaning/1_b_annual_data.R; script/analysis/fig7/preprocess_fig7.do; paper Figure 6 note
• Net wealth and liquid wealth Wealth, debt and consumption
  Net wealth = (KOEJD + BANKAKT + KURSAKT + OBLAKT + PANTAKT) - (BANKGAELD + GAELD_TIL_OFF + OBLGAELD + PANTGAELD), missing public debt set to 0. Liquid wealth = BANKAKT + KURSAKT + OBLAKT + PANTAKT. Pension wealth is the sum of PF_FORMUE across contracts in PENSFORM.
  Registers: IND, PENSFORM
  Codes: KOEJD, BANKAKT, KURSAKT, OBLAKT, PANTAKT, BANKGAELD, GAELD_TIL_OFF, OBLGAELD, PANTGAELD; PF_FORMUE
  replication code · script/cleaning/1_b_annual_data.R
• Monthly UI, cash benefits and retirement transfers Unemployment and benefits
  From ILME monthly non-employee e-income records, amounts across A/B income fields are summed per person-month and classified by the income-type code of the largest payment: UI if VMO_INDK_ART codes 2,5,8,16,104; cash benefits if type code 6 or art codes 40,41,43,84,87; retirement if art codes 6,9,10,13,20,24,25,44,45,74,83,86,113,115.
  Registers: ILME
  Codes: VMO_INDKOMST_ART_KODE UI: 2,5,8,16,104; CB: VMO_INDKOMST_TYPE_KODE 6 or ART 40,41,43,84,87; Retirement ART: 6,9,10,13,20,24,25,44,45,74,83,86,113,115
  replication code · script/cleaning/1_c_monthly_data.R
• Public vs private sector Sector and industry
  Main job (highest LOENTIMER in IDAN per year) is public if ARB_SEKTORKODE is non-missing and not 99, and recoded private if ARB_SEKTORKODE is in 18, 28, 89, 81, 58, 48, 42, 52. Industry groups come from DB07 ranges of ARB_HOVED_BRA_DB07 (e.g. >=840000 public services, 850000 teaching, 860000 health).
  Registers: IDAN
  Codes: ARB_SEKTORKODE private: 18,28,89,81,58,48,42,52; ARB_HOVED_BRA_DB07 6-digit DB07 ranges
  replication code · script/cleaning/1_b_annual_data.R
• Occupation group Occupation and job hierarchy
  Main job's DISCO_KODE first digit mapped to 9 groups (1 management, 2 high knowledge, 3 medium knowledge, 4 office, 5 sales/services, 6 farming, 7 craft, 8 skilled, 9 unskilled); military (DISCO below 100000) dropped. Tenure = year - ANSAAR.
  Registers: IDAN
  Codes: DISCO_KODE first digit 1-9; DISCO_KODE<100000 excluded; ANSAAR
  replication code · script/cleaning/1_b_annual_data.R
• Education level Education and test scores
  Highest completed education HFAUDD (latest available year) mapped through the audd format's H1 level: 10 elementary; 20, 25 high school; 30, 35 vocational; 40, 50, 60 bachelor; 65, 70 master.
  Registers: UDDA
  Codes: HFAUDD via uddan_2014_audd H1: 10; 20,25; 30,35; 40,50,60; 65,70
  replication code · script/cleaning/1_b_annual_data.R
• Population survey weights Sample definition and timing
  Inverse predicted propensity scores from a probability model of survey participation estimated on 2020 administrative data on the demographics of Danes active in the labour market.
  paper · Footnote 10 (details in Online Appendix B)

Replication package

• Labour market participation, part-time and full-time work Employment and labour force status
  Participation means at least 481 annual hours (e.g. 37 hours for 13 weeks) on BFL pay slips together with annual income of at least 50,000 DKK; part time is 481-1,664 annual hours; full time is more than 1,664 hours. Monthly hours and income are scaled down by the share of the month registered on parental leave in the OF register.
  Registers: BFL (eIndkomst), OF
  Codes: thresholds 481 and 1,664 annual hours; 50,000 DKK
  working paper · Section 2; Supplemental Material A
• Annual labour income (BFL) Earnings and income
  12-month sum of monthly labour income from BFL pay slips, by calendar year for age profiles and centred on the birth month for event-study moments (e.g. income from January-February before a March birth counts as year -1); excludes income registered while on parental leave; deflated by CPI, in 1000 DKK. Estimation sample 2010-2018.
  Registers: BFL (eIndkomst), OF
  working paper · Section 2; Supplemental Material A
• Personal income (tax sample) Earnings and income
  Personal income from annual tax data: labour income from wage work, profits from own firms and labour market transfers; deflated by CPI.
  Registers: IND
  Codes: PERINDKP
  working paper · Supplemental Material A
• Couples / partners Family links, fertility and household
  Partners follow Statistics Denmark's family definition, including married and cohabiting couples; cohabitants are two adults at the same address registered as parents to a child, or two opposite-sex adults with an age difference under 15 years at the same address. Constructed from EFALLE in BEF; couple age is the woman's age. Estimation sample keeps couples with at most five years age difference.
  Registers: BEF
  Codes: EFALLE; FOED_DAG; KOEN
  working paper · Section 2, footnote 8; Supplemental Material A
• High skilled vs less skilled Education and test scores
  High skilled if at least 180 months of education (at least a bachelor's degree), measured as the highest degree earned in the sample period; less skilled otherwise.
  Registers: UDDA
  Codes: HFPRIA >= 180 months
  working paper · Section 2; Supplemental Material A
• Labour market experience Employment and labour force status
  Statistics Denmark's experience measure based on accumulated mandatory ATP pension payments, which depend on hours worked; part-time employment counts as 2/3 of full time.
  Registers: IDAP
  Codes: ERHVER, ERHVER79
  working paper · Section 2; Supplemental Material A
• Sample exclusions by main activity Sample definition and timing
  Individuals aged 25-60 with an opposite-sex partner; observations dropped if mainly student, self-employed, retired or on disability pension, based on main activity at end of November.
  Registers: IDAP/AKM (socio-economic status)
  Codes: SOCSTIL, SOCSTIL_KODE
  working paper · Section 2; Supplemental Material A
• Mechanical net-of-tax rate change (instrument) Taxes and tax rates
  Four-year mechanical change log(1 - tau_{t+4}(z_t)) - log(1 - tau_t(z_t)) in the marginal net-of-tax rate, holding income and characteristics at year-t values, computed with a Danish tax simulator; used to instrument own and partner net-of-tax rate changes and virtual income. Restricted to couples where both have personal income of 50,000-600,000 DKK in the base year. Outcome is the 4-year difference in number of children born.
  Registers: IND
  working paper · Section 3; Supplemental Material A.1
• Household wealth Wealth, debt and consumption
  Bank deposits, bonds, stocks and property net of loans and mortgages at end of year; registered property values raised by 10% (following Leth-Petersen 2010); deflated by CPI, in 1000 DKK.
  Registers: IND
  Codes: FORM; FORMREST_NY05 (after 1996)
  working paper · Supplemental Material A
• Child linkage and age of children Family links, fertility and household
  Children and their ages are linked to parents through the parent identifiers in BEF; almost all children can be linked to a mother.
  Registers: BEF
  online appendix · Supplemental Material, Data Appendix (p. S2)
• Estimation sample vs tax sample Sample definition and timing
  Two samples: the BFL estimation sample of all individuals aged 25-60 with partners, 2010-2018; and the tax sample of women aged 25-40 with partners, 2004-2018 (used with personal income PERINDKP). Both split by less skilled vs high skilled.
  Registers: BFL (eIndkomst), IND, BEF
  online appendix · Supplemental Material, Table A.1

• First sudden parental death (treatment) Health outcomes, diagnoses and mortality
  Sudden deaths = deaths from heart diseases, cerebrovascular diseases, and traffic or other (external) accidents; non-sudden = all remaining causes. Sample: first parental deaths 1985-2014 for children aged 25-50 in the year of the first parental death with two known parents.
  Registers: DODSAARS/DODSAARG, BEF
  Codes: cause categories only (no ICD list given in appendix); child age 25-50; deaths 1985-2014
  online appendix · Supplemental Appendix Figure B.2 and B.5 notes
• Matched control group Sample definition and timing
  Alternative control group following Fadlon and Nielsen (2021): individuals treated 6 years later serve as controls. Spillover analysis on in-laws restricts to married/cohabiting individuals matched to married/cohabiting controls who lose no parent-in-law in the window, also matching on whether own parents are alive. SEs clustered at individual-by-match ID.
  Registers: BEF, DODSAARS/DODSAARG
  online appendix · Supplemental Appendix A.C, A.F
• Indexed earnings Earnings and income
  Earnings from employment and self-employment, inflated to 2020 levels and indexed by the average earnings of men (women) one year before the parental death.
  Registers: IND
  online appendix · Supplemental Appendix Figure B.5 and B.9 notes
• Participation and hours proxy (ATP) Hours worked
  Participation = strictly positive ATP pension contributions; intensive margin = amount of ATP contributions (paid proportionally to hours worked), following Kleven, Landais and Sogaard (2019).
  Registers: ATP contributions (income registers)
  Codes: ATP > 0
  online appendix · Supplemental Appendix Figure B.5 notes
• Psychologist and psychiatrist visits Health care use
  Psychologist visit data available 1990-2019; GP visits also used as an outcome.
  Registers: SYSI/SSSY
  online appendix · Supplemental Appendix Figures B.9-B.10 notes
• Mental health prescriptions Health care use
  Any prescription classified as mental-health related: ATC codes N05*, N06A*, N06B*, N06C*. Prescription data available from 1995; effects measured within 5 years after first parental death.
  Registers: LMDB
  Codes: ATC N05*, N06A*, N06B*, N06C*
  online appendix · Supplemental Appendix Figure B.10 notes
• Opioid and alcohol-dependence prescriptions Health care use
  Opioid painkillers: ATC N02A*; prescriptions for treatment of alcohol addiction: ATC N07BB. Prescription data from 1995.
  Registers: LMDB
  Codes: ATC N02A*; ATC N07BB
  online appendix · Supplemental Appendix Figures B.9-B.10 notes
• Hospital visits Health care use
  In- and outpatient visits at non-psychiatric hospital wards (data 1994-2018). Split into self-screening visits, defined by ICD-10 Z00 (general examination and investigation of persons without complaint and reported diagnosis) and Z01 (other special examinations and investigations of persons without complaint or reported diagnosis), including subcategories, versus non-screening diagnoses. Surviving parent's health status is based on their hospital visits in the 5 years before the first parental death.
  Registers: LPR_ADM, LPR_DIAG
  Codes: ICD-10 Z00*, Z01* (screening)
  online appendix · Supplemental Appendix Figures B.10, B.11, B.15 notes
• Formal childcare uptake Family links, fertility and household
  Uptake of formal childcare for children under age 5 and above age 5; only 2005-2019 used due to low coverage before 2005.
  Registers: DAGI/BOERNFB
  online appendix · Supplemental Appendix Figure B.14 notes
• Parent-child residential distance Geography and neighbourhood
  Four categories: same municipality (99 municipalities); same NUTS3 region but not municipality (11 NUTS3 regions); same NUTS2 region but not NUTS3 (5 NUTS2 regions); greater distance.
  Registers: BEF
  Codes: municipality / NUTS3 / NUTS2
  online appendix · Supplemental Appendix A.E

• Daycare availability in 1970 parish (treatment) Geography and neighbourhood
  Mothers are placed in their parish of residence in the 1970 Census. vg = 1 if a daycare institution for ages 0-2 (nursery or integrated institution, from digitised institution data) existed in the parish in the year of first birth; bh = same for ages 3-6 (kindergarten); availability is also tracked for each year from child age 0 to 6 (all/some/no years).
  Registers: FT70 (1970 Census, parish), Daycare institution yearbooks/registers
  Codes: institution number NUMMER < 120000 = nursery (vuggestue), > 250000 = integrated; parish (SOGNNR) x year; vg, bh, daycare0-daycare6
  replication code · dofiles/createfirst.do, 'FOLKETÆLLING 1970', 'DAYCARE 0-2', 'DAYCARE 3-6'
• Mother sample and timing of first birth Sample definition and timing
  Native (IE_TYPE 1) women born 1920-1959 with a first child born 1964-1975, aged at least 13 at first birth, with a 1970 parish observed and surviving until the first child is 16. First-birth year is the earliest birth year among children linked through MOR_ID (supplemented by family-register child counts).
  Registers: BEF, Fødselsdata (birth data), FAMILIE, FT70
  Codes: IE_TYPE == 1; MOR_ID; yrfirst in 1964-1975
  replication code · dofiles/createfirst.do, 'SAMLING AF BEF...' block
• Hours (ATP-based) and full-time employment Hours worked
  Annual ATP pension contributions (sum of orer/100 per person-year) are divided by the full-time contribution amount for the year (194.4 up to 1964, 259.2 up to 1971, 302 in 1972, 432 in 1973-1981, 1166.4 from 1982) and capped at 1. Full-time = ratio of 1; for workers the ratio is converted to weekly hours as 9 + 26 x ratio. Measured for child ages 0-35; missing set to 0 if the mother is alive.
  Registers: ATP
  Codes: ATP field ORER; divisors 194.4/259.2/302/432/1166.4 by year; full = hours>=1
  replication code · dofiles/createfirst.do, '*ATP' block and final forvalues x=0/35 loop
• Labour force participation by child age Employment and labour force status
  For child ages 0-15, participation = positive ATP contributions in the year (hours > 0); for ages 16-40, participation = positive wage earnings (non-missing log earnings), set to 0 if missing and the mother is alive.
  Registers: ATP, IND
  Codes: work = hours>0; xwork = learn (loglon non-missing) for ages 16-40
  replication code · dofiles/createfirst.do, final loops
• Log annual earnings Earnings and income
  Wage income (LONIND, replaced with corrected wage income CORLOEN in 2010-2015) converted to 2016 DKK with fixed year factors and to US dollars (6.7327), negative values set to 0, top-coded at 400,000, and logged (also inverse hyperbolic sine); measured child ages 5-40. Gross income (BRUTTO) is treated the same way for 1976 onward.
  Registers: IND
  Codes: LONIND, CORLOEN, BRUTTO
  replication code · dofiles/createfirst.do, 'LØN INDKOMST' and 'BRUTTO INDKOMST'
• Mother's education before first birth Education and test scores
  Years of education (normed months/12) completed by the year of first birth, taken from the highest, vocational or general education fields with completion dates before the first birth (or inferred from age 5 + length), with self-reported sources dropped and minimum 7 years; grouped into basic (main level 10), medium (20, 25, 35) and high.
  Registers: UDDA
  Codes: HFPRIA, EKPRIA, ALMPRIA; HFAUDD, ERHAUDD, ALMAUDD; HF_VFRA, ERH_VFRA, ALM_VFRA; drop HF_KILDE in 3, 9, 10, 17, 18; H1 10 = basic, 20/25/35 = medium
  replication code · dofiles/createfirst.do, 'UDDANNELSE FØR FØRSTE BARN'
• Separation from the first child's father Family links, fertility and household
  divor = 1 if the mother's partner ID (C_FAELLE_ID) in the population register when the first child is 16 differs from the first child's father ID (FAR_ID); missing if father unknown.
  Registers: BEF
  Codes: C_FAELLE_ID at age 16 of first child vs FAR_ID
  replication code · dofiles/createfirst.do, 'DIVORSE' block and divor definition
• Fertility and birth spacing Family links, fertility and household
  Number of children (more1-more4 indicators for at least 2-5 children) and time from first to second birth (space1-space3: second birth within 1, 2, 3 years).
  Registers: BEF, FAMILIE
  Codes: numkid; yrnext - yrfirst
  replication code · dofiles/createfirst.do; resfirst_final_EJ.do, Table 7
• Early retirement / pension Retirement and pensions
  From IDA labour market position, early retirement pension = PSTILL 50, 55 or 56 and old-age pension = PSTILL 92, 93 or 94; first year observed used to compute pension age.
  Registers: IDA
  Codes: PSTILL 50, 55, 56 (early pension); 92, 93, 94 (pension)
  replication code · dofiles/createfirst.do, '* PENSION'

Replication package

• Experimental sample Sample definition and timing
  Full stock of individuals registered as unemployed and receiving UI benefits in Denmark at the start of the experiment on March 15, 2019 (92,098 individuals); welfare recipients and later entrants excluded. Outcomes measured over windows after entrydate.
  Registers: STAR experiment data
  Codes: entrydate == mdy(3,15,2019)
  replication code · _README.pdf (Subject pool); code/data_preparation.do
• Total working hours within 6/12 months Hours worked
  Sum of paid hours (AJO_LOENTIMER) over all BFL monthly job records whose AJO_JOB_START_DATO falls after entrydate and within q x 30.42 days (q=6, 12); zero if no records.
  Registers: BFL
  Codes: AJO_LOENTIMER summed for AJO_JOB_START_DATO in (entrydate, entrydate+q*30.4167]
  replication code · code/data_preparation.do (Monthly labor market outcomes; hours4_total)
• Total earnings within 6/12 months Earnings and income
  Sum of AJO_SMALT_LOENBELOEB over BFL records with job start date in the window after entrydate, zero if none.
  Registers: BFL
  Codes: AJO_SMALT_LOENBELOEB (earnings4_total)
  replication code · code/data_preparation.do
• Monthly employment rate Employment and labour force status
  emp_q = 1 if cumulative BFL hours increase between month q-1 and q after entry (emp1 = any hours in month 1); employment rate over 6 or 12 months is the mean of these monthly indicators.
  Registers: BFL
  Codes: emp`x' = (hours4_total`x' - hours4_total`x-1' > 0); emp_rate12 = rmean(emp1..emp12)
  replication code · code/data_preparation.do
• Employment in recommended / stored occupations Occupation and job hierarchy
  Hours, earnings and monthly employment in BFL jobs whose DISCO code (VMO_DISCO_KODE truncated to 4 digits) matches one of up to five recommended occupations (Alternative_DISCO mapping of the job seeker's profile occupations) or an occupation stored in the job search profile (jsjs CVFAGID), within 183 and 365 days of entry.
  Registers: BFL, STAR jsjs
  Codes: floor(VMO_DISCO_KODE/100) == floor(forslag/1000) or floor(discoams_key/1000)
  replication code · code/data_preparation.do (occupations_rec, occupations_stored)
• Share of applications to recommended occupations Occupation and job hierarchy
  Among applications logged in Joblog within 1 or 12 months (30.25-day months) after entry, the share whose DISCO-AMS occupation equals a recommended occupation (or one stored in the profile).
  Registers: STAR Joblog
  Codes: share_recommended_t1/t12, share_stored_t1/t12
  replication code · code/data_preparation.do
• Labour market tightness Unemployment and benefits
  Vacancies divided by job seekers per municipality x occupation in the experiment's starting week (week 1442), averaged over the occupations in the individual's search profile.
  Registers: STAR vacancies, STAR jsjs
  Codes: tightness = no_vacancies/jobseekers, week==1442
  replication code · code/data_preparation.do
• Treatment intensity Geography and neighbourhood
  Baseline: municipality regime (10 super-control, 44 medium with 60% treated, 44 high with 90% treated). Alternatives: share treated in own plus bordering municipalities; commuting-weighted shares using the fraction of workers residing in j who commuted to m in the three prior years; and an individual measure over zip-code x 3-digit-occupation markets applied to in the two prior years.
  Registers: BFL, STAR Joblog
  Codes: geogroup 1/2/3; intensity_alt1
  online appendix · OnlineAppendix.pdf Section B.1; _README.pdf; code/data_preparation.do
• Pre-intervention labour market history Earnings and income
  From BFL records with period start before entry: mean monthly wage (lon) and employment degree (grad) in each of the three years before entry (year t-1 excludes the last 30 days), set to zero if absent; last pre-intervention occupation is the 1-digit DISCO of the most recent BFL spell.
  Registers: BFL
  Codes: wage_tminus1-3, grad_tminus1-3; last_occ = floor(VMO_DISCO_KODE/100000)
  replication code · code/data_preparation.do

Replication package

• Manager identification Occupation and job hierarchy
  A worker is a manager if their ISCO-88 occupation is 1200-1399 or their job position code is top management (pstill 31); a manager status is carried forward within an establishment in years where it has no identified manager. If no manager is found, the top earner (or anyone in the top 5% of earners) is flagged when not in a non-managerial ISCO>=2000 occupation, earns at least 1.6 times the average of co-workers, or earns more than 1.5 times the second-highest earner; further fallbacks assign the top earner with a unique occupation/position earning >1.3 times the second highest.
  Registers: IDA
  Codes: isco 1200-1399; PSTILL==31 (top manager), PSTILL 32 exception; ratio = lonind*(n-1)/(wagebill-lonind) >= 1.6; lonind/second-highest > 1.5
  replication code · clean_manager.do
• Top manager per firm-year Occupation and job hierarchy
  Among identified managers, one manager per establishment-year is kept by successively preferring ISCO 1210 (CEOs), then pstill 31, managers with longer tenure in the establishment, and higher within-establishment earnings rank; only managers observed in at least two firms are retained (mobility needed for identification), establishments that were acquisition targets are excluded, and at the firm (CVR) level the manager of the largest establishment is used.
  Registers: IDA
  Codes: isco==1210; pstill==31; target_cumu>=1 dropped
  replication code · clean_manager.do
• Log hourly wage (AKM outcome) Earnings and income
  Log of the IDA hourly wage, deflated to constant prices; when the hourly wage is zero it is replaced by annual labour income divided by 1800. Observations with missing firm ID, labour income <=1, missing education or experience, or unemployment degree above 500 are dropped; firms whose maximum employment is 5 or fewer are excluded.
  Registers: IDA
  Codes: timelon; lonind; replace timelon=lonind/1800 if timelon==0; arledgr (unemployment degree) >500 dropped; maxemp<=5 dropped
  replication code · AKM_merger_cvrnr.do
• Mergers and acquisitions Firm characteristics and productivity
  M&As are identified from establishment (workplace) IDs changing firm (CVR) IDs: a firm-year is a merger if its establishments were owned by more than one distinct firm in the previous year (ignoring spurious ownership changes where an establishment's firm ID reverts); the acquirer is the establishment whose firm ID is unchanged, and targets are establishments whose previous-year firm ID differs when the firm also has an acquirer.
  Registers: IDA, IDAS
  Codes: lbnr (establishment), cvrnr (firm); merger = nvals(cvrnr_prevyear)>1 within cvrnr-year; acquirer: cvrnr==cvrnr_prevyear; target: cvrnr!=cvrnr_prevyear & nacquirer>0
  replication code · clean_merger.do
• Firm spinouts and exits Firm characteristics and productivity
  Share of a firm's workers in establishments that move to a different firm ID next year (spinout) or disappear next year before 2011 (exit).
  Registers: IDA
  Codes: spinout = cvrnr_nextyear!=cvrnr; exit = cvrnr_nextyear missing & year<2011
  replication code · clean_merger.do
• Education groups Education and test scores
  Four education groups from the first two digits of the highest completed education code: basic (<=25), vocational (35), short/medium higher (40-50), long higher (>=60).
  Registers: IDA, UDDA
  Codes: max_educ = int(HFFSP/1e6): <=25 basic; 35 vocational; 40-50; >=60
  replication code · AKM_merger_cvrnr.do; clean_merger.do
• Firm industry Sector and industry
  Establishment industry is the modal worker industry code if held by more than 50% of workers; firm industry (127- and 36-group) is assigned only if at least 70% of the firm's workers share it, otherwise missing.
  Registers: IDA
  Codes: persbrc mapped to pb07, ind127, ind36; thresholds 0.5 (establishment) and 0.7 (firm)
  replication code · clean_merger.do
• Hiring and firing rates Firm characteristics and productivity
  Hiring rate = workers whose firm ID differs from the previous year divided by lagged firm employment; firing rate = workers whose firm ID differs next year divided by current employment.
  Registers: IDA
  replication code · AKM_merger_cvrnr.do

Replication package

• Work accident causing loss of earning capacity (treatment) Health outcomes, diagnoses and mortality
  Work accidents recorded by the Labour Market Insurance (AES) that caused a loss of earning capacity, restricted to physical (not psychological) injuries, non-military workers, and workers with full-time employment in each of the three years before the accident. An injury qualifies for compensation if personal impairment exceeds 5% or earning capacity loss exceeds 15%.
  Registers: Industrial Injury Register (Arbejdsskaderegisteret, AES), IDA
  working paper · Sections 2.1, 2.2.1 and 3.1
• Injury severity (earning capacity loss, personal impairment) Health outcomes, diagnoses and mortality
  Earnings capacity loss = loss of earning capacity in percent as assessed by The Labour Market Insurance (AES); personal impairment = degree of personal impairment in percent based on injury diagnosis. Temporary injuries = accidents AES assesses did not cause permanent loss of earning capacity or personal impairment. Non-injured controls are assigned the injury year of their matched injured worker.
  Registers: AES Industrial Injury Register (Arbejdsskaderegisteret)
  online appendix · Supplemental Appendix Table A.1, rows 26-28; temporary-injury figure notes
• Matched non-injured control workers Sample definition and timing
  Injured workers matched one-to-one to non-injured workers in cells of event year, pre-event occupation, age, gender and level of education (baseline specification 5); occupational earnings premium computed within Year-Occupation-Industry-Education-Age-Gender match cells.
  Registers: IDA; AES; education register
  online appendix · Supplemental Appendix Table C.2 and Figure C.1 notes; Table F notes
• Access to higher education (instrument) Education and test scores
  Direct access to higher education through a high school diploma or a vocational degree. Travel time to higher education = shortest car travel time (minutes) from residence zip code to the zip code of the nearest facility offering a relevant higher degree (as in Harmon 2015), measured separately for Craft, Care and Other educational groups; imputed for non-access workers as a residence-zip weighted average.
  Registers: Education register (UDDA); residence zip code
  online appendix · Supplemental Appendix Table A.1, rows 24-25
• Degree and course enrolment/completion Education and test scores
  Degree courses = formal education programmes in the Education Register (UDDA); non-degree courses = Course Participation Register (VEUV); courses classified as Basic (primary/high school), Vocational, Higher (post-secondary). Training rate = share enrolling in a higher degree within six years after the accident; participation flow = enrollment in higher degrees in the year; stock = accumulated enrollment. Reskilling (2SLS treatment) = pursuing a higher degree within ten years after the accident; completers vs drop-outs defined by completion within ten years.
  Registers: UDDA (Education Register); VEUV (Course Participation Register)
  Codes: Within 6 years (training rate); within 10 years (reskilling)
  online appendix · Supplemental Appendix Table A.1, rows 31-38; Section D.1; Figure D notes
• Disability and other social transfers Unemployment and benefits
  Effects on transfers (incl. disability and unemployment benefits) are scaled with transfer rates linked to the transfer codes of the DREAM register; education costs include tuition (~$16,500 per full-time student-year, from government rate catalogs) and SU and reskilling benefits; 6% real discount rate.
  Registers: DREAM register
  online appendix · Supplemental Appendix Section J
• Occupational task content / physical ability Occupation and job hierarchy
  Physical ability requirement = standardized average O*NET importance of Static Strength, Explosive Strength, Dynamic Strength, Trunk Strength and Stamina. Cognitive ability requirement = standardized average importance of Fluency of Ideas, Originality, Problem Sensitivity, Deductive Reasoning, Inductive Reasoning, Information Ordering, Category Flexibility, Mathematical Reasoning and Number Facility. Physical jobs = physical requirement > 0.5 SD above the mean. Occupational injury rate = accepted (not necessarily compensated) work accidents per full-time employee.
  Registers: O*NET mapped to DISCO occupation codes; AES accident records
  Codes: Physical job: > +0.5 SD
  online appendix · Supplemental Appendix Table A.1, rows 21-23; Section D.6.2 figure notes
• Health care utilisation (incl. antidepressant use) Health care use
  Hospitalisations (inpatient and outpatient) with diagnoses from LPR, contacts with primary-care physicians and specialists from the Health Insurance Register, and prescription drug purchases (e.g. antidepressants, painkillers/opioids) from LMDB, 1995-2017.
  Registers: LPR, Sygesikringsregisteret, LMDB
  working paper · Section 2.2.2; Section 3.1
• Hours worked Hours worked
  2008-2017: yearly number of paid hours. 1995-2007: yearly labor income divided by hourly wage rate.
  Registers: Employment/earnings registers (Statistics Denmark)
  Codes: Break in 2008
  online appendix · Supplemental Appendix Table A.1, row 11
• Labor income Earnings and income
  Total labor market income, including bonuses, amenities, wages paid under sick and parental leave, and employer contributions to pension saving schemes.
  Registers: Income registers
  online appendix · Supplemental Appendix Table A.1, row 12
• Hourly wage Earnings and income
  2008-2017: yearly total labor income divided by yearly hours worked. 1995-2007: average hourly wage rate in the November job; reliable only for individuals working more than 20 hours a week in that period.
  Registers: Employment/earnings registers
  Codes: >20 hours/week for 1995-2007
  online appendix · Supplemental Appendix Table A.1, row 13
• Job tenure Firm characteristics and productivity
  Number of consecutive years in which the main job in November is registered with the same firm identifier.
  Registers: IDA employer-employee data
  online appendix · Supplemental Appendix Table A.1, row 14
• Labor market experience Employment and labour force status
  Labor market experience measured in years since 1964.
  Registers: IDA / ATP-based experience
  online appendix · Supplemental Appendix Table A.1, row 15
• Public sector Sector and industry
  Dummy for work within the public sector, measured by 2-digit industry codes.
  Registers: Employment registers (industry codes)
  Codes: 2-digit industry codes (specific codes not listed)
  online appendix · Supplemental Appendix Table A.1, row 16
• Union membership Other
  Dummy for union membership, measured by a positive deductible amount (union fees) reported by unions to the tax authorities.
  Registers: Tax records
  Codes: Deductible union fees > 0
  online appendix · Supplemental Appendix Table A.1, row 17
• Sick leave Unemployment and benefits
  Share of weeks within a year in which the individual (or the employer) received sickness benefit transfers; the sick-leave labour-market state is a dummy for more than 15% of weeks in the year on sickness benefits.
  Registers: DREAM transfer register
  Codes: > 15% of weeks
  online appendix · Supplemental Appendix Table A.1, rows 18 and 43
• Labor market states (Employed, School, Sick Leave, DI, Other) Employment and labour force status
  Employed = full-time work while not in school, on sick leave or receiving DI; School = enrollment in a higher degree in the year; Sick Leave = >15% of weeks on sickness benefits; DI = any weeks in the year receiving disability insurance; Other = residual (mainly part-time work, other unemployment and non-participation).
  Registers: IDA; education register; DREAM
  Codes: Sick leave > 15% of weeks; DI any weeks
  online appendix · Supplemental Appendix Table A.1, rows 41-45
• Debt-to-income and savings-to-income ratios Wealth, debt and consumption
  Debt = debt to banks, pension funds, insurance and financial companies, credit card debt and study loans in banks, divided by labor income. Savings = mainly liquid savings: bank deposits, bond values and value of mortgage deeds, divided by labor income. Property owner = public cash valuation of owned property > 0 DKK.
  Registers: Wealth/tax registers
  Codes: Property valuation > 0 DKK
  online appendix · Supplemental Appendix Table A.1, rows 5, 19-20
• Highest completed education categories Education and test scores
  Years of schooling = prescribed years of study of highest completed degree counting from grade 1. Primary = pre-school, primary, preparatory courses or Danish language courses at language centers; Vocational = VET, qualifying programmes or AMU labour market educations; High School = upper secondary; Post-Secondary = short-cycle higher, vocational bachelor, bachelor, master or PhD.
  Registers: Education register (UDDA)
  online appendix · Supplemental Appendix Table A.1, rows 6-10
• Primary school GPA (age 16) Education and test scores
  Overall GPA = average of all grades given in grade 9 (compulsory) and grade 10 (non-compulsory), normalized to 0-100; Math GPA = same for the subject Math.
  Registers: Grade registers (Statistics Denmark)
  Codes: Normalized 0-100
  online appendix · Supplemental Appendix Table A.1, rows 29-30
• Hospital visits and days in hospital Health care use
  Number of hospital visits (admissions, outpatient treatment and ER visits); days hospitalized, with admission and release dates for admitted patients and outpatient/ER visits coded as 1-day visits.
  Registers: National Patient Register (LPR)
  Codes: Outpatient/ER = 1 day
  online appendix · Supplemental Appendix Table A.1, rows 39-40
• Long-term displacement and manual jobs (mass layoffs) Unemployment and benefits
  Mass layoffs defined as in Davis and von Wachter (2011). A worker is long-term displaced if he enters long-term unemployment (official definition: 52 consecutive weeks) within two years of the mass layoff; manual jobs = Autor-Levy-Murnane manual ability requirement > 0.5 SD above the mean; automation exposure = Webb (2019) robot exposure > 0.5 SD above mean; offshoring exposure = non-offshorability index (average of O*NET decision-making and face-to-face) > 0.5 SD below mean.
  Registers: IDA; unemployment registers; DISCO occupation codes
  Codes: 52 consecutive weeks; +/-0.5 SD cutoffs
  online appendix · Supplemental Appendix Section H.1

• Antidepressant expenditure Health care use
  Annual sum of prescription drug expenditure on antidepressants per person, from the prescription register 1995-2016; set to 0 for person-years with no purchases.
  Registers: LMDB
  Codes: ATC3 = N06A; eksp_anti = sum of eksp where atc3='N06A'
  replication code · Programs/1_data_job_spell/antidepressants2.sas; HMXdata.sas
• Heart and stroke drug expenditure Health care use
  Annual sum of prescription drug expenditure per person on cardiac therapy drugs (heart) and antithrombotic agents (stroke), 1995-2016; 0 if none. Sleep medication defined analogously.
  Registers: LMDB
  Codes: heart: ATC2 = C01; stroke: ATC2 = B01; sleep: ATC3 = N05C; depression: ATC3 = N06A (eksp summed)
  replication code · Programs/1_data_job_spell/stressdrugs2.sas; 3_data_event_study/drugs1.sas
• Stress, depression and burnout hospitalizations Health outcomes, diagnoses and mortality
  Person-year indicators for hospital contacts with an action (primary) diagnosis in the listed ICD-10 groups: severe stress, depression, mild depression, work problems, burnout; broad burnout = burnout + work problems excluding unemployment/job change/job loss; narrow burnout = burnout + stressful work + mental strain.
  Registers: LPR
  Codes: ADIAG: severe stress DF43; depression DF32; mild depression DF320, DF321, DF329; work problems DZ56 (excl. DZ560 unemployment, DZ561 job change, DZ562 job loss for workproblems2); stressful work DZ563; mental strain DZ566; burnout DZ730-DZ733
  replication code · Programs/1_data_job_spell/stressdiagnoses4.sas
• Heart attack, stroke and circulatory hospitalizations Health outcomes, diagnoses and mortality
  Person-year indicators for hospital contacts with action diagnoses for heart attack, hemorrhage, cerebral infarction, hypertension (bloodpressure) and a broad circulatory group.
  Registers: LPR
  Codes: ADIAG: heart attack DI21; hemorrhage DI61; infarction DI63; bloodpressure DI10-DI13, DI15; circulatory DI20-DI25, DI44-DI50, DI512, DI513, DR001
  replication code · Programs/1_data_job_spell/stressdiagnoses4.sas
• Alcohol, digestive, accident and suicide hospitalizations Health outcomes, diagnoses and mortality
  Person-year indicators for alcohol-related and stress-related digestive diagnoses following Browning and Heinesen (2006), plus accidents and suicide attempts identified from the contact-cause field.
  Registers: LPR
  Codes: alcohol: DF100, DF102, DF104, DF105, DI85, DK70, DK860, DT500, DT510; digestive: DK20, DK222, DK223, DK25-DK31, DK902; accident: KONTAARS=2; suicide: KONTAARS=4; self-harm DX7*, DX8*; poisoning DT36-DT39, DT4*, DT50
  replication code · Programs/1_data_job_spell/stressdiagnoses4.sas
• Annual hours, overtime and sick hours (job-spell analysis) Hours worked
  Hours paid, overtime hours and absence hours from the wage statistics register 1997-2012, merged to the worker-firm link by person, firm and year; for 2011-2012 earnings components are per-hour rates multiplied by hours.
  Registers: LON, FIDA
  Codes: 1997-2010 lon: timprae (total hours), timover (overtime), timfra (absence), sftj (gross earnings), smftj (net), bfra (sick pay); 2011-2012 lonn: timprae, timover, timfra, fortj_prae*timprae etc.
  replication code · Programs/1_data_job_spell/hoursdata2.sas
• Monthly hours and earnings (event study) Hours worked
  Monthly paid hours and narrow/broad wage amounts per person and firm summed from monthly employer payroll reports.
  Registers: BFL (eIndkomst)
  Codes: ajo_indberettede_loentimer, ajo_loentimer, ajo_loentime_kode, ajo_smalt_loenbeloeb, ajo_bredt_loenbeloeb
  replication code · Programs/3_data_event_study/BFLhours1.sas
• Bartik sales instrument (HS4) Firm characteristics and productivity
  For each firm, HS4 product shares of Prodcom sales in a pre-sample year (first year the firm is in both Prodcom and the estimation sample, or its designated pre-sample year), multiplied by aggregate Danish Prodcom sales of each HS4 product in later years and summed to the firm-year; used in logs after deflating by the GDP deflator.
  Registers: VARS (Prodcom)
  Codes: bartik = sum_hs4 share_presample * aggvrd(hs4, year), years after presample year only
  replication code · Programs/1_data_job_spell/Part5sales_hs4.sas
• Job-spell estimation sample Sample definition and timing
  Worker-firm-years in firms with positive imports and exports, workers aged 20-60, full-time employees, non-missing output, capital-labour ratio, skill share and the import-price instrument; firms kept from the first year with at least 10 employees onward; singleton job spells (person-firm) dropped.
  Registers: IDA, FIDA, FIRM, UHDI
  Codes: heltid==1 (full-time, HDTID2 codes 1-4 per code comment); age 20-60; empl>=10
  replication code · Programs/1_data_job_spell/HMXdata2.do
• Occupation codes and management layers Occupation and job hierarchy
  4-digit DISCO/ISCO-88 codes harmonised across years, missing codes filled within job spells (constant code, then forward/backward fill); hierarchical layers counted per firm-year from ISCO88 class (2-5), set missing if more than 25% of the firm-year's workers lack a class. CEO/manager dummy = ISCO major group 1.
  Registers: IDA
  Codes: ceo = disco 1000-1999; disco set missing if 0, 99, 110, 999, 9999; layers = sum of class 2-5 present
  replication code · Programs/1_data_job_spell/HMXdata2.do

Replication package

• ATP-based hours (1980-2018) Hours worked
  Hours inferred from mandatory ATP pension contributions, which employers pay based on a binned measure of hours; for monthly-paid workers annual contributions depend on monthly hours in four bins, giving 37 annual hours bins. Capped at full time in all 12 months.
  Registers: ATP pension contributions
  working paper · Section 3, footnotes 17-18
• Monthly uncapped hours (from 2008) Hours worked
  Uncapped hours worked for all workers at monthly level, available from 2008; used where precise hours variation is needed (aggregated to quarters in the promotion event study).
  Registers: Monthly payroll records
  working paper · Section 3; Section 4 Fact 3
• Job (firm x occupation cell) and job switch Occupation and job hierarchy
  Job is a firm x occupation cell; a switcher moves between firm x occupation cells after the reform. Occupations use DISCO-08 from 2010 (563 occupations), DISCO-88 1991-2009 (372) bridged with the Humlum (2021) crosswalk, and an older Danish classification before 1991 (299). Switches where more than 50% of an individual's coworkers move to the same job cell in the same year are dropped.
  Registers: Employer-employee matched data
  Codes: DISCO-08; DISCO-88; pre-1991 Danish occupation codes
  working paper · Section 3, footnote 15; Section 5.1
• Mass layoff switchers Firm characteristics and productivity
  Mass layoffs are layoffs in which firms with at least 20 employees reduce their workforce by at least 30% in a single year; switchers are workers who move firm following such a layoff. Used in an event study of log earnings vs a linear pre-layoff trend.
  Registers: Employer-employee matched data
  Codes: firm size >= 20 employees; workforce reduction >= 30% within one year
  online appendix · Online Appendix, Figure A.8 notes
• Labour earnings outcome Earnings and income
  Total labour compensation, including contributions to employer-administered retirement savings plans; outcome is the log change between 2008 and 2010 (placebo 2006-2008).
  Registers: Income tax records
  working paper · Section 5.1, footnote 26
• Reform treatment and simulated net-of-tax rate instrument Taxes and tax rates
  Treatment assigned by pre-reform baseline taxable income (labour income, transfers, pensions, alimony and certain capital income) in bins relative to the 2009 reform threshold (around the 70th percentile). The instrument is the simulated log change in the marginal net-of-tax rate from tax code changes, holding income at baseline (Gruber-Saez). Baseline sample: taxpayers with baseline income 250,000-1,000,000 DKK.
  Registers: Income tax records
  working paper · Section 5.1, footnote 27; Table 2 notes
• Promotion Occupation and job hierarchy
  Within-firm switch to an occupation cell whose median earnings are at least 10% higher (robustness 5% and 20%); promoted workers matched to unpromoted co-workers in the same firm.
  Registers: Employer-employee matched data, Monthly payroll records
  working paper · Section 4, Fact 3
• Occupational rank / top vs bottom occupations Occupation and job hierarchy
  Occupation cells (6-digit occupation codes) ranked by mean wage earnings; for top/bottom occupations the ranking is computed among workers aged 45-50 (mean, P10 and P90 of wage earnings reported per cell). Occupational rank used in balance checks orders occupation cells by their mean earnings.
  Registers: Employer-employee matched data (DISCO occupation codes)
  Codes: 6-digit DISCO occupation codes
  online appendix · Online Appendix, Table A.1 and Figure A.6 notes
• AKM firm wage premium Firm characteristics and productivity
  Firm fixed effects from an AKM model log z_it = alpha_i + psi_J(i,t) + X_it beta + nu_it, with controls for year, age and tenure-in-current-firm dummies, estimated on pre-reform data 2002-2005 for firms with at least 10 employees; estimated firm effects merged onto the 2009 tax-reform sample to measure changes in firm premia for firm switchers.
  Registers: Employer-employee matched data, Income tax records
  Codes: firms >= 10 employees; estimation window 2002-2005
  online appendix · Online Appendix, Section C and Figure A.9
• Survey hours (LFS validation) Hours worked
  Validation measure of hours: actual, uncapped hours worked from the Danish component of the EU Labour Force Survey, compared with the administrative ATP pension-based hours measure; the survey measure is much noisier, especially at the top.
  Registers: Danish Labour Force Survey (AKU), ATP pension contributions
  online appendix · Online Appendix, Figure A.1 notes
• Number of job switches (ages 20-50) Occupation and job hierarchy
  Count of job (firm x occupation cell) switches between ages 20 and 50 in a balanced panel of workers observed at every age 20-50, split by earnings percentile at age 50 (top 10%, top 1%, bottom 50%).
  Registers: Employer-employee matched data
  online appendix · Online Appendix, Figure A.3 notes
• Demographic controls (job-switcher regressions) Sample definition and timing
  Robustness controls: dummies for age, gender, marital status and number of children, alone or combined with initial job-cell fixed effects (baseline uses job-cell fixed effects only).
  Registers: Population register (BEF)
  online appendix · Online Appendix, Figure A.7 notes

• Annual employment in full-time equivalents Employment and labour force status
  Employment in annual full-time equivalents (FTE = 1 for a full year of work), from a register variable recording cumulative employment in a calendar year, observed for everyone in Denmark on 1 January. For YSR = 1 only, FTE can exceed 1 because the cumulative variable includes work in the year residency was granted.
  Registers: Income registers, CPR
  paper · Section 5 'Data on Refugee Immigrants' and footnote 17
• Danish language proficiency Education and test scores
  Indicator for having passed the standard Danish language test at the intermediate or higher level (test levels: basic, intermediate, high); observed from YSR = 1; passing is treated as absorbing, so the person counts as proficient at that level in all later periods.
  Registers: Danish language test results
  paper · Section 4, footnote 13; Section 5
• Treatment and control cohorts Sample definition and timing
  Refugees who applied for asylum after 28 February 2002 and were aged 18-64 when granted temporary residency; treatment = temporary residency granted January-October 2004 (exposed to the 1 May 2007 permanent-residency reform), control = January-October 2003.
  Registers: Immigration registers, BEF
  paper · Sections 2, 4 and 5
• Years since residency (YSR) Sample definition and timing
  For persons granted temporary residency during year y, YSR = 1 is year y+1 (first full calendar year with employment observed as of 1 January), followed for seven years (YSR 1-7).
  Registers: Immigration registers
  paper · Section 4 and Section 5
• Pre-reform employment performance type Employment and labour force status
  HIGH if cumulative employment by YSR = 2 is above 0.75 FTE, otherwise LOW (20% HIGH, 80% LOW). The reform's requirement is 2.5 FTE of cumulative employment.
  Registers: Income registers
  paper · Section 5, 'Classifying workers into performance types'
• Pre-reform language performance type Education and test scores
  HIGH if the person had passed the Danish language test at the intermediate level or higher by YSR = 2 (about 16%), otherwise LOW; combined with employment type into LOW/LOW, LOW/HIGH, etc.
  Registers: Danish language test results
  paper · Section 5, 'Classifying workers into performance types'
• Matched native comparison group Sample definition and timing
  Danish natives selected by propensity-score matching to treated (control) refugees on YSR = 1 employment (2005 for treated, 2004 for control), age, gender, being single, having any children and county of residence (initial assignment county for refugees), used in a triple-difference check for the 2009 recession.
  Registers: BEF, Income registers
  paper · Section 6.3 'Comparison with natives'

• Treatment assignment and study population Sample definition and timing
  Work units (314 units of eldercare workers across 110 employers in 11 municipalities) were randomised in September 2008, with 144 units (46%) assigned to Healthy-at-Work. The population is individuals employed at participating work units in the month just before randomisation (7,541 of 7,660 linked to registers). Treatment status is fixed at the initial randomisation month, so post-randomisation moves are ignored and estimates are intention-to-treat.
  Registers: Healthy-at-Work experimental records, Statistics Denmark registers (CPR-linked)
  working paper · Sections 2.1, 3 and 4 (treatment assignment)
• High-risk group Health care use
  Individuals with an above-median number of primary care physician consultations in the six months before randomisation; the low-risk group is at or below the median.
  Registers: Primary care physician consultation records (Statistics Denmark)
  working paper · Notes to Tables 2-5
• Primary care physician consultations Health care use
  Number of consultations (and any consultation) with the primary care physician, from register data on fee-for-service payments to primary care physicians; follow-up to 2020. The paper does not give the register name or speciality codes.
  Registers: Primary care physician fee-for-service records (Statistics Denmark)
  working paper · Section 3; Section 4 results; cost-benefit section
• Prescription drug purchases Health care use
  Any prescription drug purchase and months' supply per year (through 2019), covering drugs for hypertension, cholesterol, diabetes, severe pain relief and depression.
  Registers: Prescription drug purchase records (Statistics Denmark)
  Codes: ATC C9 (hypertension, as printed), C10, A10, N2, N05A, N05B, N05C, N06A
  working paper · Table 3 notes
• Hospitalisations (psychiatric / somatic) Health care use
  Any hospital contact and number of days admitted to hospital, separately for contacts with a psychiatric diagnosis and with a somatic diagnosis; follow-up to 2018. Diagnosis code lists are not given.
  Registers: Hospital contact records (Statistics Denmark)
  working paper · Table 4 panels A-D
• Sickness absence Employment and labour force status
  Any absence and yearly absence rate (share of scheduled hours missed), conditional on hours > 0. Absence data cover about 95% of the sample.
  Registers: Absence register (Statistics Denmark; name not given)
  working paper · Table 5 panels A-B; Table 1 notes
• Hours, employer continuity and earnings Employment and labour force status
  Any hours worked, hours worked per year, continued employment with the same employer (turnover measure), and earnings per year.
  Registers: Employment registers (Statistics Denmark; names not given)
  working paper · Table 5 panels C-F
• Spouses (spillover sample) Family links, fertility and household
  Individuals married to or cohabiting with individuals employed at participating work units in the month just prior to randomisation.
  Registers: Population register (Statistics Denmark)
  working paper · Table 6 notes (spillovers to spouses)

Replication package

• Radiotherapy receipt Health care use
  Indicator for receiving adjuvant radiation therapy, from the DBCG clinical Breast Cancer Database, which records treatments administered and dates of diagnosis and major interventions.
  Registers: DBCG clinical Breast Cancer Database
  working paper · Section 3 (Data), Treatment Variable
• T95 group and instrument Sample definition and timing
  Instrument is the interaction between an indicator for belonging to the T95 group and an indicator for diagnosis in or after 1995. The comparison group consists of 'always eligible' and 'never eligible' women (eligible or not eligible for radiotherapy regardless of year of diagnosis); full analysis sample N = 2,823. Robustness drops women diagnosed in 1994, women residing in areas with pilot universal screening programs, or women with breast-conserving surgery.
  Registers: DBCG clinical database
  online appendix · Notes to Appendix Tables A4, A9 and Figures A4-A5
• Analysis sample Sample definition and timing
  Female breast cancer patients diagnosed 1990-1998 (26,900), restricted to DBCG89 trial enrollees who are pre-menopausal and high-risk, excluding always-eligible patients whose radiotherapy intensity changed, patients missing radiotherapy, tumour size or micro-radical removal information, and women aged 55+ at diagnosis (final N = 2,823).
  Registers: DBCG clinical Breast Cancer Database
  working paper · Section 3 (Analysis Sample); Appendix Table A1
• All-cause and breast cancer mortality Health outcomes, diagnoses and mortality
  Indicators for death from any cause and from breast cancer, using the exact date and ICD-coded cause of death in the Register of Causes of Death, measured for each year from the date of diagnosis up to 10 years after.
  Registers: Register of Causes of Death
  working paper · Section 3, Outcome Variables
• Employed / unemployed / out of labour force Employment and labour force status
  Indicators of labour market status as of November from the Register-Based Labour Force Statistics, for each calendar year from diagnosis up to 10 years later. Non-survivors are assigned one for out of the labour force and zero for all other labour market outcomes.
  Registers: RAS
  working paper · Section 3, Outcome Variables
• Labour earnings and gross income Earnings and income
  Monetary outcomes are expressed in thousands of 2015 DKK; labour-market outcomes are averaged over 1-5 and 6-10 years after diagnosis; pre-diagnosis characteristics (employment status, labour earnings, gross personal income) are averaged over 2-4 years before diagnosis.
  Registers: Income Statistics Register
  online appendix · Notes to Appendix Table A1 and Figures A4-A5
• Sickness and disability benefit receipt Unemployment and benefits
  Transfer outcomes (any government transfer/any welfare, sickness benefits, disability pension) are indicators measured cumulatively within 5 (or 10) years after diagnosis, not as annual averages.
  Registers: DREAM
  online appendix · Appendix Table A3; Figure A4-A5 notes
• Clinical eligibility controls Health outcomes, diagnoses and mortality
  All possible interactions among indicators for: mastectomy, age <45 at diagnosis, number of positive nodes (0, 1-3, 4+), tumor larger than 50mm, and tumor removed micro-radically; plus separate indicators for age at diagnosis in full years, tumor size in cm, number of positive nodes (0, 1-3, 4+), micro-radical removal, type of surgery (lumpectomy, mastectomy, or lumpectomy followed by mastectomy), and chemotherapy type (DBCG89 trial arm: CMF, CEF, OA, tamoxifen regimens, pamidronate arms); and year-of-diagnosis fixed effects. Standard errors clustered by hospital.
  Registers: DBCG clinical database
  Codes: nodes 0/1-3/4+; tumor >50mm; age <45
  online appendix · Notes to Appendix Tables A2-A4, A9
• Radiotherapy eligibility under DBCG89 protocol Health care use
  DBCG89 protocol (age eligibility <75 years). Low risk (as printed in the appendix protocol table): tumor <=5cm and positive nodes and grade I (if premenopausal); high risk: tumor >5cm or positive nodes or (premenopausal and ductal grade II-III). Radiotherapy indicated after lumpectomy, when the carcinoma was not removed microradically, and additionally for: age <=45 with >=4 positive nodes (until January 1995); premenopausal with positive nodes or tumor >5cm (from January 1995). Positive nodes = cancer spread to at least one lymph node.
  Registers: Danish Breast Cancer Cooperative Group (DBCG) clinical database
  Codes: tumor >5cm; >=4 positive nodes; age <=45; age <75
  online appendix · Appendix Figure A1 (DBCG89 protocol)
• Cumulative hospital contacts Health care use
  Cumulative number of inpatient and outpatient hospital contacts from diagnosis until each year after diagnosis (separate outcomes).
  online appendix · Appendix Figure A8 notes

• Reform-induced GPA (treatment) Education and test scores
  Deviation of a student's recoded (post-1 August 2007, 7-point scale) GPA from a cubic fit of post-recoding GPA on pre-recoding (13-scale) GPA; grades were recoded per the Ministry of Education mapping. Final GPA is a weighted average of recoded grades and grades from exams after the recoding.
  Registers: University student records (Aarhus University, University of Copenhagen)
  Codes: 13-scale to 7-point scale mapping (paper Table 1)
  working paper · Sections 2 and 3.4; Table 1; Figure 3
• Master's student sample Sample definition and timing
  Students enrolled in a master's program on 1 August 2007 at Aarhus University (incl. Aarhus School of Business) or University of Copenhagen with at most 40 ECTS points remaining on the recoding date, who graduated before 2011 (3,813 students).
  Registers: University student records
  working paper · Section 3.1
• Log total gross earnings after graduation Earnings and income
  Log total gross earnings in each of the first five calendar years after graduation; individuals with zero earnings excluded. A separate extensive-margin outcome is an indicator for positive earnings. Monetary values in 2015 prices (CPI).
  Registers: Statistics Denmark registers
  working paper · Section 3.2; Table 2 notes
• Other labour market outcomes Earnings and income
  Log disposable income (after transfers and taxes), unemployment, public-sector employment, job changes, annual earnings growth, and earnings growth with the same employer, in each of the first five years after graduation.
  Registers: Statistics Denmark registers
  working paper · Section 5; Table 7
• Parental background controls Family links, fertility and household
  Parental income (mean total disposable income across observed parents, calendar year before graduation/reform), parental unemployment (mean annual unemployment), indicator for at least one parent with a university degree, and parental origin (non-Western vs Western per Statistics Denmark); missing values set to zero with missing dummies.
  Registers: Statistics Denmark registers
  working paper · Sections 3.2 and 4; Table 2 notes
• Earnings while studying (heterogeneity) Earnings and income
  Low earnings while studying used as a proxy for little relevant labour-market experience; Aarhus University attendance used as a further proxy.
  Registers: Statistics Denmark registers
  working paper · Section 6; Table 8 panels E-F

• Reform treatment (birthdate cutoff) Retirement and pensions
  Those born on or after 1 January 1954 learned in 2011 that their early retirement age (ERA), incentivized retirement age and full retirement age (FRA) rose from 60, 62 and 65 to 60.5, 62.5 and 65.5; those born one day earlier were unaffected. Running variable is birth date relative to the cutoff; RD sample is those born within 56 days (8 weeks).
  Registers: Population register (exact date of birth)
  working paper · Sections 2.3 and 4.2
• VERP participant analysis sample Sample definition and timing
  Four cumulative restrictions: (1) Danes born within six months of the cutoff date January 1, 1954 (71,095 individuals); (2) regular participatory VERP contributors, defined as making contributions in 70% of pre-announcement years 2001-2010 (47,047); (3) balanced panel: present in the data in every year 2006-2018 (43,348); (4) drop the self-employed, defined over the pre-announcement period using Statistics Denmark's definition of an individual's main source of income (final analysis sample 40,042).
  Registers: Population register, VERP contribution records, Statistics Denmark main source of income classification
  Codes: birthdate within +/-6 months of 1954-01-01; VERP contributions in >=70% of years 2001-2010; balanced 2006-2018
  online appendix · Online Appendix Table A.1 and notes
• Contributions to employer-sponsored and personal retirement accounts Retirement and pensions
  Annual contribution flows to employer-sponsored DC accounts in levels (winsorized at the 95th percentile by year) and indicators for positive contributions to personal retirement accounts (also indicators for contributions of various sizes).
  Registers: Labour-market/income registers (tax-reported pension contributions)
  working paper · Section 4.1
• Changes in bank, stock and property values Wealth, debt and consumption
  Year-t minus year-(t-1) year-end values of bank account balances and stock market account balances (reported by financial institutions to tax authorities) and of property cash values as assessed by tax authorities; winsorized at the 5th and 95th percentiles each year (also 10th/90th).
  Registers: Tax-reported asset registers
  working paper · Section 4.1, footnote 20
• Pre-tax earnings Earnings and income
  Income on which individuals pay the 8 percent labor market tax, winsorized by year at the 95th percentile; 2010 prices, 5.56 DKK per USD.
  Registers: Income register
  working paper · Section 4.1
• Monthly retirement age Retirement and pensions
  Absorbing-state measure from the monthly eIncome register (2008-2017): age in the last month with positive earnings before earnings permanently fall to zero.
  Registers: eIncome (eIndkomst) monthly earnings register
  working paper · Section 4.1, footnote 17
• VERP benefit claiming Retirement and pensions
  Annual Voluntary Early Retirement Pension (VERP) benefit payments.
  Registers: VERP benefit records
  working paper · Section 4.1
• Firm default contribution rate Firm characteristics and productivity
  All individuals over 18 are assigned to workplaces in 2010; individual contribution rate = employer-sponsored plan contributions divided by earnings; the firm default rate is the median contribution rate among the firm's workers in 2010. Predicted contributions = current earnings times the 2010 firm default rate.
  Registers: Linked employer-employee data, Pension contribution records
  working paper · Section 6 (employer defaults), footnote 30
• VERP participation indicator (qualified UI-fund contributions) Retirement and pensions
  Participatory VERP contribution in a year is measured as an indicator for making qualified (VERP) contributions to unemployment insurance (UI) funds in that year; VERP eligibility requires contributions in 25 of the previous 30 years.
  Registers: VERP contribution records
  online appendix · Online Appendix Table A.7 notes; Appendix C.2
• RD control variables Other
  Baseline RD controls: indicator for being male, indicator for being married in 2010, and indicators for region of residence in 2010 (five regions: Hovedstaden, Sjaelland, Syddanmark, Midtjylland, Nordjylland). Regressions use separate linear polynomials in birthdate on each side of the cutoff with triangular weights.
  Registers: Population register
  Codes: 5 regions as of 2010
  online appendix · Online Appendix Table A.2 notes
• Retirement at eligibility-age spikes Retirement and pensions
  Control group (born before cutoff): first spike = retirement age between 60y0m and 60y5m; second spike = 62y0m to 62y5m. Treatment group: first spike = 60y6m to 60y11m; second spike = 62y6m to 62y11m. 15.8% retire at the first spike and 10.6% at the second as defined.
  Codes: monthly retirement-age windows 60:00-60:05, 62:00-62:05 (control); 60:06-60:11, 62:06-62:11 (treated)
  online appendix · Online Appendix B
• Pre-reform personal retirement plan users Sample definition and timing
  Subsample of individuals who had used personal retirement plans before the reform announcement: those who made contributions to personal plans in two or three of the years 2008-2010.
  Registers: Tax-reported pension contribution records
  Codes: contributions in >=2 of 2008, 2009, 2010
  online appendix · Online Appendix table notes (personal-plan subsample)
• Outcome winsorization and time windows Sample definition and timing
  Anticipation-period outcomes are averaged over 2011-2013 and pre-announcement placebo outcomes over 2008-2010; robustness uses stricter winsorization at the 10th and 90th percentiles; an additional robustness subsample excludes individuals with real estate transactions.
  online appendix · Online Appendix Tables A.5-A.6, placebo table notes