Health insurance as economic stimulus? Evidence from long-term care jobs

Hackmann, Martin B.; Heining, Jörg; Klimke, Roman; Polyakova, Maria; Seibert, Holger Working Paper Health insurance as economic stimulus? Evidence from long-term care jobs IAB-Discussion Paper, No. 03/2025 Provided in Cooperation with: Institute for Employment Research (IAB) Suggested Citation: Hackmann, Martin B.; Heining, Jörg; Klimke, Roman; Polyakova, Maria; Seibert, Holger (2025) : Health insurance as economic stimulus? Evidence from long-term care jobs, IAB- Discussion Paper, No. 03/2025, Institut für Arbeitsmarkt- und Berufsforschung (IAB), Nürnberg, https://doi.org/10.48720/IAB.DP.2503 This Version is available at: https://hdl.handle.net/10419/315970 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-sa/4.0/deed.de ISSN 2195-2663 IAB-DISCUSSION PAPER Articles on labour market issues 03|2025 Health Insurance as Economic Stimulus? Evidence from Long-Term Care Jobs Martin Hackmann, Jörg Heining, Roman Klimke, Maria Polyakova, Holger Seibert Health Insurance as Economic Stimulus? Evidence from Long-Term Care Jobs Martin Hackmann (University of California, Los Angeles) Jörg Heining (Institut für Arbeitsmarkt- und Berufsforschung (IAB)) Roman Klimke (Harvard University) Maria Polyakova (Stanford University) Holger Seibert (Institut für Arbeitsmarkt- und Berufsforschung (IAB)) Mit der Reihe „IAB-Discussion Paper“ will das Forschungsinstitut der Bundesagentur für Arbeit den Dialog mit der externen Wissenschaft intensivieren. Durch die rasche Verbreitung von Forschungsergebnissen über das Internet soll noch vor Drucklegung Kritik angeregt und Qualität gesichert werden. The “IAB-Discussion Paper” is published by the research institute of the German Federal Employment Agency in order to intensify the dialogue with the scientific community. The prompt publication of the latest research results via the internet intends to stimulate criticism and to ensure research quality at an early stage before printing. Contents 1. Introduction............................................................................... 7 2. Institutional Setting and Data......................................................... 13 2.1. Long-Term Care Insurance in Germany .............................................. 13 2.2. Data Sources and Sample Construction ............................................. 14 2.2.1. Sampling Frame ............................................................... 14 2.2.2. Characteristics of Workers .................................................... 15 2.2.3. Mortality ........................................................................ 16 2.2.4. Means-Tested and Universal LTC Beneficiary Counts ..................... 16 2.3. Summary Statistics ..................................................................... 17 3. Research Design ........................................................................ 18 3.1. Source of Variation ..................................................................... 18 3.1.1. Partial Equilibrium ............................................................. 20 3.2. General Equilibrium .................................................................... 21 3.3. Mortality ................................................................................. 21 4. Results .................................................................................. 22 4.1. Partial Equilibrium...................................................................... 22 4.1.1. Industry Expansion ............................................................ 22 4.1.2. Anatomy of Expansion ........................................................ 24 4.2. General Equilibrium .................................................................... 25 4.3. Mortality ................................................................................. 27 5. Welfare .................................................................................. 28 5.1. Approximation Approach .............................................................. 29 5.2. Model-based Approach................................................................. 32 5.2.1. Theory ........................................................................... 33 5.2.2. Taking Theory to Data ........................................................ 35 5.2.3. Welfare and Counterfactuals ................................................. 38 6. Conclusion .............................................................................. 40 References...................................................................................... 42 Appendix .................................................................................... 64 A1. Data Appendix .......................................................................... 65 A1.1. Processing of the Raw Data .................................................. 65 A1.2. Variable Definitions ............................................................ 66 A1.3. Appendix Figures and Tables ................................................ 68 A2. Model Details ............................................................................ 90 A2.1. Equilibrium Wages and Queues.............................................. 90 IAB-Discussion Paper 03|2025 3 A2.2. Estimation and Model Fit ..................................................... 92 A2.3. Parameter Estimates .......................................................... 96 A2.4. Counterfactuals ...............................................................100 IAB-Discussion Paper 03|2025 4 Abstract We leverage decades of administrative data and quasi-experimental variation in the introduction of universal long-term care (LTC) insurance in Germany in 1995 to examine whether health insurance expansions can stimulate local economies. We find that the LTC insurance rollout led not only to sizeable growth of the target LTC sector, but also to an aggregate fall in unemployment and an increase in the labor force participation. Quantitatively, a 10 percentage point increase in the share of insured LTC patients led to 4 more nursing home workers per 1,000 individuals age 65 and older (12 percent increase). Wages did not rise in the LTC sector or other sectors of the economy. The quality of newly hired nursing home workers declined, but this had no negative effect on old-age life expectancy. Overall, the insurance expansion brought lower-skilled workers into new jobs rather than reallocating workers away from other productive sectors. Our marginal value of public funds (MVPF) analysis suggests that the reform paid for itself when taking the positive fiscal externalities in the labor market into account. To understand which market primitives underpin our findings and to inform the external validity of our results, we develop and estimate a general model of labor markets with product-market subsidies in the presence of wedges, such as income taxes. Our model simulations show that the aggregate welfare effects of insurance expansions are theoretically ambiguous and depend centrally on the magnitude of frictions in input markets. Zusammenfassung Wir nutzen administrative Daten und quasi-experimentelle Variationen bei der Einführung der allgemeinen Langzeitpflegeversicherung (LTC) in Deutschland im Jahr 1995, um zu untersuchen, ob die Ausweitung der Sozialversicherung sich positiv auf die lokale Wirtschaft auswirken kann. Wir stellen fest, dass die Einführung der Langzeitpflegeversicherung nicht nur zu einem beträchtlichen Wachstum des Zielsektors der Langzeitpflege führte, sondern auch zu einem Gesamtrückgang der Arbeitslosigkeit und einem Anstieg der Erwerbsbeteiligung. Quantitativ führte ein Anstieg des Anteils der versicherten Langzeitpflegepatienten um 10 Prozentpunkte zu 4 mehr Pflegeheimmitarbeitern pro 1.000 Personen im Alter von 65 Jahren und älter (12 Prozent Anstieg). Die Löhne stiegen weder im Pflegesektor noch in anderen Wirtschaftszweigen. Die Qualität der neu eingestellten Pflegeheimkräfte nahm ab, was sich jedoch nicht negativ auf die Lebenserwartung im Alter auswirkte. Insgesamt führte die Versicherungsausweitung dazu, dass weniger qualifizierte Arbeitskräfte neue Arbeitsplätze bekamen, anstatt Arbeitskräfte aus anderen produktiven Sektoren zu verlagern. Unsere Analyse des Grenzertrags öffentlicher Mittel (MVPF) legt nahe, dass sich die Reform unter Berücksichtigung der positiven externen fiskalischen IAB-Discussion Paper 03|2025 5 Effekte auf dem Arbeitsmarkt bezahlt gemacht hat. Um zu verstehen, welche Marktkomponenten unseren Erkenntnissen zugrunde liegen, und um die externe Validität unserer Ergebnisse zu ermitteln, entwickeln und schätzen wir ein allgemeines Modell von Arbeitsmärkten mit Produktmarktsubventionen bei Vorhandensein von ökonomischen Zusatzlasten, wie z. B. Einkommenssteuern. Unsere Modellsimulationen zeigen, dass die aggregierten Wohlfahrtseffekte von Versicherungsausweitungen theoretisch mehrdeutig sind und zentral vom Ausmaß der Friktionen in den Inputmärkten abhängen. JEL D58, H0, H51, I0, I13, I31, I38, J08, J14, J23, J64 Keywords General Equilibrium, Fiscal Externalities, Health Insurance Expansion, Moral Hazard, Marginal Value of Public Funds (MVPF) Acknowledgements We are grateful to Naoki Aizawa, Sydnee Caldwell, Amitabh Chandra, Raj Chetty, Rebecca Diamond, Amy Finkelstein, Matthew Gentzkow, Matthias Gnewuch, Joshua Gottlieb, Daniel Haanwinckel, Lawrence Katz, Timothy Layton, Attila Lindner, Adriana Lleras-Muney, Nicole Maestas, Matthew Notowidigdo, Andreas Peichl, Yotam Shem-Tov, Mark Shepard, Isaac Sorkin, and many seminar participants for helpful comments, as well as Marissa Hindelang and Jacob Morris for outstanding research assistance. Polyakova gratefully acknowledges grant support from the W.E. Upjohn Institute for Employment Research and from the National Institute on Aging (K01AG05984301). IAB-Discussion Paper 03|2025 6 1 Introduction Arrow (1963) and Feldstein (1971, 1977) have argued that the proliferation of health insurance significantly contributes to the growth of the healthcare sector. The textbook view of insurance is that it mutes the price signal inducing inefficiently high demand for healthcare—a phenomenon commonly referred to as moral hazard. Optimal insurance, in turn, trades off the welfare losses from moral hazard vis-a-vis the welfare benefits of risk protection. In this paper, we investigate whether, in practice, publicly funded health insurance programs are also a form of an industrial policy that channels consumers, workers, and capital into a subsidized industry. In the United States alone, more than $2 trillion in government expenditures flow to healthcare through insurance programs each year, underscoring the importance of understanding this spending’s broader economic effects. Yet the aggregate impacts of health insurance, as envisioned by Arrow and Feldstein, have been difficult to capture empirically.1 Two main challenges arise when trying to measure the broad economic consequences of health insurance expansions. First, many changes in health insurance programs are incremental and are likely to produce only small and diffuse effects on the healthcare sector (Finkelstein, 2007). Second, data limitations often hamper our ability to precisely capture how healthcare workers, firms, and capital reallocate in response to insurance expansions. In this paper, we overcome some of these challenges by exploiting a major social insurance reform—the introduction of universal long-term care (LTC) insurance in Germany in 1995—and leveraging uniquely comprehensive administrative labor market data. Governments commonly justify health insurance policy in part by its effects on healthcare employment (Ku/Brantley, 2021). However, Baicker/Chandra (2012) and Cutler (2018) 1 There are a handful of exceptions. Finkelstein (2007) has documented a significant expansion of the U.S. hospital sector following the introduction of Medicare in 1965. Her estimates suggest that the increase in spending was more than six times larger than what the estimates from the RAND Health Insurance Experiment would have predicted. This was likely attributable to the high fixed costs of investments in new technologies or capacity, as well as spillover effects. Kondo/Shigeoka (2013) study the 1961 expansion of universal insurance in Japan, finding increase in utilization, but no evidence of an increase in the number of physicians or nurses. Dillender (2022) finds that healthcare employers post more job vacancies and hire additional healthcare workers in response to Medicaid expansions in the U.S. Geddes/Schnell (2023) demonstrate that the impact of insurance expansions on retail clinic entry in the U.S. hinges on price regulations in output markets. Their analysis reveals that Medicaid expansion under the Affordable Care Act discourages entry (and the expansion of care) when regulated reimbursement rates fall below the market-clearing price (consistent with evidence from Grabowski/Gruber (2007) in the case of U.S. LTC). Conversely, and in alignment with our findings, they show that when insurance subsidizes prices set by private firms, insurance expansions stimulate increased retail clinic entry. Gottlieb et al. (forthcoming) find evidence of changes in the income and labor supply of physicians in response to the relatively large insurance expansion in the U.S. following the Affordable Care Act. IAB-Discussion Paper 03|2025 7 caution that viewing healthcare jobs as drivers of economic growth may be misguided if such jobs do not improve patient health or if they simply crowd out employment in more productive sectors. Thus, an essential question—for both fiscal accounting and welfare analysis—is whether insurance expansions create new jobs on net or merely reallocate existing workers from productive employment elsewhere in the economy (Dustmann et al., 2021). To inform this debate, we examine both the partial equilibrium effects of LTC insurance on the LTC market itself, as well as the general equilibrium effects on employment in the economy as a whole. We find that the German LTC insurance program created jobs and improved welfare. Yet we also show that this stimulus effect is attributable to many frictions that characterized the German labor market in the mid-1990s and may not generalize to other contexts. Our paper makes three main contributions. First, we add extensive new evidence to a small but growing literature that has been able to trace out the effects of health insurance expansions on healthcare sector employment (Finkelstein, 2007; Kondo/Shigeoka, 2013; Dillender, 2022). We do so in an important and unique setting, the rollout of universal long-term care insurance in a large economy. Second, to the best of our knowledge, we provide some of the first evidence of factor substitution between sectors of the economy in response to a health insurance expansion. Third, we show that the nature of factor substitution and market frictions is qualitatively and quantitatively important for the welfare implications of public spending on health insurance. Our positive analysis proceeds in two steps. We start by measuring the effect of LTC insurance on its primary target: the long-term care sector itself. We focus specifically on employment in nursing homes as our key outcome measure. Nursing home care is one of the most labor-intensive parts of the healthcare system and accounts for the majority of spending in long-term care. We then take a broader perspective, examining whether the new public funds flowing to the long-term care sector affected the overall economy. Our causal research design is similar in spirit to the approach in Finkelstein (2007) and Kondo/Shigeoka (2013). We take advantage of the geographic variation in partial public assistance for long-term care services that existed prior to the implementation of the universal insurance program. The results of our positive analysis provide several key insights about the relationship between insurance and the supply side of care. First, we observe a dramatic increase in the number of nursing homes and workers in this labor-intensive industry following LTC insurance expansion. We estimate that a 10 percentage point increase in the share of insured LTC patients led to 0.06 (6 percent) more inpatient LTC firms and four (12 percent) more workers per 1,000 age 65 and older individuals in Germany. Scaling this to the aggregate level of expansion, which offered IAB-Discussion Paper 03|2025 8 1975 and 2008.13 We use this sample to provide a deep-dive into the reform’s effects on employment in LTC, which we refer to as the partial equilibrium analysis. We focus on workers in nursing homes, who account for the lion’s share of LTC employment, and can be best identified in our data.14 Our second data extract comprises full labor market biographies for a 10 percent random sample of workers observed in 1975 to 2008. We refer to the subset of these data that is restricted to years 1985 to 2004 as the “Labor Market Sample (LMS)”. Starting the analysis in 1985 ensures a minimum ten-year look-back period for identifying workers who re-enter the labor market. Moreover, it mitigates potential concerns that geographic variation in overall employment in 1970s may correlate with geographic variation in the eligibility for welfare programs among older adults in the 1990s. Limiting the sample to pre-2005 data further avoids years with substantial changes in the unemployment benefit system introduced by the 2005 Hartz IV reforms (Price, 2019). These reforms led to significant gaps in unemployment benefit histories, compromising the reliability of longitudinal unemployment analyses for years following 2004 (Antoni et al., 2019). 2.2.2. Characteristics of Workers We observe the year of birth, sex, nationality, and educational attainment for individuals in both samples. For employment spells, we further observe the (anonymized) employer identifier, employer’s industry code, geographic location (county) of the employer, whether employment was full-time or part-time, and the employee’s average daily wage. See Appendix A1.2 for additional information on these variables. We construct several measures of labor market experience for all individuals in our samples. 13 We define nursing homes as establishments categorized under WZ73 industry codes for private and for-profit institutions or “homes” (710), private and not-for-profit homes (711), and homes in public ownership (712). We use time-consistent industry codes following the procedure of Eberle et al. (2011) and consider only “regular” employment, as defined by the IAB convention—see Appendix A1.2 for details. Not-for-profit institutions, typically affiliated with or owned by Catholic and Protestant churches, hold a significant market share in the German LTC sector. While for simplicity of exposition we refer to all “homes” that fall under these WZ73 industry codes as nursing homes, in practice the industry classification is coarse and includes several different types of care and assisted living institutions that may have different formal certifications. Some establishments may function more like retirement homes with minimal assistance for daily living, while others may provide intensive skilled nursing care (yet others may serve populations outside the elderly, such as adults and children with disabilities). All such facilities could be affected by all types of LTC insurance benefits and hence we include them in our analysis. However, not all establishments that we classify as nursing homes qualify for inpatient LTC insurance benefits, which require licensing under the German SGB XI. The latter certification typically requires specific skilled nursing services to be offered on site. LTC insurance statistics report 6,564 such licensed facilities (Bundesministerium für Gesundheit (2001), Anlage 8), while our sample includes 11,401 facilities—approx. 74% more. We account for this discrepancy in the welfare analysis presented in Section 5. 14 IEB industry codes have limited precision in the years we analyze, making it challenging to identify formal outpatient care providers. IAB-Discussion Paper 03|2025 15 For workers in the nursing home sample we construct a measure of the general labor market experience by counting the number of years an individual was employed over a fixed 15-year look-back window. Analogously, we construct a measure of nursing home experience by counting the number of years the individual worked in a nursing home establishment over the prior 15 years. The 15-year look-back window restricts the number of years that we can include in regressions that use experience as an outcome variable, to years 1990 to 2008, but helps avoid a censoring bias in the early years of our sample period. For each nursing home employment spell, we additionally classify each worker-year observation into a nursing home incumbent or a new hire. A worker is considered to be a new hire in year t if the person was regularly employed in a nursing home in year t but not in year t − 1. For workers in the labor market sample, we classify each worker-year observation as being either employed or unemployed (see precise definitions in Section A1.2). If a worker is unemployed in year t, we also create an indicator for long-term unemployment status, which requires being unemployed in two consecutive years. We consider a worker to be rejoining the labor force in year t if the worker was not observed in the data in year t − 1, but was observed in at least one year prior to t − 1, and was age 55 or younger (and hence not close to early retirement) in year t. 2.2.3. Mortality We use two sources of mortality data. First, we obtain county-level mortality rates for individuals age 75 and older from the vital statistics of West German states, covering the years 1991 to 2008.15 The second source is the Human Mortality Database, which allows us to compute age group-specific mortality for years 1991–2008 for West Germany and 27 other countries. 2.2.4. Means-Tested and Universal LTC Beneficiary Counts We use historical statistical reports to compute the number of individuals who received means-tested Hilfe zur Pflege in 1993. The most granular geographic resolution at which we were able to obtain Hilfe zur Pflege recipient counts consists of 15 geographic regions covering the former West Germany. These regions include state-level observations for all states (Länder) and seven sub-state regions (Regierungsbezirke) within Bavaria. We sourced 15 Data from 1994 onwards are publicly available as online resources (Statistische Ämter des Bundes und der Länder, 2021a). Mortality data for years 1991–1994 were obtained through written requests from the statistical offices of the respective German federal states. IAB-Discussion Paper 03|2025 16 state-level recipient counts from Statistisches Bundesamt (1993: p. 96) and obtained counts for Bavaria’s Regierungsbezirke from Statistical Office of Bavaria (1993: p. 297). We obtained LTC claim counts from federal LTC insurance program reports (Statistisches Bundesamt, 1999b: p. 7). We use counts from 1999, which is the first year with reliable state-level LTC insurance statistics. We extracted corresponding data for Bavaria’s seven Regierungsbezirke from Statistical Office of Bavaria (1999). 2.3 Summary Statistics Table 1 provides descriptive statistics for the two analytic samples. The nursing home sample summarized in column (1) consists of 25.3 million observations for 1.6 million unique workers across years 1975 to 2008. Individuals in the sample are, on average, 36 years old. 78 percent are female, and 94.5 percent are German nationals. About 7 percent of the workers have completed the upper tier of high school (Abitur). 58 percent of observations are for employment in the healthcare sector, while 8.7 percent are unemployment episodes. The employment spells span nearly a million unique establishments of any kind, of which 22,629 are nursing home establishments.16 Nursing home spells, summarized in column (2), account for 10.2 million individual-year observations. Nursing home spells are more likely to occur at a slightly older age (40 vs. 36 years old) and be part-time (25 percent of all employment spells vs. 29 percent for nursing home spells). The share of women is 82%. Workers earn 2.5 percent higher wages during their nursing home spells (78.60 EUR/day vs. 76.70 EUR/day, in 2020 Euros). About 59 percent of nursing home employment episodes are in not-for-profit (mainly church-owned) nursing homes, 27 percent in for-profit, and 14 percent in publicly-owned institutions. The labor market sample, summarized in column (3) of Table 1, has 44 million observations for 3.8 million individuals in years 1985 to 2004. This sample is representative of the German socially insured workforce. Workers are on average 38 years old and 93 percent are German nationals. 41.5 percent of workers are women. 7.7 percent of workers are unemployed, 13 percent work part-time. The average daily wage is 96.4 EUR. In sum, we see that individuals who work in nursing homes at some point in their careers are predominantly women, are more likely to work part-time, and earn about 20 percent lower wages than the average worker in Germany. 16 Recall that we are only able to count establishments and not firms in our data. Also see footnote 13 for the exact definition of which establishments are included in our count of nursing homes. IAB-Discussion Paper 03|2025 17 3 Research Design 3.1 Source of Variation Our empirical strategy takes advantage of the geographic variation in the prevalence of the means-tested Hilfe zur Pflege program prior to the rollout of universal LTC insurance. After the introduction of LTC insurance in 1995, all individuals with medically documented LTC needs became eligible for the new insurance benefits.17 The new program, however, had little financial impact on individuals already receiving Hilfe zur Pflege, as their care needs were effectively already covered by an existing public program. Consequently, regions with a higher pre-reform share of Hilfe zur Pflege coverage were likely less affected by the insurance expansion.18 The variation in pre-existing coverage reflects differences in local Hilfe zur Pflege eligibility determinations as well as the historical financial circumstances of older individuals in 1995.19 To capture this geographic variation in exposure to the new insurance program, we compute the share of individuals with LTC needs who were already “covered” through the Hilfe zur Pflege program prior to 1995 for 15 geographic regions of Germany r (defined in Section 2.2.4). The denominator of this share is the number of claims for LTC insurance in 1999, adjusted back to 1993 to account for region-specific demographic growth. This measures the latent demand for LTC prior to the expansion under the assumption that after the expansion all individuals with medical needs revealed their demand by claiming benefits. We define the numerator to be the count of Hilfe zur Pflege recipients in 1993, just before the new insurance program was signed into law. Our exposure measure Er is then: Hilfe zur Pfleger,1993 Er = 100% − , (1) gr,1993,1999 × LTC Insurance Claimsr,1999 17 Medical necessity for LTC is determined by independent medical assessors who have no financial incentive to approve or deny applications. According to Nadash/Doty/von Schwanenflügel (2018), benefit determinations, including denials, have generally been accepted as reliable and fair, and, if appealed, are rarely overturned. 18 There are two possible countervailing channels that may complicate this interpretation. Patients qualifying for Hilfe zur Pflege can select cash benefits over formal care benefits in the post-reform period. This is primarily relevant for patients with relatively minor care needs, who prefer informal over formal home care, and arguably less relevant for patients considering nursing home care. This potential substitution effect away from formal care could have been more pronounced in regions with a higher pre-reform share of Hilfe zur Pflege coverage, thus attenuating the reform’s impact in these regions (and thus amplifying our estimates of the causal effect). On the other hand, areas with a higher share of Hilfe zur Pflege payments that were now displaced by federal funds would have experienced a higher windfall to local budgets, which could have independently affected local economies, thus amplifying the reform’s impact in these regions (and thus attenuating our estimates of the causal effect). 19 Notably, our measure of variation does not simply reflect differences in contemporaneous incomes across areas. Figure A3 shows that exposure is uncorrelated with average area-level income in 1993. IAB-Discussion Paper 03|2025 18 The deflation factor gr,1993,1999 captures region-specific aging trends, computed as the ratio of the population aged 65 and older in 1993 to that in 1999. gr,1993,1999 thus deflates the 1999 count of LTC insurance recipients by the change in the number of older individuals in each region r between 1993 and 1999. Intuitively, Er measures the share of individuals with LTC needs who did not have insurance coverage for these needs prior to the reform.20 The geographic variation in Er is visualized in Figure A4, which plots the region-level variation in Er and the mapping of counties to regions. On average, across counties,21 68.7 percent of individuals with LTC medical needs gained insurance coverage. This exposure rate exhibited notable regional disparities, ranging from less than 60 percent in North Rhine-Westphalia to nearly 80 percent in parts of Bavaria. Using this source of variation, we estimate an event study specification to examine whether areas with greater exposure to insurance expansion experienced differential changes in the outcomes of interest. For an outcome Y in county c within region r in year t, we c(r)t estimate: t=2008 X Yc(r)t = αc × 1(countyc) + δt × 1(yeart) + λt × (Er) × 1(yeart) + c(r)t (2) t=1975 where αc are county fixed effects that capture time-invariant differences in outcomes across counties, and δt are year fixed effects that non-parametrically capture a common time trend across counties. The coefficients of interest, λt, multiply the interaction between year fixed effects and the (time-invariant) measure of exposure Er. These coefficients measure whether areas with greater exposure to the LTC insurance expansion experienced differential growth in outcomes. This flexible specification does not impose a trend break in any specific year, allowing the data to reveal whether any changes in trends emerged after the expansion. The identifying assumption for a causal interpretation of λt is that, absent the insurance expansion, outcomes would have evolved in parallel across geographic areas with different levels of exposure. To simplify the discussion of effect magnitudes, we also report results from a difference-in-differences specification that pools the coefficients of the transition years 20 We cannot distinguish between individuals ineligible for Hilfe zur Pflege and those not taking it up. While incomplete take-up is common and likely present in our context, this distinction is not essential for our analysis, as we are interested in the effective change in the share of subsidized versus unsubsidized LTC patients irrespective of the reason for not having a subsidy. 21 We assign each county the level of Er for region r in which the county is located. Using counties as the unit of analysis allows for more flexible specifications of time- and place-invariant factors and provides a comparable spatial unit of analysis across the different exposure regions. It also better aligns with natural labor markets. We report region-level specifications in the Appendix; the results are similar. IAB-Discussion Paper 03|2025 19 (1994 to 1996) and the post-reform years (1997 to 2008): t=1992 X Yc(r)t = αc × 1(countyc) + γt × 1(yeart) + δt × (Er) × 1(yeart) t=1975 (3) + δ94−96 × (Er) × 1(year94−96) + δ97−08 × (Er) × 1(year97−08) + c(r)t 3.1.1. Partial Equilibrium In our partial equilibrium analysis of the nursing home market, we consider three sets of outcomes Yc(r)t: the number of nursing home workers and establishments, the wage of new hires and incumbent workers, and the demographic characteristics and labor market experience of new hires. We construct all outcomes from individual-level data and aggregate to the county-year level by summing counts (for the first set) or by taking averages (for worker characteristics). To account for varying county sizes and potential differences in aging trends, we scale all count-based outcomes into per capita terms using the count of the population age 65 and older. We also use these population counts as weights in wage regressions. Throughout, we cluster standard errors at the county level. In the Appendix, we report the results of several alternative specifications for our main outcomes of interest. We show specifications that re-estimate Equation (2) at the region r rather than county level (top row of Figure A6 and columns (1)-(4) of Table A1), or that keep the county-level specification, but cluster standard errors at the region r level (bottom row of Figure A6 and columns (5)-(8) of Table A1). We also present the results of specifications that include county-specific time trends (top row of Figure A7 and columns (9)-(12) of Table A1), control for the share of the population age 65 and older in the county (bottom row of Figure A7 and columns (5)-(8) of Table A2), or control for the county-year counts of people aged 65 and older (top row of Figure A8 and columns (1)-(4) of Table A2). Further, we show two specifications that use alternative ways of defining exposure (bottom row of Figure A8, Figure A9, and Table A3). Figure A5 shows that alternative definitions of exposure are highly correlated with our baseline measure. Across all specifications, our main takeaways remain qualitatively the same. IAB-Discussion Paper 03|2025 20 3.2 General Equilibrium We next move beyond nursing homes and consider the overall effects of the LTC insurance expansion on the local economy.22 We consider several county-level employment outcomes, including local rates of unemployment (including long-term unemployment), labor force size, the number of workers re-entering the labor force, average wages, and the total wage bill. A practical challenge is that overall local labor market trends are heavily influenced by place-specific macroeconomic fluctuations, making it hard to separate trend changes related to treatment from the overall volatility. To address this, for our general equilibrium analysis we adopt the synthetic difference-in-differences method, which weakens the reliance on the parallel trends assumption (Arkhangelsky, Dmitry and Athey, Susan and Hirshberg, David A. and Imbens, Guido W. and Wager, Stefan, 2021). This data-driven procedure optimally reweights treatment and control units, ensuring that treatment units are compared to controls that had more similar trends prior to treatment (in our case, this would mean comparing units with more similar macro-economic fluctuations). To apply this method, we convert the continuous exposure Er to a dichotomous indicator. Counties are classified as treated if their exposure is above the median, while those at or below the median are used as control units. This binary measure corresponds to a 9.1 percentage point difference in exposure between treated and control counties. 3.3 Mortality A natural and key question is whether any of the changes precipitated by the rollout of universal LTC insurance ultimately resulted in improved care for the older population—one of the stated goals of the LTC reform (Rothgang, 1997). While we lack direct measures of care quality, we consider mortality at older ages as one way to capture the net effect of the LTC reform on (potential) patients. We examine deaths among the population aged 75 and older at the county level. We estimate Equation (2) as our baseline specification, but also report results from the synthetic difference-in-differences specification in Figure A10a. In addition, we estimate a synthetic control specification that leverages variation in mortality among the elderly (aged 75 and older) between Germany and other high-income countries, rather than variation in exposure across geographic regions within Germany (Figure A10b). In this specification we consider West Germany to be “treated” after 1995, and construct a counterfactual mortality time trend (a “synthetic” West Germany) by reweighing mortality trends in other countries following Abadie/Diamond/Hainmueller (2010). 22 This analysis also captures the effects on the formal outpatient LTC sector and LTC cash benefits, which we cannot investigate separately. IAB-Discussion Paper 03|2025 21 4 Results 4.1 Partial Equilibrium 4.1.1. Industry Expansion Figure 1 plots the time series of the number of nursing home workers (Panel 1a) and the number of nursing home establishments (Panel 1b) per 1,000 individuals aged 65 and older between 1975 and 2008. We show the unweighted average of each outcome separately for counties in regions with above-median exposure (blue solid line) and below-median exposure (red dashed line). Both time series are normalized to the overall unweighted mean across all counties in 1993 (Figure A11 shows the raw time series without this normalization). The grey shaded area marks the years of the insurance rollout. Three facts emerge. First, inpatient LTC employment per capita saw persistent growth over the three decades that we study, with the number of nursing home workers per 1,000 individuals age 65 and older more than tripling from 12 to 40 workers.23 Second, the growth in the nursing home workforce was accompanied by growth in the number of nursing home facilities per capita until the late 1990s, increasing to more than one facility per 1,000 older adults in 1990. However, growth in establishments slowed relative to population growth after this period. Third, we observe that in the post-expansion years, both the numbers of nursing home establishments and of workers per 1,000 elderly were growing visibly faster (or declining slower) in counties that were more exposed to the insurance expansion. In Panels 1c and 1d of Figure 1, we present the results of estimating the event study in Equation (2), with point estimates also reported in Table 2. The estimates suggest that prior to 1995, the rate of growth in the number of workers and in the number of establishments per 1,000 older adults did not differ across geographic areas with different levels of means-tested coverage in 1993. After the rollout of universal insurance, however, growth was more pronounced in areas that were more exposed to insurance expansion.24 The acceleration flattened out around eight years after the new insurance was signed into law. Table 2 presents pooled difference-in-differences estimates from Equation (3). The results in columns (1) and (4) indicate that, on average, a 10 percentage point increase in exposure to 23 This corresponds to an annual growth rate of 3.6%, in contrast to a 0.2% annual growth rate for the overall West German population and 1.2% for the 65+ population during this period (Human Mortality Database). 24 Figure A11 suggests that places with higher exposure (and hence a lower prevalence of subsidized care prior to insurance expansion) had lower levels of workers and establishments per capita prior to the reform. These areas caught up with less exposed areas in the number of workers per capita after the insurance expansion. IAB-Discussion Paper 03|2025 22 the reform (equivalent to a 15 percent increase relative to the mean) led to an additional four nursing home workers per 1,000 individuals age 65 and older (a 12 percent increase) and 0.06 more nursing homes per 1,000 elderly (a 6 percent increase). The increase in the number of nursing home workers is approximately evenly distributed between part-time and full-time workers (columns 2 and 3), suggesting a significantly larger relative effect for part-time workers. Before the reform, the number of part-time workers was less than half that of full-time workers (nine part-time workers per 1,000 elderly in 1993 versus 24 full-time workers). Figure A12 illustrates the time series and event studies for both worker types. A 10 percentage point change in exposure is close to the 9.1 percentage point difference in mean exposure between counties with above- and below-median exposure. We refer to this difference as “in-sample variation” in the third panel of Table 2. Multiplying these “in-sample” per capita effects by the count of individuals aged 65 and older, we estimate that offering LTC insurance to an additional 9.1 percentage points of uninsured elderly adds 38,979 nursing home workers and 542 nursing home establishments. We also calculate an out-of-sample estimate of the aggregate impact of LTC insurance. This extrapolation assumes linear scaling with exposure, which is inherently more speculative. The out-of-sample prediction suggests that expanding universal coverage by 68.7 percentage points (i.e. moving from 31.3 percent of the potential patient population being insured to full coverage; or from 0 percent to 68.7 percent) would result in 0.42 more nursing home establishments per 1,000 older adults (a 40 percent increase relative to the mean of 1.04 in 1993) and nearly a doubling of the nursing home workforce, adding close to 300,000 nursing home jobs. To contextualize these employment effects, we construct a coarse arc elasticity by dividing the extrapolated increase in employment by the implicit change in the out-of-pocket price for an average consumer: ΔQ/(Q1 + Q2) arc = = −0.74 . (4) ΔP /(P 1 + P 2) The numerator considers changes in employment per 1,000 older adults (ΔQ=30.5 workers) relative to the average employment between the pre-reform period (Q1=33.1 workers as seen in column (1) of Table 2), and the post-reform period (Q2 = Q1+ΔQ=63.6 workers). The denominator considers the change in prices for inpatient care. Pre-reform, on average 68.7 percent of potential patients paid the full price of care out-of-pocket, while the remaining 31.3 percent were fully insured through means-tested Hilfe zur Pflege. Hence, the average out-of-pocket price pre-reform was P 1 = 0.687Pmarket. Post-reform, cost-sharing of the newly insured drops to 40 percent (cf. footnote 4). The new average out-of-pocket IAB-Discussion Paper 03|2025 23 price then becomes P 2 = 0.687 × 0.40Pmarket, or 0.28Pmarket.25 Put together, this suggests an arc elasticity of -0.74, which significantly exceeds the elasticity estimates in the RAND or the Oregon experiments (Newhouse/Insurance Experiment Group, 1993; Finkelstein et al., 2012). Our data on firm entry suggest that fixed costs of investment may be one explanation for these differences (Finkelstein, 2007). 4.1.2. Anatomy of Expansion In this section, we characterize the nature of the nursing home industry expansion. We consider the change in wages and the composition of newly hired workers. Figure 2 summarizes all point estimates. Price effects We start by examining whether the large expansion of the nursing home workforce was accompanied by growth in wages. We expect that firms increased their wages in order to attract more new hires or to improve the retention of incumbent workers. Table 3 displays the results from estimating Specifications (2) and (3) for (log) daily wages of new full-time hires (columns 1 and 2) and of incumbent full-time workers (columns 3 and 4).26 We find no evidence of systematic increases in wages on average, for either the new hires or the incumbents. For the new hires, the point estimates are close to zero when we control for experience (column 2), suggesting that the wage decline in column 1 is driven by a shift towards lower-skilled workers. For incumbents, we find no evidence for a change in wages. This can be most clearly seen in the event study coefficients for the first and second year post expansion (λ1997 and λ1998), as these coefficients mostly capture the effects for workers who were hired prior to the reform. The lack of an increase in wages in a rapidly expanding sector is consistent with either a perfectly elastic labor supply curve among lower-skilled workers, or with the existence of labor market frictions and the prevalence of excess labor supply prior to the insurance expansion. For example, nursing home labor markets may not (just) clear in prices in the presence of search and matching frictions that interact with tax wedges, unemployment insurance benefits, and collective bargaining agreements, all of which characterized the 25 If we assume (1) that the entirety of the employment expansion is due to new patients without Hilfe zur Pflege, (2) that the proportion of inpatient patients among all patients under Hilfe zur Pflege is the same as under full insurance in 1999, and (3) that the staffing ratio of patients to workers is fixed, then we obtain an even higher elasticity. In this case, the relative increase in patients is larger (considering a smaller base of non-Hilfe zur Pflege patients). At the same time, the observed increase in demand combines a substitution and an income effect, which we consider in the structural model outlined in Section 5.2. Isolating the substitution effect, the model estimates suggest a demand elasticity of -1.19. 26 Event studies are in Figure A13. To allow for individual-specific controls in county-level regressions, we first residualize individual-level log wages with respect to worker fixed effects and then compute county-year level averages of these residuals. IAB-Discussion Paper 03|2025 24 saving or cost for the government and on the margin (under simplifying assumptions revisited in Section 5.2) no additional willingness to pay by workers.41 This is the point highlighted by Baicker/Chandra (2012)—reallocating workers to subsidized sectors from other sectors of the economy generally does not generate welfare gains on labor markets. Our results in Section 4, however, suggest that some marginal workers entered the nursing home sector out of unemployment or from out of the labor force. For these workers, wages represented a source of surplus, while the government collected additional taxes and reduced its unemployment insurance expenditures. To quantify these benefits, we make several assumptions. First, we assume that for every extra Euro in wages, the German government collected 60 cents in taxes and social insurance contributions.42 Second, following Mui/Schoefer (2024), we assume that, on average, workers’ reservation wages were 90 percent of the after-tax (and after-social insurance contribution) income.43 This implies that a Euro in gross wages generates 4 cents (10 percent of the post-tax 40 cents) in worker surplus for successful employment matches among workers who were formerly unemployment or out of the labor force. Adding this worker surplus, government tax revenue, and unemployment insurance savings to our MVPF estimates above, we obtain a combined MVPF of the insurance expansion in the skilled inpatient nursing home market of 1.06. In other words, when we account for workers who found new jobs as a result of subsidizing patients’ demand for care, we get positive returns on a dollar of public spending. General Equilibrium We now turn to the general equilibrium analysis, asking: What was the overall economic return on each dollar of public funds spent subsidizing long-term care through the universal long-term care insurance program? For this computation, we use the most conservative version of our general equilibrium labor market estimates—the lower bound of the 95 percent confidence intervals. The general equilibrium effects are driven primarily by fiscal externalities in the labor market. The government gains additional income tax revenue and saves on unemployment benefits as workers transition into higher-wage jobs, enter the labor market, and exit unemployment. Our estimates indicate that the government saves 4.7 billion EUR in unemployment insurance payments and collects an additional 26 billion EUR in income taxes and contributions to social insurance programs (see column (3) of 41 Workers could have experienced gains from having a more robust labor market, higher job finding rates from increased vacancy posting, potential gains in amenities, or other factors, none of which we are able to capture in this computation. We revisit the role of vacancy posting and non-wage compensating differentials in Section 5.2. 42 In 1999, a single worker with no children paid 18.6% in income tax and 41.6% in social insurance contributions, the latter statutorily split equally between the employer and the employee (Bundeszentrale für Politische Bildung, 2023). 43 Mui/Schoefer (2024) estimate the median reservation “raise” (which aims to capture a notion of reservation wages) to be circa 90% of observed post-tax and post-contributions wages in the German context. IAB-Discussion Paper 03|2025 31 Table 8). These fiscal gains far exceed the 8.4 billion EUR in additional spending on long-term care subsidies, for 5.9 billion EUR in patient surplus. In summary, the policy achieves an infinite MVPF in general equilibrium—it not only pays for itself but also generates surplus for the government. Summary To summarize, the MVPF analysis of the universal LTC insurance expansion when focused only on partial equilibrium without accounting for labor market effects concludes that the MVPF is below 1. This reflects the familiar textbook case of deadweight loss from moral hazard and is a common finding for many healthcare programs, especially aimed at older adults. Our point estimate is very similar to other MVPF estimates from health insurance expansions to adult populations, as discussed in Hendren/Sprung-Keyser (2020) and summarized in Figure A16. In contrast, when we add the willingness to pay of workers who were affected by changes in consumer demand, as well as the corresponding fiscal externalities in taxes and UI savings, we conclude that a marginal dollar invested into LTC insurance generates a substantial fiscal payoff. This puts health insurance expansions to adult populations more on par with programs for children which frequently achieve infinite MVPF. However, the mechanism here is different, as it operates through the indirect benefits of the economic stimulus to local labor markets rather than the direct benefits to program recipients. We emphasize that our estimates are specific to the German labor market context of the 1990s. The large employment gains and significant positive fiscal externalities we document are tied to high contemporary baseline unemployment rates and low labor force participation. These conditions, in turn, were at least partially the result of substantial labor market wedges—high unemployment benefits and tax rates (Dustmann et al., 2014). As such, our MVPF estimate here is not a structural parameter of health insurance expansions, but is a reflection of the economic circumstances and existing tax and transfer programs under which the reform was implemented. In the next section, we offer a formal model that allows us to speak to the external validity of our estimates by quantifying the role of these incumbent economic conditions. 5.2 Model-based Approach We specify and estimate a structural model of the labor market. Building on Acemoglu/Shimer (1999) and Wright et al. (2021), we consider a directed search model, in which finding a job is a costly process and markets do not clear solely in prices. We IAB-Discussion Paper 03|2025 32 incorporate worker and firm heterogeneity, unemployment benefits, income taxes, and collective bargaining to reflect the institutional context. We pin down the parameters of the model using our causal estimates from Section 4 as target moments. We then use the model to revisit the welfare calculation. 5.2.1. Theory Environment: The economy is populated by a continuum of potential firms and workers. Workers differ in their skill level φ. We assume that each firm has a production technology that requires one worker. Let j ∈ J index the sectors of the economy—all firms in j have homogeneous production technologies. Workers and firms meet via search. Firms first decide whether to enter a sector and which skill segment to enter in. Conditional on entry, each firm posts a φ vacancy with a wage w ; the wage will be common among all firms in j in equilibrium. If j is j φ subject to binding wage-setting frictions, e.g. due to collective bargaining, firms set w to j equal the wage floor. In the next stage, workers observe all wage offers for their skill type and decide whether to search and apply to one job offer, or whether to stay out of the labor market. We denote the ratio of the number of applicants to the number of vacancies in each φ sector j and skill segment φ with q > 0 and refer to this as the queue length. Each j φ φ applicant is hired with a probability µ(q ). If hired, the worker earns wage w , pays taxes at j j rate τ, and produces φ output units. Otherwise, the applicant remains unemployed and obtains unemployment benefits as well as benefits from home production. The probability of being hired decreases in the queue length. Conversely, the probability of filling a φ φ φ φ vacancy, denoted with η(q ) = µ(q ) × q , increases in qj . Intuitively, firms have higher j j j chances of filling vacancies in labor markets that have more applicants for any fixed number of open positions. Payoffs: There are three types of agents in the economy—firms, workers, and consumers of the final good. We describe each of their payoffs in turn. A firm’s payoff is its profit. For a firm in sector j that has entered skill segment φ, the profit is given by: φ φ φ π = η(q ) × (Pj × φ − w ) − cj (φ). (5) j j j cj (φ) denotes the cost of vacancy posting, capturing recruiting and retention costs. These costs are incurred with certainty—irrespective of whether the firm manages to fill the vacancy. We allow these costs to vary across sectors and worker skill, reflecting that it may IAB-Discussion Paper 03|2025 33 φ be costlier to hire higher-skilled workers. With probability η(q ), the firm fills its vacancy, j produces output, collects revenue, and pays wages. This is captured by the first term of the φ profit function. The firm pays wage w to matched workers, who produce φ units of output j that the firm can sell at price Pj per unit. Worker i maximizes her expected utility by directing her search toward sector j or staying out of the labor force. The expected utility of worker i with skill φ is given by: ⎧ ⎨ φ φ φ u µ(q ) × ue((1 − τ) × w ) + (1 − µ(q )) × u if i searches in j j j j + ij Uij = , (6) ⎩ ool u + i0 if i doesn’t search where ue denotes the payoff from employment when matched to a vacancy. This payoff φ depends on the after-tax wage (1 − τ) × wj . uu is the payoff from unemployment, which occurs if the worker searches but is not matched and increases in unemployment benefits. uool is the flow utility from staying out of the labor force and not searching. Workers who do not search are assumed to obtain utility from home production and leisure, as well as the benefit of foregoing the hassle cost of applying for jobs along with potential stigma effects of not being formally employed. ij denotes idiosyncratic preference shock that worker i may have for sector j. Prior experience in sector j, or living closer to firms in sector j, or having a small child at home and needing more flexibility and proximity to work may lead the worker to obtain a higher utility from choice j, all else equal. Finally, we consider a representative consumer who has preferences over output produced in all sectors of the economy. The consumer solves where v(Q) is the utility from consuming the vector of output quantities Q and max v(Q) Q X (7) s.t. y ≤ Pj × Qj (8) j P Pj × Qj j denotes consumer expenditures on all goods j ∈ J. y denotes income derived from earnings and transfers, as detailed below. Equilibrium: Building on proposition 1 in Acemoglu/Shimer (1999), we define the search equilibrium as a tuple of wages, queue lengths, output prices, and output quantities that maximizes the worker’s expected utility, Uij , subject to the following constraints. First, firms maximize expected profits, taking output prices and any institutional wage-setting constraints as given (Equation 10). Second, under free entry, firms’ expected profits are zero (Equation 11). IAB-Discussion Paper 03|2025 34 Third, output markets clear in each sector (Equation 12). max Uij (9) φ φ w ,q φφ s.t. w = w if j is constrained by wage floor (10) j φ φ η(q ) × (Pj × φ − w ) − cj (φ) = 0 (11) j j QD j (Pj) = QS j (Pj ) (12) In equilibrium with directed search, firms post efficient wages that depend on the elasticity of the matching function (Moen, 1997; Acemoglu/Shimer, 1999). In our setting, wages and vacancy postings are affected by unemployment insurance benefits, income tax wedges, and collective bargaining agreements. A2.1 provides additional model details. 5.2.2. Taking Theory to Data Parametric assumptions We make several parametric assumptions to take this theoretical model to our data. First, we assume a single national labor and product market, abstracting from the geographic variation. We index “markets” by t ∈ 0, 1, distinguishing between the observed post-reform period and the counterfactual without insurance expansion. Second, for tractability, we convert all types of employment to full-time equivalents. Third, we assume that worker heterogeneity is a function of the number of years of health care experience ehc, which ranges from 0 to 18 in our setting, defining 19 different types of workers. We allow for the importance of healthcare experience to be different across sectors, and we also allow for a discontinuity in skill at zero experience, as empirically workers with no healthcare experience are much less likely to work in nursing homes. We impose the following relationship between years of healthcare experience and skills: φj (ehc) = ζ210 j + ζj × ehc + ζj × 1{ehc > 0}. (13) Next, we assume that the economy has three sectors: for-profit nursing homes, public and not-for-profit nursing homes, and all other sectors of the economy. We distinguish between for-profit and not-for-profit nursing homes as the latter may face different wage constraints due to collective bargaining, and may differ in non-wage marginal costs capturing differences in firm objectives (Lakdawalla/Philipson, 1998). The wages in not-for-profit and public nursing homes are then determined exogenously according to a second-order polynomial in the marginal revenue product net of non-wage marginal costs; see Sections IAB-Discussion Paper 03|2025 35 A2.2 and A2.3 for details. Search and matching frictions help reconcile the large changes in employment absent significant changes in wages. Following Buchholz (2022), we assume the following form of the matching function:  φ  φ jt η(q ) = 1 − exp(−q ) . (14) jt φ λj We allow the parameter governing the matching efficiency λ to vary flexibly by skill level φ and sector. Vacancy posting is costly, and we allow the vacancy costs c to vary by skill level j and between for-profit and non-for-profit/public nursing homes. Turning to the worker’s flow payoffs, we assume: e,φ φ u = (1 − τ) × w + ξj + ν ∗ 1{ehc > 0} ∗ 1{NH} (15) jt jt u u = b + ξu (16) ool u = κ0 + κ1 ∗ 1{ehc > 0} (17)   ijt = γ × ϑig + (1 − ρ) × ˜ijt . (18) In the flow utility from employment, ξj captures compensating differentials of working in sector j. Parameter ν allows compensating differentials to vary by health experience to capture differential attachment to working in nursing homes for workers with some health care experience. We normalize the compensating differential in the outside sector to zero. In the flow utility from unemployment, b denotes unemployment benefits and ξu captures monetized non-pecuniary utility from unemployment. Similarly, uool captures monetized non-pecuniary utility from being out of the labor force which we allow to vary based on whether workers have any health experience. Lastly, we impose an assumption on the distribution of idiosyncratic worker preferences that leads to a nested logit choice problem for the worker. We group the two nursing home sectors into one nest g, keeping the outside sector and being out of the labor force as separate nests. ϑig denotes a taste shock that worker i may have for nest g, whereas ˜ijt is an identically and independently distributed extreme value shock for sector j. The nesting parameter 0 ≤ ρ < 1 governs the correlation of worker utility across different types of nursing homes, and γ denotes a scaling parameter. IAB-Discussion Paper 03|2025 36 Finally, we assume that the representative consumer has CES preferences, and that nursing homes produce homogeneous outputs. We then write the consumer problem as:  σ−1  σ σ−1 Qo,t σ−1 σ v(Qt) = αo × Jo × ( Jo ) σ + αNH × QNH,t (19) ˜ s.t. QNH,t = QNH,t + HzPNH (20) yt + Tt ≥ Pot × Qot + (PNH,t − sNH,t) × QNH,t . (21) We assume that the outside sector comprises Jo homogeneous subsectors. The α denote scaling parameters and σ denotes the elasticity of substitution. QNH denotes the demand for nursing home care services paid out of pocket. HZPNH denotes the demand by patients insured via Hilfe zur Pflege, which we treat as exogenous. Income y is given by the sum of the wage earnings of workers, firm profits (zero in equilibrium), vacancy costs (earned by recruiters), and other marginal costs. T denotes a lump-sum transfer from the government to the consumer, which is income taxes net of government spending on nursing home care. The main policy parameter of interest for our analyses of insurance expansion is sNH , which captures the product subsidy that the government provides to the nursing home sectors. Estimation Our estimation strategy proceeds in two steps. In the first step, we estimate the parameters governing worker preferences, production, and the matching technology via generalized method of moments (GMM). In the second step, we impose the market clearing conditions in the product market to recover the preference parameters of the representative consumer. We use two primary sets of moments for estimation: first, the observed employment shares and mean wages by sector and years of health care experience in 1999; and second, the causal estimates from Section 4.1.1. Figure 6 illustrates the key empirical moments and model fit. Section A2.2 describes the details of the estimation approach. The details of all parameter estimates are reported in Section A2.3. We estimate an elasticity of demand to out-of-pocket prices for nursing home care of -1.19, slightly larger than our estimate of the arc-elasticity in Equation (4). The model also allows for (and estimates) substantial income effects from the insurance expansion. The importance of income effects in our context is similar to the role of income effects found in Finkelstein/Hendren/Luttmer (2019) for Medicaid coverage in the U.S. IAB-Discussion Paper 03|2025 37 5.2.3. Welfare and Counterfactuals Table 9 summarizes our welfare analysis. Columns (1) through (3) keep the tax and transfer environment fixed. We report the equilibrium labor market allocation in Panel A and surplus in Panel B. Moving from column (1) to column (2) shows how the allocation and surplus change when we introduce LTC insurance into the economy.44 As in the causal evidence of Section 4.1.1, the model simulation of LTC subsidies results in an increase in nursing home employment and overall labor force participation. We decompose the change in welfare into three components: ΔW = ΔConsumer Welfare + ΔWorker Welfare + ΔGovernment Surplus . (22) We measure the change in consumer welfare through the equivalent variation, considering changes in out-of-pocket prices but holding pre-reform income y0 + T0 fixed. Consumers benefit from lower out-of-pocket prices for nursing home care but are harmed by the price increase for the outside sector good (not considered in Section 5.1).45 Worker welfare is also measured through the equivalent variation, capturing changes in vacancy postings, after-tax wages, and hence earnings.46 Finally, the government surplus is defined as tax revenues minus LTC subsidy spending and UI spending. We find that on net, consumer welfare with LTC insurance goes down in the model. This decline in consumer welfare as well as the extra government outlays on LTC subsidies are, however, offset by welfare gains in the labor markets as well as by a substantial increase in tax revenues collected by the government. As in the model-free calculation, we get that in the general equilibrium environment overall social welfare goes up when LTC insurance is rolled out and that the overall government surplus (tax revenues minus spending on LTC insurance and unemployment benefits) increases, i.e. the insurance expansion pays for itself. Quantitatively, we find a net welfare gain of 0.78 billion EUR per month, or 2,294 EUR per full-time-equivalent nursing home worker and month. 44 We operationalize this by treating column (2) as data—it is the observed environment with insurance in place in 1999 as simulated by the model. Column (1) then reports the results of a counterfactual simulation that removes long-term care subsidies. 45 We first calculate consumer utility at pre- and post-reform out-of-pocket prices, holding income fixed at pre-reform levels. We refer to the before and after expansion prices and utility as P0, u0 and P1 − s, u1, respectively. The equivalent variation is then given by e(P0, u1) − e(P0, u0), where e() denotes the expenditure function. 46 The equivalent variation equals the compensating variation given the worker’s linear indirect utility function and is given by nP P P U ˜ 1 P P P U ˜ 0o ij ij EV = γ ×log[ ( exp( ))1−ρ] − log[ ( exp( ))1−ρ] , where i g j∈gγ×(1−ρ) i g j∈gγ×(1−ρ) U ˜ t= Ut ij ij − ij . IAB-Discussion Paper 03|2025 38 To what extent are these welfare gains from a health insurance expansion likely to generalize to other contexts? We use our model to help us think about this key policy question. Specifically, we explore the sensitivity of our net welfare numbers to the generosity of unemployment insurance (UI) benefits, the income tax rate, and the productivity of the outside sector. We focus our discussion on relatively large changes in these primitives and present analogous results on more marginal changes in Appendix A2.4. In our first counterfactual, we revisit the withversus without-expansion outcomes after setting UI benefits to zero.47 We find a lower unemployment rate but also a lower labor force participation rate suggesting that more individuals refrain from searching for jobs altogether. Columns (4)-(6) of Table 9 document these facts. We find that the insurance expansion in this context has similar effects on social welfare suggesting that the generosity of UI benefits plays a secondary role in driving the welfare results.48 We next consider a 50 percent drop in the income tax rate. We estimate a larger labor force participation in the no-insurance counterfactual and find that under these circumstances the insurance expansion is less successful in increasing labor force participation further; see columns (7)-(9). In the context with lower tax rates and higher labor-force participation, expanding insurance leads to a significant share of new nursing home employees to be hired away from productive jobs in the outside sector. In this environment with much lower taxes, we no longer find that the insurance expansion pays for itself and estimate an overall welfare loss of 10 million EUR per month. Finally, we simulate the interaction between productivity growth in the outside sector and the insurance expansion. As wages increase in the outside sector, this also drives up wages in nursing home care. Without productivity growth in the nursing home care sector, wage increases will lead to higher output prices, illustrating Baumol’s cost disease (Baumol/Bowen, 1965). We consider a 50 percent increase in productivity and again estimate substantially higher labor force participation, see columns (10)-(12).49 In contrast to the tax counterfactual, we now estimate substantially higher nursing home wages and, as a result, higher nursing home output prices. We find smaller increases in nursing home employment and larger increases in prices, implying that the insurance expansion is less effective in expanding nursing home jobs. We again estimate smaller gains in labor force participation. Overall we estimate smaller welfare gains from the insurance expansion both 47 To ensure weakly positive job finding and vacancy filling rates, we gradually increase the matching parameter λ in 1% increments for the respective skill-sector type until the job finding rates for each skill and sector are bounded between 0 and 100%. 48 Here we note, however, that our model can only account for a small share of the reform-induced decline in unemployment in our baseline model, thereby possibly understating the role of UI benefits; see Section A2.2, Table A8. 49 The calibrated productivity increase reflects the cumulative growth in productivity from the mid-1990s to the present. This serves to assess how differences in productivity between these environments might influence the welfare effects of the LTC reform. IAB-Discussion Paper 03|2025 39 6 Conclusion in absolute terms and even more strikingly in relative terms—social welfare in column (10) exceeds baseline welfare in column (1) by 77 percent. In summary, we find that our welfare estimates are sensitive to the economic environment in important and policy-relevant ways. Lower income taxes or higher productivity in the outside sector render policy efforts to stimulate the economy less effective. We also find smaller positive fiscal externalities and gains in the labor market surplus in economic environments with higher labor force participation to begin with. When workers define the shorter market side, policy-induced employment expansions will largely involve the reallocation of workers across sectors of the economy, muting potential welfare gains in the labor market (Baicker/Chandra, 2012). Arrow (1963) and Feldstein (1971, 1977) hypothesized that demand-side moral hazard induced by health insurance can trigger supply-side expansions in healthcare markets. Capturing this phenomenon empirically has proven challenging. In this paper, we combine unusually comprehensive administrative labor market data with a rarely observed large-scale, universal insurance rollout—Germany’s 1995 introduction of national long-term care (LTC) insurance—to examine how a public insurance expansion can reallocate workers across sectors. We find that the LTC insurance rollout led to a dramatic expansion in LTC employment. While many more workers were hired into LTC jobs, average wages did not increase, consistent with considerable slack in the labor market. Notably, this surge in LTC hiring drew in workers who likely faced obstacles to employment, such as those with less experience, lower skill level, or extended gaps in work histories due to long-term unemployment or non-participation in the labor force. By providing “good” jobs for for lower-skilled workers, the reform aligns with an aim of industrial policy recently highlighted by Rodrik (2022). We further find that the effect of insurance expansion on the labor market was not limited to LTC workers. Instead, we estimate a reduction in overall local unemployment, and increased labor force participation. To the best of our knowledge, this provides some of the first evidence of factor substitution between sectors of the economy following a health insurance expansion. Our normative analysis considers the external validity of the German experience and leads to two main policy-relevant takeaways. First, public spending on programs for older adults IAB-Discussion Paper 03|2025 40 Propper, Carol; Van Reenen, John (2010): Can Pay Regulation Kill? Panel Data Evidence on the Effect of Labor Markets on Hospital Performance. In: Journal of Political Economy, Vol. 118, No. 2, p. 222–273. Rodrik, Dani (2022): An Industrial Policy for Good Jobs. The Hamilton Project, Brookings Institution. Rothgang, Heinz (1997): Ziele und Wirkungen der Pflegeversicherung: Eine Öekonomische Analyse. In: Campus Verlag. Schmieder, Johannes F.; Von Wachter, Till; Bender, Stefan (2012): The Effects of Extended Unemployment Insurance over the Business Cycle: Evidence from Regression Discontinuity Estimates over 20 Years. In: The Quarterly Journal of Economics, Vol. 127, No. 2, p. 701–752. Schwandt, Hannes; von Wachter, Till M. (2020): Socio-Economic Decline and Death: The Life-Cycle Impacts of Recessions for Labor Market Entrants. NBER Working Paper No. 26638. Sojourner, Aaron J.; Frandsen, Brigham R.; Town, Robert J.; Grabowski, David C.; Chen, Min M. (2015): Impacts of Unionization on Quality and Productivity: Regression Discontinuity Evidence from Nursing Homes. In: Industrial and Labor Relations Review, Vol. 68, No. 4, p. 771–806. Staiger, Douglas O.; Spetz, Joanne; Phibbs, Ciaran S. (2010): Is there Monopsony in the Labor Market? Evidence from a Natural Experiment. In: Journal of Labor Economics, Vol. 28, No. 2, p. 211–236. Statistical Office of Bavaria (1999): Pflegebericht Bavaria 1999. Table: “Pflege: Kreise, Pflegebedürftige nach Art des Pflegeheims, Pflegegeldempfänger, Jahr”. Last accessed August 2024. Statistical Office of Bavaria (1993): Regionalbericht 1993. Statistische Ämter des Bundes und der Länder (2021a): Statistik der Sterbefälle. Last accessed September 2021. Statistische Ämter des Bundes und der Länder (2021b): Einkommen der privaten Haushalte in den kreisfreien Städten und Landkreisen der Bundesrepublik Deutschland 1995 bis 2019, Reihe 2, Band 3. Statistisches Bundesamt (2013): Statistik der Sozialhilfe - Hilfe zur Pflege. Statistisches Bundesamt (1999a): Pflegeheime, Vergütung, u.a. nach Region (1999-2015). Last accessed August 2024. Statistisches Bundesamt (1999b): Pflegestatistik 1999 - Pflege im Rahmen der Pflegeversicherung. 2. Kurzbericht: Ländervergleich - Pflegebedürftige. Statistisches Bundesamt (1993): Reihe 2 Sozialhilfe, Fachserie 13, 1993. IAB-Discussion Paper 03|2025 47 Stevens, Ann H.; Miller, Douglas L.; Page, Marianne E.; Filipski, Mateusz (2015): The Best of Times, the Worst of Times: Understanding Pro-Cyclical Mortality. In: American Economic Journal: Economic Policy, Vol. 7, No. 4, p. 279–311. Wright, Randall; Kircher, Philipp; Julien, Benoît; Guerrieri, Veronica (2021): Directed Search and Competitive Search Equilibrium: A Guided Tour. In: Journal of Economic Literature, Vol. 59, No. 1, p. 90–148. IAB-Discussion Paper 03|2025 48 Figure 1.: Introduction of Universal LTC Insurance and the Nursing Home Market High Exposure Low Exposure 10 20 30 40 50 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year High Exposure Low Exposure .7 .8 .9 1 1.1 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year (a) NH Workers per 1,000 65+ Population (b) NH Establishments per 1,000 65+ Population -20 0 20 40 60 80 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year Point estimates -.5 0 .5 1 1.5 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year Point estimates (c) NH Workers, Event Study (d) NH Establishments, Event Study Notes: Panels in the top row plot the number of regular nursing home workers (A) and nursing home establishments (B) per 1,000 individuals age 65 and over, on average across counties for each year 1975 to 2008. The average in each year is computed separately for the group of West German counties with (region-level) exposure variable Er above (“high exposure”) and below (“low exposure”) the level of exposure of the median county. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for long-term care prior to the rollout of universal LTC insurance (mean and median of Er=0.69) All counties with exposure level at the median are assigned to the below median group. For visual clarity, both time-series are normalized to the aggregate mean of the y-axis variable across all counties in 1993 (see Figure A11 for raw time-series without the normalization). Appendix A1.2 provides the definition of nursing homes and “regular” workers. Er is derived in Equation 1 and its geographic variation is visualized in Figure A4. Panels in the bottom row display λt coefficients and 95% confidence intervals from estimating the specification in Equation 2 with the number of regular nursing home workers (C) and nursing home establishments (D) per 1,000 individuals age 65 and above in a county as an outcome. λt are coefficients that multiply exposure variable Er; λt is normalized to zero in the pre-reform year t = 1993. More regression details are reported in columns (1) and (4) of Table 2. Regressions are estimated on the nursing home sample; see column (2) of Table 1 for summary statistics. Source: Own calculations. IAB-Discussion Paper 03|2025 49 Figure 2.: Changes in Nursing Home Wages and Worker Sorting (a) Incumbent and Starting Daily Wage (EUR) 1. Incumbent Wages, No Controls Pre-Mean: 4.58 2. Incumbent Wages, Individual FEs Pre-Mean: 4.58 3. Starting Wages, No Controls Pre-Mean: 4.38 4. Starting Wages, Experience Controls Pre-Mean: 4.38 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 Implied Reform Impact (b) Characteristics, New Hires (c) Origins, New Hires 1. Age (Divided By 10) 2. German National 3. Female 4. Abitur (A-Level Equiv.) 5. NH Apprentice at t-1 6. Part-Time at t 7. Rolling 15-Year LM Experience (Divided By 10) 8. Rolling 15-Year NH Experience (Divided By 10) -0.6 -0.4 -0.2 0.0 0.2 Implied Reform Impact 1. Employed, in Healthcare at t-1 2. Employed, Not in Healthcare at t-1 3. Unemployed at t-1 4. Temporarily Not in LF at t-1 5. Never in LF Before t -0.2 -0.1 0.0 0.1 0.2 0.3 Implied Reform Impact Notes: This panel displays aggregate treatment effects δ97−08 and 95% confidence intervals from estimating the difference-in-differences specification in Equation 3. The coefficient measures the effect of universal long-term care insurance expansion per unit of exposure to expansion, Er. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for long-term care prior to the rollout of universal LTC insurance (mean and median of Er=0.69). Er is derived in Equation 1 and its geographic variation is visualized in Figure A4. The top panel (A) displays the treatment effects on log regular full-time daily wages of nursing home incumbents and new hires. A new hire in year t is defined as an individual not in nursing home employment in year t − 1. Regression details, including coefficients of the event study version, are reported in Table 3. In the bottom row, the outcomes in panel (B) are the characteristics and in panel (C) the immediate employment history of new regular nursing home hires. Tables 4 and 5 report regression details for panels (B) and (C), respectively. Source: Own calculations. IAB-Discussion Paper 03|2025 50 Figure 3.: Changes in Overall Employment 1. Share Unemployment 2. Share Long-Term Unemployment 3. Log Labor Force Size 4. Log Labor Force that is Rejoining the LF 5. Log Average Wage 6. Log Total Wage Bill -.01 0 .01 .02 .03 .04 Implied Reform Impact Notes: This panel displays aggregate treatment effects and 95% confidence intervals from estimating the synthetic difference-in-differences (SDID) specification version of Equation 3, following Arkhangelsky, Dmitry and Athey, Susan and Hirshberg, David A. and Imbens, Guido W. and Wager, Stefan (2021) methodology. The underlying data is the sample of all German labor market biographies—Labor Market Sample (LMS) summarized in column (3) of Table 1. The treatment effect coefficient measure the effect of more exposure to universal long-term care insurance expansion on outcomes specified on the y-axis. We use a binary measure of exposure to the reform, which defines regions with an abovemedian exposure variable Er as treated units (i.e. more exposed) and counties at or below median exposure as control units. Er is derived in Equation 1 and its geographic variation is visualized in Figure A4. “Long-term unemployment” is defined as unemployment observed on in at least two consecutive years. Table 6 reports regression details for all outcomes. Source: Own calculations. IAB-Discussion Paper 03|2025 51 Figure 4.: Introduction of Universal LTC Insurance and Overall Employment (a) Share Unemployed (b) Log Labor Force Size High Exposure Low Exposure SDID ATT: -0.005 (0.001) .04 .06 .08 .1 .12 1985 1988 1991 1994 1997 2000 2003 Year High Exposure Low Exposure SDID ATT: 0.016 (0.005) 8.4 8.45 8.5 8.55 8.6 1985 1988 1991 1994 1997 2000 2003 Year (c) Share Unemployed, Event Study (d) Log Labor Force Size, Event Study -.015 -.01 -.005 0 .005 1985 1988 1991 1994 1997 2000 2003 Year Point estimates -.01 0 .01 .02 .03 .04 1985 1988 1991 1994 1997 2000 2003 Year Point estimates Notes: Panels in the top row display treatment and matched control time series from synthetic difference-in-difference (SDID) specifications, following Arkhangelsky, Dmitry and Athey, Susan and Hirshberg, David A. and Imbens, Guido W. and Wager, Stefan (2021) methodology, for the rate of unemployment in a county (A) and the (log) size of the labor force in a county (B). The underlying data is the sample of all German labor market biographies—Labor Market Sample (LMS) summarized in column (3) of Table 1. The share unemployed uses the size of the labor force in the denominator, which is defined as the count of all employed and unemployed individuals. The treatment group are West German counties with (region-level) exposure variable Er above (“high exposure”) and below (“low exposure”) the level of exposure of the median county. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for long-term care prior to the rollout of universal LTC insurance (mean and median of Er=0.69) All counties with exposure level at the median are assigned to the below median group. For visual clarity, both time-series are normalized to the aggregate mean of the y-axis variable across all counties in 1993 (see Figure A11 for the time-series without the normalization). The bottom panel shows event study charts for the same outcomes, using b = 100 bootstrap draws. Coefficients are normalized to zero in the pre-reform year t = 1993. More regression details are reported in Table 6. Source: Own calculations. IAB-Discussion Paper 03|2025 52 Figure 5.: Introduction of Universal LTC Insurance and Old-Age Mortality (a) Deaths per 100 Age 75+ (b) Deaths per 100 Age 75+, Event Study High Exposure Low Exposure 7.5 8 8.5 9 9.5 10 1991 1994 1997 2000 2003 2006 Year -3 -2 -1 0 1 2 1991 1994 1997 2000 2003 2006 Year Point estimates Notes: Panel a plots the raw time series of mortality per 100 age 75 and over population, on average across counties for each year 1991 to 2008. The average in each year is computed separately for the group of West German counties with (region-level) exposure variable Er above (“high exposure”) and below (“low exposure”) the level of exposure of the median county. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for longterm care prior to the rollout of universal LTC insurance (mean and median of Er=0.69) All counties with exposure level at the median are assigned to the below median group. For visual clarity, both time-series are normalized to the aggregate mean of the y-axis variable across all counties in 1993. Er is derived in Equation 1 and its geographic variation is visualized in Figure A4. Panel b display λt coefficients and 95% confidence intervals from estimating the specification in Equation 2 with the same mortality measure as the outcome. More regression details and alternative specifications are reported in Table 7. Source: Own calculations. IAB-Discussion Paper 03|2025 53 Figure 6.: Cost of Employment and Employment Rates by Sector and Experience (a) Log Monthly Employment Cost (EUR) (b) Employment Type to Population Ratio Nursing Homes Outside Sector 7.8 8 8.2 8.4 8.6 0 3 6 9 12 15 18 Years of Healthcare Experience Data Model Fit Nursing Homes Outside Sector 0 .2 .4 .6 .8 0 3 6 9 12 15 18 Years of Healthcare Experience Data Model Fit Notes: Panel 6a plots average log monthly cost of employment (full-time gross wages and employer social insurance contributions, in EUR) for all full-time workers age 20–64 in the Labor Market Sample in 1999 (i.e. after the full insurance expansion). For each worker, we calculate years of prior employment in the healthcare sector before 1999. Wage averages are then computed within one-year bins of healthcare experience, separately for nursing homes employees (red lines) and workers in other sectors (black lines). Solid lines report observed data; dashed lines show model predictions (Section 5.2). Panel 6b, constructed analogously, displays the full-time equivalent employment shares in nursing homes and other sectors in 1999, using all workers age 20–64. Full-time equivalents are calculated by considering two part-time jobs as one full-time position. Employment categories omitted from the exhibit are formal unemployment and being out of the labor force. Refer to Section A1.2 for definitions of the healthcare sector and nursing homes. Source: Own calculations. IAB-Discussion Paper 03|2025 54 Table 1.: Summary Statistics Nursing Home Sample Labor Market Sample All Spells Nursing Home Spells All Spells 1975-2008 1975-2008 1985-2004 (1) (2) (3) No. of Individual-Year Observations 25,273,408 10,184,624 44,096,416 Individuals No. of Unique Individuals 1,639,576 1,639,576 3,787,607 Mean Age 36.2 40.2 38.0 % Female 78.2 81.8 41.5 % German 94.5 95.0 92.9 % High School Education (Abitur) 7.32 7.01 8.00 % in Healthcare Sector 58.2 100.0 9.65 % Unemployed 8.65 0 7.75 Mean 15-Year Labor Market Experience (yrs) 8.24 8.69 9.18 Mean 15-Year Nursing Home Experience (yrs) 3.50 5.80 0.14 % Part-Timea 25.0 29.2 13.3 Mean Daily Wage (EUR)b All Observations 76.7 78.6 96.4 Nursing Home Observations 78.6 78.6 83.0 Establishments No. of Unique Establishments Any 991,814 22,629 1,748,171 Nursing Homes 22,629 22,629 13,880 Of Nursing Home Employment Spells, Share in For-Profit Nursing Home 0.27 0.27 0.27 Church-Owned Nursing Home 0.59 0.59 0.59 Publicly-Owned Nursing Home 0.14 0.14 0.14 a Conditional on being employed. b In constant 2020 Euros. Notes: The table reports summary statistics for two main analytic samples “Nursing Home Sample (NHS)” and “Labor Market Sample (LMS).” NHS is an extract from the annualized universe of the German Integrated Employment Biographies for years 1975-2008. The extract contains full labor market biographies of individuals who were employed in a nursing home in at least one year in 1975-2008. Employment spells are annualized by taking the spell observed on June 30th of a given year. Nursing homes are defined as establishments with WZ73 industry codes for private and for-profit institutions or “homes” (710), private and not-for-profit homes (711), and homes in public ownership (712). LMS is a 10% draw of the annualized universe of labor market biographies for years 1985 to 2004. In both samples, individual-year observations are excluded if the place of employment is in the (former) East Germany, Berlin, or Bremen. Source: Own calculations. IAB-Discussion Paper 03|2025 55 Table 2.: Event Study Results: Aggregate Response Outcome (per 1,000 Age 65+ Population) Workers Full-Time Part-Time Firms (1) (2) (3) (4) Pooled Coefficients δ97−08 44.4 21.9 22.5 0.62 (8.35) (4.87) (4.90) (0.17) Event Study Coefficients 1-Year Effect, λ1997 24.4 12.3 12.1 0.30 (6.74) (4.21) (3.25) (0.14) 3-Year Effect, λ1999 33.3 18.8 14.6 0.39 (7.91) (4.79) (4.07) (0.17) 5-Year Effect, λ2001 42.9 24.8 18.1 0.54 (8.69) (5.27) (4.74) (0.17) 10-Year Effect, λ2006 53.4 22.6 30.7 0.89 (9.57) (5.55) (6.24) (0.23) Implied Impact Using In-sample Variationa 4.02 1.98 2.03 0.06 Aggregate Impact, West Germanyb 38,979 19,253 19,726 542.1 Level of Outcome in 1993 Mean 33.1 23.9 9.19 1.04 S.D. 13.8 10.3 4.57 0.47 Years 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 No. of Observations 10,948 10,948 10,948 10,948 a Multiplies δ97−08 by 9.1 percentage point difference in mean exposure between counties with above and below the median level of exposure b Scales estimates by 9,699 thousand people age 65+ in West Germany (excluding Berlin and Bremen) in 1993. Notes: This table reports the results of estimating the event study specification in Equation (2) and the pooled difference-in-differences specification in Equation (3). The coefficients measure the impact of long-term care insurance expansion—per unit of exposure (Er)—on the outcome variables as specified in column titles. Exposure measure Er is derived in Equation (1). Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for long-term care prior to the rollout of universal LTC insurance (mean and median of Er=0.69). Outcome variables are: the number of regular nursing home workers, in total (column 1), and separately by part-time and full-time status (columns 3 and 4), as well as the number of nursing home establishments (column 4) per 1,000 individuals age 65 and older in a county. Event study results in columns (1) and (4) are visualized in Figure 1; the estimates in columns (2) and (3) are visualized in Figure A12. All specifications include county and year fixed effects. Standard errors clustered at the county-level are reported in parentheses. Source: Own calculations. IAB-Discussion Paper 03|2025 56 IAB-Discussion Paper 03|2025 Table 9.: Counterfactuals and Welfare Baseline 100% Reduction in UEB 50% Reduction in Income Tax 50% Increase in Productivity Insurance Diff Insurance Diff Insurance Diff Insurance Diff No Yes No Yes No Yes No Yes (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Panel A: Employment and Earnings Nursing Home Employment (million FTE) 0.18 0.34 0.16 0.19 0.34 0.16 0.23 0.47 0.24 0.25 0.36 0.11 Total Labor Costs to Employers (billion EUR/month) 70.6 73.6 2.99 68.8 71.9 3.04 111.5 114.4 2.81 178.7 182.1 3.36 Unemployment (million FTE) 1.69 1.66 -0.03 0.73 0.72 -0.006 1.85 1.80 -0.05 0.88 0.85 -0.04 Labor Force (million FTE) 19.4 19.9 0.50 18.6 19.1 0.53 30.3 30.5 0.28 30.8 31.0 0.20 Panel B: Welfare (billion EUR/month) Consumer Welfare 81.0 80.3 -0.69 81.7 81.0 -0.69 130.2 129.1 -1.10 203.0 201.3 -1.72 Worker Welfare 148.7 149.2 0.59 147.7 148.3 0.60 172.1 173.5 1.35 175.4 176.5 1.12 Government Surplus 33.4 34.3 0.88 33.9 34.8 0.88 25.9 25.7 -0.25 87.8 88.8 0.99 LTC Subsidy Spending -0.42 -1.05 -0.63 -0.43 -1.06 -0.63 -0.42 -1.41 -0.99 -0.57 -1.28 -0.71 UI Spending -1.31 -1.29 0.02 0 0 0 -1.44 -1.40 0.04 -0.68 -0.66 0.03 Tax Revenues 35.2 36.7 1.49 34.3 35.8 1.51 27.8 28.5 0.70 89.1 90.7 1.67 Total Welfare 263.1 263.9 0.78 263.3 264.1 0.78 328.2 328.2 -0.01 466.2 466.6 0.38 Notes: This table summarizes the model-simulated effects of a universal LTC insurance expansion on a number of outcomes in different economic environments. For each environment, we present outcomes (i) without universal LTC insurance, where LTC support is provided via Hilfe zur Pflege only and (ii) with universal LTC insurance coverage. Universal LTC insurance is modeled as a subsidy of s=1,200 EUR per month that can only be applied to the cost of nursing home care by individuals who would have otherwise had to pay the full price of care out-of-pocket. This way of modeling the subsidy maintains the availability of Hilfe zur Pflege. We also present the difference between (i) and (ii). We consider four environments. Columns (1)-(3) present results for our baseline economic environment, columns (4)-(6) revisit the results after reducing the level of unemployment insurance benefits (UEB) from 775 EUR per month to zero. Columns (7)-(9) revisit the results after reducing the income taxes and contributions by 50% (from about 50 p.p. to 25 p.p.). Columns (10)-(12) revisit the results after increasing the productivity in the outside sector for each skill segment, φout, (ehc) by 50%. For each environment, we report several descriptive measures of the resulting allocation (Panel A): nursing home employment, total gross wage payments by employers in all sectors of the economy, the count of workers in unemployment, and the number of individuals in the labor force (employed or unemployed). Panel B reports the corresponding normative measures. We report the level of consumer welfare, worker welfare, and net government spending, which captures fiscal externalities. Extra government outlay is the spending on LTC insurance. The government can also spend more or less on unemployment insurance benefits and can collect more or less revenue in taxes and social insurance contributions. Total welfare add consumer and worker welfare along with changes in government spending. We do not weight the change in government spending by a cost of public funds. Source: Own calculations. 63 Appendix Online Appendix to: Health Insurance as Economic Stimulus? Evidence from Long-Term Care Jobs Martin Hackmann Jörg Heining Roman Klimke Maria Polyakova Holger Seibert IAB-Discussion Paper 03|2025 64 A1 Data Appendix A1.1. Processing of the Raw Data The primary data source for this paper is the Integrated Employment Biographies (IEB) database, which contains the universe of employment and unemployment spells for workers subject to social security contributions in Germany. We use data spanning the years 1975 to 2008. The IEB covers individuals across five employment categories: 1) employment subject to social security (included since 1975), 2) marginal part-time employment (included since 1999), 3) unemployment benefit receipt under the German Social Code Books III (included since 1975) or II (included since 2005), 4) registered job-seekers (included since 2000), and 5) participants in active labor market programs (included since 2000). All spells are recorded with day-level precision. For each employment spell, the database records the industry code of the establishment. To ensure consistency over time, we harmonize industry codes of the three-digit WZ73 classification, following the methodology of Eberle et al. (2011). For spells with missing WZ73 codes, we extrapolate values within establishments across time. When this extrapolation is not feasible, we impute missing WZ73 codes using values from later industry classifications (WZ93 and WZ03), relying on the most frequently observed WZ73 code during overlapping years. We construct two analytic samples from the IEB. The “Nursing Home Sample (NHS)” comprises complete person-histories of individuals employed in nursing homes for at least one day during 1975–2008. We define a nursing home as an establishment with one of the three time-consistent WZ73 industry codes 710, 711, and 712, for private and for-profit institutions or homes (710); private, not-for-profit homes (711); and homes in public ownership (712). We also extract the “Labor Market Sample (LMS)”, a random 10% draw of complete person-histories from the complete IEB population during 1975 to 2008. We subset the NHS and LMS to include only employment spells (originating in the IEB’s “BeH Employee History’’ source data set) and unemployment spells (from “LEH Benefit Recipient History’’ or “LHG Unemployment Benefit II Recipient History’’). We exclude spells with zero daily wage or benefit rate, marginal part-time workers (identified via values of the “employment status [erwstat]” variable equal to 109 or 209), and notifications of lump-sum payments (“reason of cancellation/notification/termination [grund]” with value 154). For included spells, we impute missing values of the workplace location variable [ao_kreis] during non-employment periods with the individual’s most recent non-missing value. If imputation is not possible, we drop the corresponding observations from the analytic samples. IAB-Discussion Paper 03|2025 65 We aggregate the NHS and LMS to the individual-year level by selecting the employment status of the spell that includes June 30th of each year. For unemployment spells that span multiple instances of June 30, we split the spells following the approach of Eberle/Schmucker (2019). If duplicates persist, we retain the spell with the highest reported daily benefit amount. After coding variables of interest, as described in Section A1.2, we subset the LMS (but not the NHS) to the years 1985 to 2004. A1.2. Variable Definitions 1. (Un)employment: Individuals are classified as unemployed in year t if their employment status [erwstat] in that year is coded as 1 (long-term unemployment assistance post-2005 for adults able to work [VeP]), 11 (Unemployment benefits [ALG]), 12 (Unemployment assistance [ALHI]), or 13 (Maintenance allowance [UHG]). All other observations with a record in the NHS or LMS are classified as employed. 2. Regular employment: Following IAB conventions, regular employees are individuals whoseemployment status [erwstat] is coded as 101 (employees subject to social security with no special features), 140 (seamen), or 143 (maritime pilots). Non-regular employees include apprentices, workers in part-time pre-retirement employment, and working students. 3. Nursing homes: Defined as establishments classified under WZ73 industry codes 710 (private and for-profit), 711 (private, not-for-profit), and 712 (public ownership). 4. Healthcare sector: Employment spells are classified as part of the healthcare sector if they occur in establishments with one of the WZ73 industry codes 710, 711, 712, 780, 781, 782, 783, 784, 880, or if the spell corresponds to one of the KldB 1988 occupation codes [beruf] 841, 842, 844, 851, 852, 853, 854, 855, 856, 857, 861, 862. 5. Labor market experience: This variable measures the number of years with any employment during a rolling 15-year look-back window for the years 1990–2008. 6. Nursing home experience: Constructed analogously to labor market experience but specific to employment in nursing homes. 7. Healthcare experience: A 15-year rolling measure specific to employment in the healthcare sector, as defined above. 8. New hires: An individual is classified as a new hire in year t if they are employed in year t but not in year t − 1. 9. Wages: For employment spells, wages are recorded in the variable “Daily wage, daily benefit rate” [tentgelt]. This represents “the employee’s gross daily wages […] calculated from the fixed-period wages reported by the employer and the duration of the (unsplit) original notification period in calendar days” (Antoni et al., 2019). We top-code daily gross wages at the annual upper earnings limit set by statutory pension insurance regulations. IAB-Discussion Paper 03|2025 66 10. Unemployment benefits: For unemployment spells, [tentgelt] records daily benefit amounts. Prior to 1998, these values refer to working days, while from 1998 onward, they refer to calendar days. 11. Schooling: We construct an indicator variable for individuals who have passed the Abitur, coded as 1 if the schooling variable [schule] equals 8 or 9 (upper secondary school leaving certificate, equivalent to A-levels). A second dummy variable is coded as 1 for individuals with a bachelor’s degree, corresponding to values of the tertiary education variable [ausbildung] equal to 11 (degree from a university of applied sciences) or 12 (university degree). IAB-Discussion Paper 03|2025 67 A1.3. Appendix Figures and Tables Figure A1.: Population Share Age 65 and Over in West Germany .1 .12 .14 .16 .18 .2 1956 1966 1976 1986 1996 2006 2016 Notes: This figure plots the count of individuals age 65 and older as a share of the total population of West Germany by year. The time series includes Berlin and Bremen that are excluded in our primary analysis. The data source is the Human Mortality Database (2021). Source: Own calculations. IAB-Discussion Paper 03|2025 68 Figure A2.: Public Program Spending on Long-Term Care (in Billion EUR) Universal LTC Insurance Means-tested Hilfe zur Pflege 0 5 10 15 20 1991 1995 1999 2003 2007 Notes: This figure plots aggregate public spending on means-tested long-term care benefits, Hilfe zur Pflege, and universal long-term care (LTC) insurance. The spending statistics cover all of Germany (East and West). Universal LTC insurance started covering outpatient services in 1995 and inpatient services in 1996. These transition years are shaded in gray. The data source for expenditures on Hilfe zur Pflege is Statistisches Bundesamt (2013), Table D5. The source for expenditures on universal LTC insurance is Bundesministerium für Arbeit und Soziales (2011), Appendix 3. Source: Own calculations. IAB-Discussion Paper 03|2025 69 Figure A3.: Geographic Variation in Baseline Exposure and Local Income (a) Er and Average Gross Income (b) Er and Average Disposable Income SH HH NI NW HE RP BW SL OBay NBay OPf OFr MFr UFr Schw p-value: 0.920 .6 .65 .7 .75 .8 Baseline Exposure Measure (Er) 14,000 16,000 18,000 20,000 22,000 Income (Per Capita EUR, 1995) SH HH NI NW HE RP BW SL OBay NBay OPf OFr MFr UFr Schw p-value: 0.941 .6 .65 .7 .75 .8 Baseline Exposure Measure (Er) 13,000 14,000 15,000 16,000 17,000 18,000 Disposable Income (Per Capita EUR, 1995) Notes: This figure estimates the relationship between the baseline measure of exposure to LTC insurance rollout and measures of per capita in come in 1995, at the level of 15 exposure regions. The baseline exposure measure, denoted with Er throughout the text, is derived in Equation (1) and measures the share of individuals in need of long-term care, who did not have means-tested support for long-term care services in 1993, prior to the rollout of universal LTC insurance (mean of Er = 0.69). This measure varies across 15 different West German geographical regions, visualized in Figure A4. The correlation of the baseline exposure measure with income per capita, visualized in A3a, is statistically insignificant at p = 0.920, and the correlation with disposable income, displayed in in A3b, is statistically insignificant at p = 0.941. Income data have been obtained from (Statistische Ämter des Bundes und der Länder, 2021b). Source: Own calculations. IAB-Discussion Paper 03|2025 70 Figure A4.: Geographic Variation in Exposure Er (.76,.79] (.72,.76] (.68,.72] (.64,.68] [.59,.64] Former East Germany Exposure Er, % of potential demand without insurance Notes: This figure plots the value of exposure to the rollout of universal long-term care insurance, Er, for each county. Exposure varies at the level of 15 regions; counties in the same region are assigned to the same level of exposure. Er is defined as the count of individuals who did not have meanstested support for long-term care services (Hilfe zur Pflege) in 1993 divided by the count of people who received universal long-term care (LTC) benefits in 1999, adjusted by county-specific 1993-1999 population growth. See Equation (1) for the exact derivation of Er. Source: Own calculations. IAB-Discussion Paper 03|2025 71 Figure A5.: Alternative Ways of Computing Exposure to Insurance Rollout (a) Baseline and Alternative Er (I) (b) Baseline and Alternative Er (II) SH HH NI NW HE RP BW OBay NBay OPf OFr MFr UFr Schw SL Slope: 0.15 (0.02) R2: 0.86 .93 .94 .95 .96 .97 Alternative Exposure Measure (I) .6 .65 .7 .75 .8 Baseline Exposure Measure (Er) SH HH NI NW HE RP BW OBay NBay OPf OFr MFr UFr Schw SL Slope: 0.18 (0.04) R2: 0.61 .08 .09 .1 .11 .12 .13 .14 Alternative Exposure Measure (II) .6 .65 .7 .75 .8 Baseline Exposure Measure (Er) (c) Alternative Er (I) and (II) SH HH NI NW HE RP BW OBay NBay OPf OFr MFr UFr Schw SL Slope: 0.73 (0.27) R2: 0.24 .09 .1 .11 .12 .13 .14 Alternative Exposure Measure (II) .93 .94 .95 .96 .97 Alternative Exposure Measure (I) Notes: This panel displays bi-variate relationships between different ways of computing exposure to long-term care insurance rollout Er. The baseline Er is derived in Equation (1) and measures the share of individuals in need of long-term care (imputed from 1999 LTC insurance claims), who did not have means-tested support for long-term care services in 1993. The “Alternative Exposure Measure I” replaces denominator of exposure with the overall count of 65 and older population in 1993; i.e. it measures the share of 65 and older population that was receiving means-tested benefits. “Alternative Exposure Measure II” is computed as the share of individuals in 1993 who were not receiving means-tested benefits, but were in need of care (imputed from 1999 LTC insurance claims) among gr,1993,1999 ∗LT CClaimsr,1999−HzPr,1993 all individuals age 65 and older. In other words, Er = (mean of 65andOlderP opulationr,1993 Er = 0.103). Solid lines plot the line of best fit. The slope coefficient and the R-squared are reported next to the line. Source: Own calculations. IAB-Discussion Paper 03|2025 72 Figure A12.: Universal LTC Insurance and the Nursing Home Market, by Type of Employment (a) Full-Time NH Workers per 1,000 65+ Popl (b) Part-Time NH Workers per 1,000 65+ Popl High Exposure Low Exposure 10 15 20 25 30 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year High Exposure Low Exposure 0 5 10 15 20 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year (c) Full-Time NH Workers, Event Study (d) Part-Time NH Workers, Event Study 0 10 20 30 40 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year Point estimates -20 0 20 40 60 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 Year Point estimates Notes: Panels in the top row plot the number of regular full-time nursing home workers (A) and parttime nursing home workers (B) per 1,000 individuals age 65 and over, on average across counties for each year 1975 to 2008. The average in each year is computed separately for the group of West German counties with (region-level) exposure variable Er above (“high exposure”) and below (“low exposure”) the level of exposure of the median county. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for long-term care prior to the rollout of universal LTC insurance (mean and median of Er=0.69). All counties with exposure level at the median are assigned to the below median group. Both time-series are normalized to the aggregate mean of the y-axis variable across all counties in 1993. See Appendix A1.2 for the definition of nursing homes and “regular” workers. Er is derived in Equation 1 and its geographic variation is visualized in Figure A4. Panels in the bottom row display λt coefficients and 95% confidence intervals from estimating the specification in Equation 2 with the number of regular full-time nursing home workers (C) and part-time nursing home workers (D) per 1,000 individuals age 65 and above in a county as an outcome. λt are coefficients that multiply exposure variable Er; λt is normalized to zero in the pre-reform year t = 1993. More regression details are reported in columns (2) and (3) of Table 2. Regressions are estimated on the nursing home sample; see column (2) of Table 1 for summary statistics. Source: Own calculations. IAB-Discussion Paper 03|2025 79 Figure A13.: Event Study Results: Nursing Home Wages (a) New Hires (b) New Hires, Experience Controls -1 -.5 0 .5 1990 1994 1998 2002 2006 Year Point estimates -.6 -.4 -.2 0 .2 .4 1990 1994 1998 2002 2006 Year Point estimates (c) Incumbents (d) Incumbents, Individual FE -.4 -.3 -.2 -.1 0 .1 1976 1980 1984 1988 1992 1996 2000 2004 2008 Year Point estimates -.2 -.1 0 .1 1976 1980 1984 1988 1992 1996 2000 2004 2008 Year Point estimates Notes: This panel display λt coefficients and 95% confidence intervals from estimating the specification in Equation 2 with the log full-time wage of regular new (Panels A and B) or incumbent (panels C and D) nursing home workers. Wages are in constant 2020 EUR. See Appendix 2 for the definition of nursing homes and regular workers. Panel A13b displays results from a specification that controls for years of prior employment (overall and in nursing homes). Panel A13d displays results from a specification that includes individual fixed effects. Coefficients λt are normalized to zero in the pre-reform year t = 1993. λt multiply the exposure variable Er that takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for LTC prior to the rollout of universal LTC insurance (mean and median of Er=0.69). The geographic variation in Er is visualized in Figure A4. All specifications include county and year fixed effects. The results are summarizes in Table 3. Source: Own calculations. IAB-Discussion Paper 03|2025 80 Figure A14.: Introduction of Universal LTC Insurance and Supply of Nursing Home Care: SDID (a) Share Nursing Home Workers (b) Share Nursing Home Workers, Event Study High Exposure Low Exposure SDID ATT: 0.002 (0.0004) .01 .015 .02 .025 1985 1988 1991 1994 1997 2000 2003 Year 0 .001 .002 .003 .004 1985 1988 1991 1994 1997 2000 2003 Year Point estimates Notes: This exhibit replicates the analysis in Figure 1 using the synthetic difference-in-difference (SDID) specification (Arkhangelsky, Dmitry and Athey, Susan and Hirshberg, David A. and Imbens, Guido W. and Wager, Stefan, 2021). The outcome is the share of regular nursing home employees among all employed and unemployed individuals. We use a binary measure of exposure to the reform, which defines regions with an above-median exposure variable Er as treated units and counties at or below median exposure as control units. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for LTC prior to the rollout of universal LTC insurance (mean and median of Er =0.69) as a measure of local exposure to the reform. Er is derived in 1 and its geographic variation is visualized in Figure A4. Panel A displays treatment and matched control time series. Both time-series are normalized to the aggregate mean between control and synthetic treated units across all counties in 1993. Panel B shows the corresponding event study chart. We use b = 100 bootstrap draws to compute the 95% confidence intervals. Coefficients are normalized to zero in the pre-reform year t = 1993. Source: Own calculations. IAB-Discussion Paper 03|2025 81 Figure A15.: Introduction of Universal LTC Insurance and Overall Employment (a) Log Population Rejoining the Labor Force (b) Log Total Regular Wage Bill High Exposure Low Exposure SDID ATT: 0.019 (0.009) 5.3 5.4 5.5 5.6 5.7 5.8 1985 1988 1991 1994 1997 2000 2003 Year High Exposure Low Exposure SDID ATT: 0.022 (0.007) 12.7 12.8 12.9 13 1985 1988 1991 1994 1997 2000 2003 Year (c) Log Population Rejoining the Labor Force, Event Study (d) Log Total Regular Wage Bill, Event Study -.05 0 .05 .1 1985 1988 1991 1994 1997 2000 2003 Year Point estimates -.02 0 .02 .04 .06 1985 1988 1991 1994 1997 2000 2003 Year Point estimates Notes: Panels in the top row display treatment and matched control time series from synthetic difference-in-difference (SDID) specifications estimated at the county-year level. The outcome used in Panel (A) is the log count of individuals rejoining the labor force during a given index year; Panel (B) uses the log of the total daily wage bill pertaining to regular employment. Both time-series are normalized to the aggregate mean between control and synthetic treated units across all counties in 1993. The bottom panel shows event study charts for the same outcomes. We use b = 100 bootstrap draws to compute the 95% confidence intervals. Coefficients are normalized to zero in the pre-reform year t = 1993 We use a binary measure of exposure to the reform, which defines regions with an abovemedian exposure variable Er as treated units and counties at or below median exposure as control units. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for LTC prior to the rollout of universal LTC insurance (mean and median of Er =0.69) as a measure of local exposure to the reform. Er is derived in 1 and its geographic variation is visualized in Figure A4. Source: Own calculations. IAB-Discussion Paper 03|2025 82 Figure A16.: MVPF Estimates in Context Notes: This exhibit overlays our MVPF estimates (in red) over a simplified version of Figure IV.B in Hendren/Sprung-Keyser (2020). The original figure reports average MVPFs and 95% confidence intervals for several categories of public policies, plotted as a function of the average age of each policy’s beneficiaries. Source: Own calculations. IAB-Discussion Paper 03|2025 83 IAB-Discussion Paper 03|2025 84 IAB-Discussion Paper 03|2025 Table A1.: Event Study Results: Aggregate Response, Alternative Specifications Outcome (per 1,000 Age 65+ Population) At Exposure Region (r) Level S.E. Clustered at Region (r) Level County-Specific Time Trend Workers Full-Time Part-Time Firms Workers Full-Time Part-Time Firms Workers Full-Time Part-Time Firms (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Event Study Coefficients 1-Year Effect, λ1997 24.7 11.3 13.4 0.36 24.4 12.3 12.1 0.30 25.6 14.9 10.7 0.33 (8.07) (5.77) (2.88) (0.37) (7.04) (4.91) (2.56) (0.24) (6.85) (4.59) (3.16) (0.16) 3-Year Effect, λ1999 47.2 26.1 21.0 0.40 33.3 18.8 14.6 0.39 35.2 22.6 12.6 0.43 (18.9) (12.6) (6.96) (0.46) (10.3) (6.64) (4.56) (0.31) (8.14) (5.43) (3.97) (0.19) 5-Year Effect, λ2001 56.5 30.1 26.3 0.65 42.9 24.8 18.1 0.54 45.4 29.9 15.5 0.60 (19.5) (12.2) (8.47) (0.41) (12.0) (6.74) (6.22) (0.35) (9.17) (6.37) (4.64) (0.23) 10-Year Effect, λ2006 59.6 21.6 38.0 1.14 53.4 22.6 30.7 0.89 57.4 31.0 26.4 0.98 (14.4) (7.67) (9.43) (0.38) (15.3) (6.46) (10.0) (0.44) (12.0) (8.71) (6.29) (0.36) Level of Outcome in 1993 Mean 32.5 23.6 8.92 0.99 33.1 23.9 9.19 1.04 33.1 23.9 9.19 1.04 S.D. 5.86 4.25 2.08 0.27 13.8 10.3 4.57 0.47 13.8 10.3 4.57 0.47 Years 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 No. of Observations 510 510 510 510 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 Notes: This table summarizes the results of alternative specifications to the baseline results reported in 1, A12, and Table 2. Displayed are λt coefficients, obtained from estimating different versions (as specified in column titles) of the difference-in-differences specification in Equation (2). Outcome variables include the number of regular nursing home workers, part-time and full-time nursing home employment, and the number of nursing home establishments, per 1,000 individuals age 65 and older. The results in columns (1)-(8) are visualized in Figure A6 and the results in columns (9)-(12) in the top row of Figure A7. All specifications include county and year fixed effects. Columns (9)-(12) also include county-specific time trends. Standard errors clustered at the county level (columns 1-4, 9-12) and at the region r level (columns 5-8) are included in parentheses. Source: Own calculations. 85 Table A2.: Event Study Results: Aggregate Response, Alternative Specifications Outcome (per 1,000 Age 65+ Population) Controlling for Count of Elderly Controlling for Share of Elderly Workers Full-Time Part-Time Firms Workers Full-Time Part-Time Firms (1) (2) (3) (4) (5) (6) (7) (8) Event Study Coefficients 1-Year Effect, λ1997 21.1 11.1 10.1 0.25 23.5 12.2 11.3 0.30 (6.69) (4.20) (3.20) (0.15) (6.67) (4.20) (3.18) (0.14) 3-Year Effect, λ1999 28.2 16.8 11.4 0.29 31.4 18.5 12.9 0.36 (7.92) (4.83) (4.04) (0.17) (7.82) (4.80) (3.95) (0.17) 5-Year Effect, λ2001 34.1 21.5 12.7 0.39 38.8 24.1 14.6 0.50 (8.73) (5.37) (4.70) (0.18) (8.53) (5.32) (4.49) (0.18) 10-Year Effect, λ2006 34.7 15.6 19.1 0.56 48.8 21.9 26.8 0.84 (9.93) (5.90) (6.29) (0.25) (9.39) (5.63) (5.71) (0.23) Level of Outcome in 1993 Mean 33.1 23.9 9.19 1.04 33.1 23.9 9.19 1.04 S.D. 13.8 10.3 4.57 0.47 13.8 10.3 4.57 0.47 Years 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 No. of Observations 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 Notes: This table summarizes the results of alternative specifications to the baseline results reported in 1, A12, and Table 2. Displayed are λt coefficients, obtained from estimating different versions (as specified in column titles) of the difference-in-differences specification in Equation (2). Outcome variables include the number of regular nursing home workers, part-time and full-time nursing home employment, and the number of nursing home establishments, per 1,000 individuals age 65 and older. Columns (1)-(4) display results of specifications additionally controlling for the county-year-level count of individuals age 65 and above, and columns (5)-(8) control for the county-year-level population share of residents age 65 and above. The results in columns (1)-(4) are visualized in the bottom row of Figure A7 and the results in columns (5)-(8) in the top row of Figure A8. All specifications include county and year fixed effects. Standard errors clustered at the county level are reported in parentheses. Source: Own calculations. IAB-Discussion Paper 03|2025 86 IAB-Discussion Paper 03|2025 87 Table A3.: Event Study Results: Aggregate Response, Alternative Specifications Outcome (per 1,000 Age 65+ Population) Binary Exposure Measurea Alternative Exposure Measure (I)b Alternative Exposure Measure (II)c Workers Full-Time Part-Time Firms Workers Full-Time Part-Time Firms Workers Full-Time Part-Time Firms (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Event Study Coefficients 1-Year Effect, λ1997 3.69 2.07 1.62 0.03 112 45.3 66.8 0.70 102 67.2 35.0 2.14 (0.80) (0.50) (0.40) (0.02) (36.5) (22.8) (18.0) (0.89) (34.1) (22.0) (15.8) (0.76) 3-Year Effect, λ1999 4.88 2.81 2.06 0.05 143 69.2 74.2 1.12 156 103 52.6 2.32 (0.91) (0.56) (0.50) (0.02) (43.7) (26.7) (22.8) (1.00) (37.6) (24.5) (18.7) (0.89) 5-Year Effect, λ2001 5.87 3.45 2.41 0.07 188 101 87.1 1.81 199 125 74.7 3.22 (1.04) (0.63) (0.58) (0.02) (47.9) (29.3) (26.6) (1.03) (41.3) (27.3) (21.7) (0.97) 10-Year Effect, λ2006 6.94 3.27 3.67 0.10 242 93.7 148 4.02 246 115 131 4.25 (1.12) (0.69) (0.75) (0.03) (52.4) (31.3) (34.4) (1.39) (44.8) (28.6) (28.2) (1.11) Level of Outcome in 1993 Mean 33.1 23.9 9.19 1.04 33.1 23.9 9.19 1.04 33.1 23.9 9.19 1.04 S.D. 13.8 10.3 4.57 0.47 13.8 10.3 4.57 0.47 13.8 10.3 4.57 0.47 Years 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 1975 - 2008 No. of Observations 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 10,948 a The binary exposure measure takes on the value of one for counties above the median of the exposure measure Er derived in Equation (1), and zero otherwise. HzPr,1993 b The alternative exposure measure (I) is defined as Er = 100% − . 65andOlderP opulationr,1993 gr,1993,1999 ∗LT CClaimsr,1999−HzPr,1993 c The alternative exposure measure (II) is defined as Er = . 65andOlderP opulationr,1993 Notes: This table summarizes the results of alternative specifications to the baseline results reported in 1, A12, and Table 2. Displayed are λt coefficients, obtained from estimating different versions (as specified in column titles) of the difference-in-differences specification in Equation (2). Outcome variables include the number of regular nursing home workers, part-time and full-time nursing home employment, and the number of nursing home establishments, per 1,000 individuals age 65 and older. The results in columns (1)-(4) are visualized in the bottom row of Figure A8 and the results in columns (5)-(12) in Figure A9. All specifications include county and year fixed effects. Standard errors clustered at the county level are reported in parentheses. Source: Own calculations. Table A4.: Event Study Results: Count of New Nursing Home Hires By Origin Outcome (per 1,000 Age 65+ Population) Count New NH Hiresa Among New Nursing Home Hiresa, Count of Employed & in HC in t − 1 Employed & Not in HC in t − 1 Unemployed in t − 1 Temporarily Not in LF in t − 1 Never in LF before t (1) (2) (3) (4) (5) (6) Pooled Coefficients δ97−08 9.37 1.32 1.02 1.93 3.91 1.19 (1.82) (0.74) (0.53) (0.42) (0.52) (0.43) Event Study Coefficients 1-Year Effect, λ1997 8.38 3.10 -0.04 2.35 2.37 0.60 (4.05) (3.14) (0.75) (0.54) (0.53) (0.49) 3-Year Effect, λ1999 7.52 0.89 1.26 1.69 2.96 0.73 (1.93) (0.72) (0.56) (0.56) (0.66) (0.49) 5-Year Effect, λ2001 10.4 1.03 2.25 1.96 3.54 1.60 (2.27) (1.03) (0.67) (0.53) (0.68) (0.47) 10-Year Effect, λ2006 8.81 0.64 0.51 2.03 4.13 1.50 (2.15) (0.84) (0.60) (0.55) (0.68) (0.53) Level of Outcome in 1993 Mean 6.85 1.28 1.30 1.13 2.05 1.09 S.D. 2.94 0.84 0.78 0.64 0.93 0.63 Years 1977 - 2008 1977 - 2008 1977 - 2008 1977 - 2008 1977 - 2008 1977 - 2008 No. of Observations 10,304 10,304 10,304 10,304 10,304 10,304 a “New Hires” are individuals who were not employed in a nursing home in the year before each index year. Notes: The top panel displays pooled δ97−08 coefficients obtained from estimating the difference in differences specification in Equation (3) at the county-year level, using Er, derived in Equation (1), as the measure of a county’s exposure to the reform. Er takes values from 0 to 1 and measures the share of potential long-term care patients who did not have public assistance for LTC prior to the rollout of universal LTC insurance (mean and median of Er =0.69). The geographic variation in Er is visualized in Figure A4. The outcome variables are counts of new regular Nursing Home hires per 1,000 population age 65 and above by employment status before each index year, as specified in the column titles. The second panel displays λt coefficients of the event study in Equation (2). Coefficients were normalized to zero in year t = 1993. Results from equivalent specifications, using the share of hires from a respective origin rather than counts as outcomes, are presented in Table 5. All specifications include county and year fixed effects. Standard errors clustered at the county-level are reported in parentheses. Source: Own calculations. IAB-Discussion Paper 03|2025 88 Table A6.: Monthly Wages by Experience and Sector in 1999 Experience Population DataFP DataNFP DataOUT ModelFP ModelNFP ModelOUT (1) (2) (3) (4) (5) (6) (7) 0 29.6 2.29 2.46 4.10 2.34 2.42 4.09 1 0.81 2.58 2.75 3.27 2.62 3.10 3.24 2 0.47 2.76 3.07 3.15 2.74 3.20 3.32 3 0.39 2.86 3.41 3.41 2.86 3.30 3.40 4 0.31 3.10 3.62 3.55 3.03 3.40 3.52 5 0.26 3.17 3.77 3.61 3.10 3.51 3.55 6 0.24 3.24 3.86 3.72 3.13 3.61 3.56 7 0.22 3.38 3.91 3.74 3.24 3.71 3.64 8 0.19 3.45 3.95 3.82 3.30 3.81 3.66 9 0.17 3.54 4.03 3.86 3.46 3.91 3.78 10 0.14 3.54 4.06 3.88 3.46 4.01 3.75 11 0.12 3.69 4.19 3.96 3.60 4.11 3.85 12 0.11 3.69 4.27 4.05 3.65 4.21 3.87 13 0.10 3.88 4.35 4.13 3.81 4.31 3.98 14 0.08 3.86 4.40 4.16 3.86 4.41 4.00 15 0.07 4.08 4.51 4.21 4.03 4.51 4.12 16 0.07 3.97 4.49 4.31 4.00 4.61 4.08 17 0.06 4.16 4.43 4.28 4.15 4.71 4.18 18 0.41 4.34 4.56 4.49 4.39 4.81 4.36 Notes: This table shows full-time equivalent monthly earnings in 1,000 EUR by years of health care experience and labor market in the post-reform years (1999). Column (1) denotes the (working age) population in millions. Columns (2)-(4) denote full-time equivalent monthly wages based on the Labor Market Sample for for-profit nursing homes (column 2), not-for-profit nursing homes (column 3), and the outside sector (column 4). Columns (5)-(7) present corresponding estimates based on the quantitative model. Source: Own calculations. Next we examine changes in unemployment and gains in labor force participation relative to the estimated changes in nursing home employment. We consider the most conservative estimates supported by the confidence intervals in column (1) of Table 6. We consider a 0.00534 − 2 × 0.00177 = 0.00299 p.p. decrease in the unemployment share. We benchmark this to the relative gain in nursing home employment. Using our partial equilibrium estimates from Table 2, we estimate an increase of 38,979 workers. Divided by the 1993 labor force of 22.7 million, we estimate an increase of 0.00172 p.p. This suggests that the drop int UI exceeds the gains in nursing home employment by 0.00299/0.00172=1.74, see the sixth row. Our estimated model can only reconcile a smaller reduction in unemployment. Likewise, we construct conservative increase in labor force participation based on the estimates in column 3 of Table 2. Dividing the gain by the gain in nursing home employment, we find a ratio of 2.94, see the seventh row. We also ensure that the job filling and vacancy filling rates are bounded between 0 and 1, see Figure A17. IAB-Discussion Paper 03|2025 95 Table A7.: Employment Share by Experience and Sector in 1999 Experience Population DataFP DataNFP DataOUT DataOOL ModelFP ModelNFP ModelOUT ModelOOL (1) (2) (3) (4) (5) (6) (7) (8) (9) 0 29.6 0.0004 0.0007 0.53 0.41 0.0004 0.0007 0.54 0.41 1 0.81 0.01 0.02 0.39 0.49 0.01 0.02 0.39 0.50 2 0.47 0.02 0.03 0.42 0.45 0.02 0.03 0.42 0.47 3 0.39 0.02 0.04 0.43 0.44 0.02 0.04 0.44 0.44 4 0.31 0.03 0.05 0.44 0.43 0.03 0.05 0.43 0.42 5 0.26 0.03 0.05 0.45 0.43 0.03 0.05 0.44 0.41 6 0.24 0.03 0.06 0.45 0.42 0.03 0.06 0.43 0.41 7 0.22 0.03 0.06 0.47 0.40 0.03 0.06 0.46 0.38 8 0.19 0.03 0.07 0.48 0.39 0.03 0.07 0.46 0.38 9 0.17 0.03 0.07 0.49 0.38 0.03 0.07 0.48 0.35 10 0.14 0.03 0.07 0.48 0.39 0.03 0.07 0.47 0.36 11 0.12 0.03 0.07 0.51 0.37 0.03 0.07 0.49 0.34 12 0.11 0.02 0.07 0.52 0.36 0.02 0.07 0.51 0.33 13 0.10 0.03 0.06 0.55 0.34 0.03 0.06 0.54 0.31 14 0.08 0.02 0.07 0.57 0.31 0.02 0.08 0.56 0.29 15 0.07 0.02 0.07 0.58 0.30 0.02 0.07 0.57 0.27 16 0.07 0.03 0.07 0.59 0.29 0.03 0.07 0.59 0.27 17 0.06 0.03 0.08 0.60 0.26 0.03 0.08 0.60 0.25 18 0.41 0.02 0.07 0.67 0.21 0.02 0.07 0.67 0.21 Notes: This table shows full-time equivalent employment shares by years of health care experience and labor market in the post-reform years (1999). Column (1) denotes the (working age) population in millions. Columns (2)-(5) denote employment shares based on the Labor Market Sample for for-profit nursing homes (column 2), not-for-profit nursing homes (column 3), the outside sector (column 4), and the share that is out of the labor force (column 5). Columns (6)-(9) present corresponding estimates based on the quantitative model. Source: Own calculations. A2.3. Parameter Estimates In this section, we present further details on the calibrated and the estimated parameters. Calibrated parameters: Table A9 summarizes the calibrated parameters. First, we calibrate the fraction of patients insured via Hilfe zur Pflege in the post-reform to 31%. Second, we set the price subsidy to the average inpatient benefits, weighted by the care level distribution of LTC beneficiaries. We calculate a price subsidy of 1,200 EUR per month. Third, we calculate use the average tax rate and social security contributions of singles without children in 1999. We use the tax rate of 18.6% and social security contribution of 20.6% paid by employee and also by the employer.52 This implies a wedge between total costs to the employer and the net takehome pay for workers of in 1999 of 18.6%+2×20.6% τ = = 50%. 1+20.6% 52 See . IAB-Discussion Paper 03|2025 96 Table A8.: Other Moments Moment Data Model (1) (2) Monthly NH Price Post in 1k Euro 2.03 2.00 Monthly NH Revenues Post in bn Euro 1.14 1.46 % Change in NH Wage -0.12 0.02 NH Employment Post/ NH Employment Pre 1.92 1.92 % Change in Experience among NH Workers -0.30 -0.13 Drop in UI Recipients/ Change in NH Employment 1.74 0.19 Change in LF/ Change in NH Employment 2.94 3.04 Notes: This table presents the model fit on various other targeted moments. Column (1) denotes the data estimate and column (2) presents the model-simulated counterpart. The first two rows present average nursing home prices and overall nursing home revenues in 1999. The third row considers the reform-induced relative change in full-time nursing home wages. The fourth row considers the reform-induced relative change in nursing home employment. The fifth row considers the reform-induced relative change in the average number of years of health care experience among nursing home workers. The sixth row considers the reform-induced relative change in the the number of individuals claiming unemployment benefits, relative to the reform-induced change in nursing home employment. The seventh row considers the reform-induced relative change in labor force participation, relative to the reform-induced change in nursing home employment. More details are provided in the text. Source: Own calculations. Finally, we set the unemployment benefits in 1999 to 775 EUR per month. Table A9.: Calibrated Parameters Moment Data (1) Share Hilfe-zur Pflege Post 0.31 LTC Price Subsidy in 1,000 Euro per month: s 1.21 % Tax and Social Security Contributions: τ 0.50 Unemployment benefits in 1,000 Euro per month: b 0.78 Notes: This table presents calibrated parameters. The first row presents the fraction of nursing home patients insured via Hilfe zur Pflege. The second row presents the price subsidy for inpatient nursing home care in the post-reform period. The third row presents income taxes and social security contributions as a share of the total labor costs to the employer. The last row presents the unemployment benefits per month in 1,000 EUR. Source: Own calculations. Parameter Estimates: Next, we turn to the estimated parameters. We present the estimated λ parameters governing the matching technology and the vacancy posting costs c, by years of health care experience and sector, in columns (1)-(6) in Table A10. The λ parameters are IAB-Discussion Paper 03|2025 97 Figure A17.: Job Finding and Vacancy Filling Probability (a) Pre-Reform Vacancy Filling Rate in Nursing (b) Pre-Reform Job Finding Probability in Nursing Homes Homes 0 .2 .4 .6 .8 1 0 3 6 9 12 15 18 Years of Healthcare Experience For-Profit Not-For-Profit Outside (c) Post-Reform Vacancy Filling Rate in Nursing Homes 0 .2 .4 .6 .8 1 0 3 6 9 12 15 18 Years of Healthcare Experience For-Profit Not-For-Profit Outside (d) Post-Reform Job Finding Probability in Nursing Homes .2 .4 .6 .8 1 0 3 6 9 12 15 18 Years of Healthcare Experience For-Profit Not-For-Profit Outside .5 .6 .7 .8 .9 1 0 3 6 9 12 15 18 Years of Healthcare Experience For-Profit Not-For-Profit Outside Notes: This figure presents the estimated vacancy and job finding probability by sector and years of health care experience. For-profit nursing homes and not-for-profit nursing homes are denoted by ‘For-Profit’ and ‘Not-for-Profit’, respectively. Figures A17a and A17c display the equilibrium vaφ cancy filling rate, η(q ), for the pre- and the post-reform period. Figures A17b and A17d display the jt φ equilibrium job finding probabilities, µ(q ), for the pre- and the post-reform period. jt Source: Own calculations. presented for for-profit nursing homes (column 1), not-for-profit nursing homes (column 2) and firms in the outside sector (column 3). Likewise, the vacancy posting costs c are presented for for-profit nursing homes (column 4), not-for-profit nursing homes (column 5) and firms in the outside sector (column 6). The last two columns present the worker skills φ, as detailed in Equation (13), in nursing homes (column 7) and the outside sector (column 8). The remaining parameter are summarized in Table A11. The first three parameters govern the demand of the representative agent, see Equations 19-21. The next eight parameters govern the worker’s utility function, see Equations 15-18. Our empirical specification adds non-wage marginal costs to the firm profit Equation (5): φ φ φ π = η(q ) × ((Pj − mcj ) × φ − w ) − cj (φ). (A8) j j j IAB-Discussion Paper 03|2025 98 Table A10.: Matching Technology and Hiring Costs Parameters Experience LambdaFP LambdaNFP LambdaOut cFP cNFP cOut SkillsNH SkillOUT (1) (2) (3) (4) (5) (6) (7) (8) 0 0.50 0.82 0.90 0.16 0.06 0.16 1.65 2.19 1 0.96 0.81 0.81 0.34 0.21 0.34 1.72 1.85 2 0.83 0.91 0.79 0.35 0.12 0.35 1.79 1.90 3 0.81 0.84 0.79 0.35 0.16 0.35 1.86 1.95 4 0.82 0.77 0.84 0.32 0.20 0.32 1.93 1.99 5 0.80 0.77 0.82 0.36 0.23 0.36 2.00 2.04 6 0.78 0.87 0.80 0.43 0.14 0.43 2.07 2.09 7 0.80 0.88 0.79 0.44 0.14 0.44 2.14 2.13 8 0.79 0.89 0.77 0.49 0.14 0.49 2.21 2.18 9 0.83 0.76 0.79 0.46 0.30 0.46 2.28 2.22 10 0.84 0.76 0.77 0.56 0.34 0.56 2.35 2.27 11 0.86 0.76 0.78 0.55 0.37 0.55 2.42 2.32 12 0.87 0.78 0.75 0.61 0.37 0.61 2.49 2.36 13 0.88 0.82 0.76 0.58 0.37 0.58 2.56 2.41 14 0.88 0.79 0.73 0.64 0.39 0.64 2.63 2.46 15 0.92 0.79 0.75 0.60 0.43 0.60 2.70 2.50 16 0.87 0.85 0.70 0.73 0.39 0.73 2.77 2.55 17 0.89 0.84 0.72 0.71 0.41 0.71 2.85 2.60 18 0.95 0.87 0.72 0.61 0.37 0.61 2.92 2.64 Notes: This table presents estimated matching technology parameters, λ, hiring costs parameters, c, and worker skills, φ, by sector and years of health care experience. For-profit nursing homes are abbreviated by ‘FP’, not-for-profit nursing homes are abbreviated by ‘NFP’, and the outside sector is abbreviate by ‘Out’. We assume that workers skills are homogeneous between for-profit and not-for-profit nursing homes, conditional on experience, which we hence denote by ‘SkillsNH’, see Equation 13. More details are provided in the text. Source: Own calculations. and the next three rows of Table A11 present the corresponding marginal cost estimates. Building on the marginal cost estimates, we can express the wages paid by not-for-profit and public nursing homes, which are determined via collective bargaining. As discussed in the main text, we model wages, determined via collective bargaining, as a second order polynomial in the marginal revenue product net of non-wage marginal costs:  2 φ w = β0 + β1 × φ × (Pj − mc) + β2 × φ × (Pj − mc) . (A9) j The next three parameters in Table A11 show the corresponding parameter estimates. Finally, we present the estimated equilibrium output prices in the last four rows. ‘Post’ refers to the post-reform prices and ‘Pre’ refers to pre-reform environment without IAB-Discussion Paper 03|2025 99 universal LTC insurance, where LTC support is only provided via Hilfe zur Pflege. Prices in nursing homes and the outside sector are denoted by ‘NH’ and ‘OUT’, respectively. Table A11.: Parameter Estimates Parameter Estimate (1) Demand: αNH 2.71 Demand: αOUT 5.85 Demand: σ 1.19 Workers: Flow Payoff Unemployment (b + ξu) 1.01 Workers: Nesting Paramter γ 0.49 Workers: Standard Deviation Shocks ρ 0.03 Workers: ξF P -0.15 Workers: ξNF P -0.09 Workers: κ 7.64 Workers: κ1 -0.47 Workers: ν 1.14 Wage: β0 2.42 Wage: β1 0.71 Wage: β2 0 Firms: FP 0.002 Firms: NFP 0 Firms: OUT 0.05 NH Price Pre 1.87 NH Price Post 2.00 OUT Price Pre 2.11 OUT Price Post 2.14 Notes: This table presents the parameter estimates governing the quantitative model. The first three parameters αNH , αOut, and σ demand of the representative agent, see Equations 19-21. The next 8 parameters govern worker’s utility function, see Equations 15-18. b + ξu denotes the flowpayoff from unemployment, γ is a scaling parameter, ρ governs the correlation between utilities for working across long-term care sectors. ξF P and ξNF P denote compensating differentials for working in for-profit and not-for-profit nursing homes (relative to working in the outside sector). ν captures an extra compensating differential for workers with some health care experience to work in nursing homes, see Equation (15). The next three rows of Table resent estimates for the non-wage marginal costs by sector, mcj , see Equation (A8). They are followed by the parameters governing the collective bargaining wage profile described in Equation (A9). Finally, we present the estimated equilibrium output prices in the last four rows. ‘Post’ refers to the post-reform prices and ‘Pre’ refers to pre-reform environment without universal LTC insurance, where LTC support is only provided via Hilfe zur Pflege. Prices in nursing homes and the outside sector are denoted by ‘NH’ and ‘OUT’, respectively. Source: Own calculations. A2.4. Counterfactuals In this section, we discuss the counterfactual analysis in more detail. We start with a discussion of baseline outcomes before turning to a discussion of additional counterfactual exercises. IAB-Discussion Paper 03|2025 100 Baseline results: We start with a more detailed discussion of our baseline environment and present the model-simulated outcomes with and without universal LTC insurance in the first two columns of Table A12 (copied from Table 9). We estimate that increasing long-term care insurance coverage from 31% to 100% creates an additional 164,000 nursing home jobs. About 32,000 individuals are removed from unemployment and labor force participation increases by 497,000 individuals. Put together, overall employment increases by 529,000 individuals implying substantive employment gains in other sectors of the economy. This is because the reform boosted incomes among new nursing home hires who then spend their incomes on all goods in the economy generating positive spillover effects to other sectors. Figure A18a plots the reform-induced employment expansion in nursing homes by experience. We find that the increase in nursing home employment was concentrated among lower experienced workers, suggesting that the baseline result is partially driven by the considerable slack in the labor market among people who were previously out-of-the labor force and thus held little health care experience. Figures A18b-A18d illustrate that labor markets do not ‘just’ clear in wages and instead depend on job posting decisions by firms. Figure A18b illustrates pre-reform and post-reform nursing home employment but also counterfactual nursing home employment had wages not adjusted. We find that more than half of the increase in nursing home employment can be attributed to changes in vacancy postings. Unemployment would have decreased even further, had wages remained the same, Figure A18c. In contrast, a larger fraction of the increases in labor force participation can be attributed to changes in wages, see Figure A18d, suggesting that changes in vacancy posting played less of a role for the employment gains in other sectors. Marginal counterfactuals: We complement the discussion of the counterfactual analysis in the main text with added results for more marginal policy interventions. Table A12 is structured as Table 9 in the main text but considers additional interventions. Starting with the results for UI benefits, we now consider a 1% reduction in UI benefits (columns 4-6) in addition to the 100% reduction in UI benefits (columns 7-9) discussed in the main text. In regards to the role of income taxes, we now consider a 1% reduction in income taxes (columns 10-12) in addition to the 50% reduction in income taxes (columns 13-15) discussed in the main text. Finally, and in regards to the role of productivity shocks, we now consider a 10% IAB-Discussion Paper 03|2025 101 Figure A18.: Mechanisms (a) Nursing Home Employment (Millions) (b) Nursing Home Employment (Millions) 0 .01 .02 .03 .04 0 3 6 9 12 15 18 Years of Healthcare Experience 1993 (Pre-Reform) 1999 (Post-Reform) .15 .2 .25 .3 .35 Pre Post with Pre-Wages Post Environment (c) Unemployment (Millions) (d) Labor Force (Millions) 1.6 1.62 1.64 1.66 1.68 Pre Post with Pre-Wages Post Environment 19 19.2 19.4 19.6 19.8 Pre Post with Pre-Wages Post Environment Notes: Figures A18a -A18d provide more details on compositional changes in nursing home employment and the role of search and matching frictions, based on estimates from the quantitative model. Figure A18a presents full-time equivalent nursing home employment in millions with and without universal LTC insurance by years of healthcare experience. Figures A18b-A18d display full-time equivalent nursing home employment, unemployment, and labor force participation in millions for three economic environments. The first bar denotes equilibrium outcomes in the prereform period (without universal LTC insurance) and the third column shows equilibrium outcomes in the post-reform period (with universal LTC insurance). The center column presents outcomes using post-reform equilibrium queue lengths but pre-reform wages. Source: Own calculations. increase in the productivity of the outside sector (columns 16-18) in addition to the 50% increase in productivity (columns 19-21) discussed in the main text. The reform effects for the more marginal interventions here are directionally consistent with the findings presented in the main text, but smaller in absolute magnitude. The last three columns consider one additional counterfactual combining the role of productivity shocks and income taxes. Specifically, we lower income taxes by 50% and also increase the productivity in the outside sector by 50%. We now find almost no expansion in labor force following the reform suggesting that the reform is mostly relocating workers from the outside sector to the nursing home sector. This reduces allocative efficiency and in turn results in a welfare loss. IAB-Discussion Paper 03|2025 102 IAB-Discussion Paper 03|2025 103 Table A12.: Counterfactuals and Welfare Baseline 1% Reduction in UEB 100% Reduction in UEB 1% Reduction in Income Tax 50% Reduction in Income Tax 10% Increase in Productivity 50% Increase in Productivity 50% Incr. in Prod., 50% Decr. in Inc. Tax Insurance Diff Insurance Diff Insurance Diff Insurance Diff Insurance Diff Insurance Diff Insurance Diff Insurance Diff No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) Panel A: Employment and Earnings Nursing Home Employment (million FTE) 0.18 0.34 0.16 0.18 0.34 0.16 0.19 0.34 0.16 0.18 0.35 0.17 0.23 0.47 0.24 0.20 0.38 0.18 0.25 0.36 0.11 0.26 0.37 0.12 Total Labor Costs to Employers (billion EUR/month) 70.6 73.6 2.99 70.6 73.6 2.99 68.8 71.9 3.04 71.7 74.7 3.02 111.5 114.4 2.81 92.0 95.4 3.43 178.7 182.1 3.36 195.1 197.6 2.55 Unemployment (million FTE) 1.69 1.66 -0.03 1.69 1.66 -0.03 0.73 0.72 -0.006 1.70 1.67 -0.04 1.85 1.80 -0.05 1.60 1.52 -0.08 0.88 0.85 -0.04 0.62 0.59 -0.03 Labor Force (million FTE) 19.4 19.9 0.50 19.4 19.9 0.50 18.6 19.1 0.53 19.7 20.2 0.50 30.3 30.5 0.28 22.6 23.0 0.46 30.8 31.0 0.20 33.6 33.6 0.02 Panel B: Welfare (billion EUR/month) Consumer Welfare 81.0 80.3 -0.69 81.0 80.3 -0.69 81.7 81.0 -0.69 82.4 81.7 -0.70 130.2 129.1 -1.10 105.5 104.6 -0.89 203.0 201.3 -1.72 223.5 221.6 -1.90 Worker Welfare 148.7 149.2 0.59 148.6 149.2 0.59 147.7 148.3 0.60 149.0 149.6 0.60 172.1 173.5 1.35 152.9 153.6 0.70 175.4 176.5 1.12 222.7 224.6 1.82 Government Surplus 33.4 34.3 0.88 33.4 34.3 0.88 33.9 34.8 0.88 33.6 34.5 0.88 25.9 25.7 -0.25 44.2 45.2 1.06 87.8 88.8 0.99 47.5 47.5 -0.08 LTC Subsidy Spending -0.42 -1.05 -0.63 -0.42 -1.05 -0.63 -0.43 -1.06 -0.63 -0.42 -1.07 -0.64 -0.42 -1.41 -0.99 -0.44 -1.15 -0.71 -0.57 -1.28 -0.71 -0.59 -1.34 -0.74 UI Spending -1.31 -1.29 0.02 -1.31 -1.28 0.02 0 0 0 -1.32 -1.29 0.03 -1.44 -1.40 0.04 -1.24 -1.18 0.06 -0.68 -0.66 0.03 -0.48 -0.45 0.03 Tax Revenues 35.2 36.7 1.49 35.2 36.7 1.49 34.3 35.8 1.51 35.4 36.9 1.49 27.8 28.5 0.70 45.8 47.6 1.71 89.1 90.7 1.67 48.6 49.2 0.64 Total Welfare 263.1 263.9 0.78 263.1 263.9 0.78 263.3 264.1 0.78 265.0 265.8 0.78 328.2 328.2 -0.01 302.5 303.3 0.86 466.2 466.6 0.38 493.8 493.6 -0.16 Notes: This table summarizes the model-simulated effects of a universal LTC expansion to on a number of outcomes in different economic environments. For each environment, we present outcomes (i) without universal LTC insurance, where LTC support is provided via Hilfe zur Pflege only and (ii) with universal LTC insurance coverage captured by the subsidy s=1,200 EUR per month of the cost of nursing home care for individuals who would otherwise pay out-of-pocket. This exercise maintains support via Hilfe zur Pflege. We also present the difference between these outcomes. We consider eight environments. Columns (1)-(3) present results for our baseline economic environment, columns (4)-(6) revisit the results after reducing the UI benefits by 1% and columns (7)-(9) present results after reducing UI benefits from 775 EUR per month to zero. Columns (10)-(12) (13-15) revisit the results after reducing the income taxes by 1% (50%, from about 50 p.p. to 25 p.p.). Columns (16)-(18) (19-21) revisit the results after increasing the productivity in the outside sector for each skill segment, φout, (ehc) by 10% (50%). Finally, columns (22)-(24) revisit the results after increasing the productivity in the outside sector by 50% and reducing income taxes by 50%. For each environment, we present nursing home demand in 1m patients, nursing home employment in 1m full-time equivalent employees, total gross wage earnings in bn EUR per month, number of employment individuals in 1m, and labor force participation in 1m individuals. Next, we present the partial equilibrium welfare effect in bn per month. Turning to welfare in general equilibrium, we denote the equivalent variation (consumer surplus) in bn per month, holding income fixed at the pre-reform level. We then present the worker surplus in the labor market in bn per month. Turning the fiscal externalities, we present government LTC subsidy spending in bn per month, government spending on UI benefits in bn per month, income tax revenues in bn per month, and overall net spending government spending in bn per month. Finally, we combine the equivalent variation, the labor market surplus, and the net impact of government spending which comprises the net welfare effect in general equilibrium in bn per month. Source: Own calculations. Imprint IAB-Discussion Paper 03|2025 Publication Date March 7, 2025 Publisher Institute for Employment Research of the Federal Employment Agency Regensburger Straße 104 90478 Nürnberg Germany All rights reserved This publication is published under the following Creative Commons licence: Attribution – ShareAlike 4.0 International (CC BY-SA 4.0) https://creativecommons.org/licenses/by-sa/4.0/deed.de Download https://doku.iab.de/discussionpapers/2025/dp0325.pdf All publications in the series “IAB-Discusssion Paper” can be downloaded from https://iab.de/en/publications/iab-publications/iab-discussion-paper-en/ Website https://www.iab.de/en ISSN 2195-2663 DOI 10.48720/IAB.DP.2503 Corresponding author Jörg Heining Telefon ++49-(0)911-178-5392 E-Mail [email protected]