The Effects of Employers’ Disability and Unemployment Insurance Costs on Benefit Inflows

In Finland, large ˝rms are partially liable for the costs of disability and un-employment bene˝ts paid to their former workers. To estimate the e˙ects of such costs, we exploit a reform that extended this cost-sharing to cover a new group of blue-collar workers. We show that experience rating in disability insurance reduces in˛ows to sickness and disability bene˝ts and increases participation in vocational rehabilitation programs, whereas employers' unemployment insurance costs reduce excess layo˙s of older workers who are eligible for extended unemployment bene˝ts until retirement age. We ˝nd no evidence of spillover e˙ects: employers' costs in one bene˝t type do not a˙ect in˛ows to other types of the bene˝ts.


Introduction
Disability and unemployment bene˝t schemes are among the largest social security programs in many countries. In 2019, public spending on sickness and disability bene˝ts amounted to 2% and unemployment expenditures 0.6% of GDP in the OECD countries.
A large part of disability insurance (DI) and unemployment insurance (UI) costs is ˝nanced through employer contributions. In most countries, ˝rms pay ˛at-rate DI and UI premiums, which may depend on the age structure of the workforce or industry but not directly on the costs of the bene˝ts received by the ˝rm's own employees. This can lead to ine°ciently high bene˝t in˛ows because ˝rms do not take into account the societal costs of new bene˝t claims when making decisions about layo˙s and investments in workplace health and safety. The government may aim to mitigate this incentive problem through experience rating or coinsurance. With experience rating, the ˝rm's insurance premium rate increases with the amount of bene˝ts awarded to its employees in some past reference period. With coinsurance, the ˝rm has to pay a lump sum contribution to the insurance provider at the time when the bene˝t is awarded to its employee. In the both cases, the ˝rm is directly liable for part of the bene˝t costs of its own employees. The objective is to make employers internalize the costs to society of new bene˝t claims.
Coinsurance and experience rating systems are controversial because the administrative costs can be high while the evidence of their e˙ectiveness in reducing bene˝t costs is limited and mixed. Such systems may also distort hiring decisions and have undesired spillover e˙ects on the receipt of other types of bene˝ts. Perhaps this uncertainty explains why only a few countries have adopted an experience rating or coinsurance system to ˝nance their UI and DI expenditures. The United States is the only country where ˝rms pay experience-rated UI premiums, whereas it is only in Finland and the Netherlands that ˝rms pay experience-rated DI premiums. In some European countries, including Finland, France, Italy and Germany, ˝rms must pay lump sum contributions for layo˙s in certain cases (Fath and Clemens 2005). Evidence on the e˙ects of experience rating and coinsurance should not only be of interest for policy makers in these countries but also for those in other countries who are considering alternative reforms to cut the DI and UI expenditures. Autor (2011) and Burkhauser and Daly (2011), for example, have suggested that the US Social Security DI program should be ˝nanced by an experience-rated payroll tax.
In this paper, using data from Finland, we show that experience rating and coinsurance systems can be useful tools to reduce disability and unemployment in˛ows. In Finland, large ˝rms pay DI premiums that depend on the past disability bene˝t claims of their former employees. Large ˝rms are also partially liable for the costs of extended unemployment bene˝ts paid to their former employees. Extended bene˝ts are available for older unemployed individuals who exhaust their regular UI bene˝ts. When a bene˝t extension is awarded, the former employer is obligated to pay a share of the estimated costs of the extended bene˝ts over the coming years as a lump sum payment to the Unemployment Insurance Fund. Small ˝rms are not subject to experience rating in DI, nor coinsurance in UI.
In 2007, a pension reform in Finland extended the coverage of experience rating and coinsurance to a new group of blue-collar workers in construction, dock work, agriculture and forestry. As a consequence, large employers became directly liable for the DI and UI costs of these workers, whereas smaller employers were not a˙ected. This worker group is of particular interest due to its exceptionally high disability and unemployment risks.
We exploit the 2007 reform to identify the causal e˙ects of experience rating in DI and coinsurance in UI using a di˙erence-in-di˙erences (DID) approach.
We ˝nd that experience rating in DI reduces in˛ows to sickness and disability bene˝ts, and increases participation in vocational rehabilitation programs. Like previous studies (e.g. Hakola and Uusitalo 2005, Kyyrä and Wilke 2007, and Kyyrä and Pesola 2020, we ˝nd a higher layo˙ risk for older workers who are eligible for extended UI bene˝ts. The layo˙ risk of this group is particularly high in large ˝rms, suggesting that large employers may exploit the extended UI bene˝t scheme as a soft way to reduce employment when downsizing. We show that coinsurance in UI alleviates this problem by reducing excess layo˙s of the oldest workers. We ˝nd no evidence of adverse spillover e˙ects: experience rating in DI does not increase layo˙s, nor does coinsurance in UI increase the in˛ow to disability bene˝ts. We also show that the desired e˙ects of experience rating on bene˝t in˛ows are not driven by employers becoming more selective when hiring new workers.
Our study contributes to the literature on the e˙ects of employer liabilities in DI and UI. A handful of papers have analyzed the Dutch DI system, providing somewhat mixed results.
1 The ˝ndings of Koning (2009) andvan Sonsbeek andGradus (2013) suggest that the introduction of experience rating in the late 1990s reduced the in˛ow to disability bene˝ts in the Netherlands. de Groot and Koning (2016) show that the removal of experience rating from small ˝rms increased the disability in˛ow by 7% over the years 20032004, but they ˝nd no e˙ect when experience rating was re-introduced in 2008. The asymmetry in the e˙ects may be due to other changes in the sickness and disability schemes between the two reforms. 2 1 Several authors have analyzed experience rating in US and Canadian Workers' Compensation programs, which cover medical costs and provide cash payments for those who are injured at work (e.g. Ruser, 1985Ruser, , 1991Bruce and Atkins, 1993;Krueger, 1990;Thomason and Pozzebon, 2002). These studies present evidence that experience rating reduces on-the-job injuries and the duration of injury spells. Because employers have less control over disability outcomes than workplace injuries, we should expect smaller e˙ects for experience rating in DI. 2 In 2005, the period of sickness bene˝ts was extended from one to two years. This extended employers' 3 In Finland, a coinsurance system in DI existed for decades until it was replaced with an experience rating system in 2006.
3 Korkeamäki and Kyyrä (2012) show that coinsurance in DI reduces the in˛ow to sickness bene˝ts and further transitions from sickness bene˝ts to disability bene˝ts. Hawkins and Simola (2018) ˝nd that higher coinsurance discourages ˝rms from hiring workers with a high disability risk. Kyyrä and Paukkeri (2018) analyze the e˙ects of experience rating in DI among worker groups that were not a˙ected by the 2007 pension reform. These worker groups (a majority of all private-sector workers) have lower disability and unemployment risks than the group studied in this paper. Kyyrä and Paukkeri (2018) ˝nd that experience rating has little or no e˙ect on in˛ows to sickness and disability bene˝ts, suggesting that replacing coinsurance with the experience rating system in 2006 may have weakened the incentive e˙ect and possibly increased overall DI costs.
4 We complement this research by showing that experience rating reduces sickness and disability bene˝t in˛ows in blue-collar occupations with exceptionally high disability rates. It should be stressed that Kyyrä and Paukkeri (2018) were unable to rule out moderate e˙ects of experience rating due to the limited statistical power of their regression-kink setting.
5 In this paper, we provide more convincing causal evidence based on the di˙erence-in-di˙erences approach.
Very little is known about the e˙ects of the experience rating of UI premiums in the US labor market due to the lack of appropriate data and exogenous variation in the UI system. Anderson and Meyer (2000) and Woodbury et al. (2004) provide some empirical evidence that a higher degree of experience rating in UI reduces layo˙s, while earlier studies such as Topel (1983) and Card and Levine (1994) ˝nd that experience rating reduces temporary layo˙s. A recent study by Johnston (2021) ˝nds that an increase in the ˝rm's UI premium due to layo˙s in the past three years reduces hiring and employment but does not a˙ect layo˙s. We complement this literature by providing causal evidence that coinsurance in UI reduces excess layo˙s of older workers compared to the counterfactual case of ˛at-rate UI premiums. 6 liabilities, given that Dutch ˝rms are responsible for the costs of sickness bene˝ts for their employees. One year later, separate disability bene˝ts were introduced for persons who are permanently fully disabled and for persons who are only partially and/or temporarily disabled. Since this reform, experience-rated DI premiums have been applied only to the disability bene˝t costs of the latter group. 3 Finland adopted International Financial Reporting standards in 2006, as required by the European Union. Because coinsurance in DI was seen problematic in the context of the new reporting environment, it was abolished and replaced with experience-rated DI premiums. The experience rating system was designed to closely mimic the incentive structure of the coinsurance system in terms of the allocation of costs between employers. 4 Their estimates imply no signi˝cant e˙ect on average in the population of all workers. However, they ˝nd a signi˝cant e˙ect on the disability in˛ow for men under the age of 50, but not for other groups. 5 Their identi˝cation strategy was based on kinks in the formula that determines the degree of experience rating in DI premiums as a function of ˝rm size. Since these kinks are quite small, the regression kink design approach produced rather imprecise estimates. 6 Prior to the current UI system in Finland, the long-term unemployed were entitled to an unemploy-4 The remainder of the paper proceeds as follows. Section 2 describes the institutional setting. Section 3 discusses how the 2007 reform changed employer liabilities for certain worker groups and employers. Section 4 describes the data used and Section 5 reports sample statistics. Section 6 discusses the econometric methods and reports the results.
2 Institutional framework 2.1 Bene˝t schemes A worker who cannot work due to a medical condition is entitled to compensation for income loss. For the ˝rst weeks (typically one to three months depending on the collective agreement), the applicant is fully compensated by the employer, after which he or she may receive a sickness bene˝t up to about one year. In case of prolonged disability, the individual may qualify either for a disability pension or rehabilitation bene˝t. 7 If it is unlikely that the applicant will be able to return to work, a disability pension is awarded for an inde˝nite period of time. Otherwise, a rehabilitation bene˝t is awarded for a speci˝c time period. The rehabilitation bene˝t is e˙ectively a ˝xed-term disability pension. Despite its name, receipt of it is not conditional on participation in rehabilitation programs. There is no automatic retesting of disability status, except for new periods of rehabilitation bene˝ts.
A worker who is at risk of becoming disabled may participate in a vocational rehabilitation program (work trial, job coaching, apprenticeship training or vocational retraining program). The health problem that hinders the worker from performing his or her current job must be veri˝ed by a doctor. For the duration of vocational rehabilitation, the worker receives either pay from their employer or a rehabilitation allowance.
The entitlement period of regular UI bene˝ts is about two years (100 weeks) for insured workers with su°cient work history. However, UI recipients who are old enough on the day when their regular bene˝ts expire are entitled to extended bene˝ts. The age threshold for the bene˝t extension is 57, 59 or 60, depending on the recipient's birth year. In addition, long-term unemployed workers born before 1950 were entitled to an unemployment pension at age 60. The unemployment pension scheme was abolished in ment pension at age 60, and the former employer had to pay a lump sum coinsurance contribution to the pension provider when the unemployment pension was awarded. Hakola and Uusitalo (2005) show that higher unemployment pension costs reduced layo˙s of older workers. The unemployment pension scheme was abolished in 2005, after which large ˝rms have been liable for the costs of extended UI bene˝ts paid to their former employees. 7 For both bene˝t types, full bene˝t is conditional on a loss of working capacity of at least 60% and partial bene˝t on a loss of 40% to 59%. unemployment pension (at age 60 for those born before 1950) or old-age pension (at age 63 for later cohorts) begins. The combination of regular and extended UI bene˝ts plus possibly an unemployment pension is often dubbed the unemployment tunnel (UT).
The UT scheme provides a seamless ˛ow of income until retirement for older unemployed individuals, and it has acted as a popular early retirement pathway.

Experience rating in disability insurance
A ˝rm's DI premium rate is de˝ned as a weighted sum of a ˛at-rate base premium rate and an experience-rated premium rate. The relative weight of the experience-rated rate depends on the ˝rm's size: it increases linearly with the payroll (measured two years earlier) from 0 to 1 for ˝rms with a payroll between 1.5 and 24 million in 2004 euros.
Thus, small ˝rms with a payroll of 1.5 million euros or less are not subject to experience rating but only pay ˛at-rate premiums, whereas larger ˝rms are partially responsible for the costs of disability pensions received by their former employees. The experience-rated premium rate is determined by the estimated costs of new disability pensions awarded to the ˝rm's employees two and three years ago. The cost of a disability pension equals the expected amount of disability pension bene˝ts until the statutory old-age retirement age.
When a ˝rm's disability pension costs are equal to the average costs of ˝rms with the same age structure, the experience-rated premium equals the base rate. At maximum, a ˝rm can earn a 90% discount on the base premium or be obligated to pay a 450% surcharge on top of the base premium.
To the extent that experience rating in DI encourages ˝rms to invest in disabilitypreventive measures, the experience rating system should reduce in˛ows to vocational rehabilitation programs, sickness bene˝ts, rehabilitation bene˝ts, and disability pensions.
However, conditional on the onset of a medical condition reducing work capacity, experience rating may increase in˛ows to other programs than disability pensions. This is because the experience-rated DI premium depends only on disability pension costs, not on rehabilitation bene˝t or rehabilitation allowance costs. This may induce large ˝rms to encourage their workers to apply for a rehabilitation bene˝t or participate in vocational rehabilitation programs rather than to apply for a disability pension. If a worker's health status improves to the extent that he or she can return to work, their employer will avoid DI costs altogether. But even if the worker eventually ends up on a disability pension, their employer will gain from postponing that event, because the older the worker is at the time when the disability pension is awarded, the smaller the e˙ect is on the ˝rm's DI premium rate. Therefore, the overall e˙ect of experience rating in DI on other health-related programs than disability pension is a priori ambiguous.
6 Firms cannot in general escape DI costs by laying o˙ workers whose health deteriorates. This is because a new disability pension claim a˙ects the DI premium rates of the ˝rms in which the claimant worked one and two calendar years prior to the year when the underlying medical condition was diagnosed. On the contrary, if workers with health issues are more keen to apply for a disability pension when unemployed than when employed, experience rating in DI may discourage layo˙s. For the oldest workers who are eligible for the UT scheme, this incentive may be weaker because laying o˙ such workers is more acceptable from society's viewpoint and because these workers may be less likely to claim a disability pension when unemployed as their income level is secured until old-age retirement due to the extension of the UI entitlement period. As such, experience rating in DI may reduce layo˙s of workers with poor health but less so for UT-eligible workers.

Coinsurance in unemployment insurance
Regular unemployment bene˝ts are ˝nanced by tax revenues and ˛at-rate insurance premiums, but large ˝rms are partially liable for the costs of extended UI bene˝ts paid to their former employees. When a bene˝t extension is awarded to a worker born in 1950 or later, the former employer, depending on its size, may have to pay a share of the estimated costs of the extended bene˝ts as a lump sum payment to the Unemployment Insurance Fund. The cost of extended bene˝ts is calculated assuming the worker will collect them until the statutory old-age retirement age irrespective of actual bene˝t duration. These costs act like severance pay and are likely to discourage layo˙s of older workers who are eligible for the UT scheme.
The share of the extended UI bene˝t costs the employer is directly liable for increases linearly from 0% to 80% as a function of the ˝rm's payroll in the year preceding the layo˙.
The payroll thresholds for the minimum and maximum cost shares are the same as those that determine the degree of experience rating in DI, i.e. 1.5 and 24 million in 2004 euros.
It follows that any two ˝rms that di˙er in the degree of experience rating in DI also di˙er in the degree they are responsible for the extended UI bene˝t costs. Nevertheless, the e˙ects of DI and UI costs can be separated because DI costs can arise from workers of all ages, but UI costs can arise only from the oldest workers born in 1950 or later.

The 2007 pension reform
Until 2006, DI and UI costs only applied to workers who were insured under the Employees' Pension Act (TEL). The vast majority of all workers in the private section were covered by TEL. The only notable exception were blue-collar workers in construction, dock work, agriculture and forestry, who were insured under the Temporary Employee's Pensions pension Acts. In construction ˝rms, for example, white-collar workers were insured under TEL and blue-collar workers under LEL. Firms of all sizes only paid ˛at-rate DI and UI premiums for their LEL-insured workers. However, large ˝rms were partially liable for the costs of disability and unemployment-related bene˝ts received by their TEL-insured workers through experienced-rated DI premiums and coinsurance payments.
In 2007, all private-sector pension acts were uni˝ed into a new Employees Pensions Act (TyEL). 9 This reform did not a˙ect the eligibility criteria or bene˝t levels of disabilityrelated bene˝ts, i.e. the content of DI from a worker's perspective, but it extended experience rating in DI to also cover blue-collar groups that were insured under LEL. As a consequence, large ˝rms with a payroll exceeding 1.5 million in 2004 euros began to pay experienced-rated DI premiums also for their former LEL-insured workers, whereas smaller ˝rms continued to pay ˛at-rate premiums as before. Due to a transitional provision, experience rating came into e˙ect gradually over time: only disability pensions awarded to former LEL-insured workers whose medical condition was diagnosed in 2008 (partial e˙ect) or later (full e˙ect) have in˛uenced the employer's DI premium rate. Another consequence of the 2007 reform was that large ˝rms became responsible for the costs of extended UI bene˝ts received by their former LEL-insured workers born in 1950 or later.
These coinsurance UI costs applied to workers who were laid o˙ in 2007 or later.
If small and large ˝rms and their employees are su°ciently similar, we can exploit the 2007 reform to estimate the causal e˙ects of experience rating in DI and coinsurance in UI using a di˙erence-in-di˙erences (DID) approach. Namely, we can use small ˝rms and their employees as a comparison group, because the reform did not a˙ected the DI and UI costs of small ˝rms. Figure 1 illustrates eligibility for the UT scheme and the timing and coverage of employer costs in UI and DI in the former LEL industries. The shaded areas in the graph show the timing and coverage of UI and DI costs for large ˝rms: experience rating in DI applies to all workers from 2008 onward (shaded area with dashed border), whereas coinsurance in UI only applies to workers born in 1950 or later at the ages when they are eligible for the UT scheme (shaded area with solid border). Since these areas only partially overlap, the e˙ects of UI and DI costs are identi˝ed separately. 10 8 The name of the pension act is a bit misleading: it was applied to all blue-collar worker groups in certain industries and certain occupation groups independently of whether they had a ˝xed-term or open-ended employment contract. Another exception were artists, journalists and persons working for households, who were insured under the Pension Act for Performing Artists and Certain Groups of Employees. We do not consider that group in our analysis. 9 The government submitted the law change to parliament in April 2005 and it was accepted in May 2006.
10 Eligibility for the UT scheme (bold numbers) is de˝ned as the lowest age at which a laid o˙ worker is able to collect UI bene˝ts, including regular bene˝ts for the ˝rst 100 weeks and extended bene˝ts Calendar year Year of birth thereafter, until entry into an unemployment pension (pre-1950 cohorts) or old-age pension (later cohorts). This age threshold varies between 55 and 58 depending on the birth year, which allows us to distinguish the e˙ect of UT eligibility from the age e˙ect. In addition, we can identify the e˙ect of replacing the unemployment pension with extra weeks of extended UI bene˝ts by comparing workers born before and after 1950 at ages 60+. We are not really interested in this change but may want to control for its e˙ect in the econometric analysis.  While the in˛ow rate to vocational rehabilitation programs is very small, it almost triples in the large ˝rms and roughly doubles in the small ˝rms. In the large ˝rms, the dis- Notes: A ˝rm is classi˝ed as large if its average payroll was more than 1.5 million in 2004 euros in the pre-reform years. Degree of experience rating refers to the relative weight of the experience-rated premium rate in the ˝rm's DI premium rate formula, and it is computed using the average payroll in the pre-reform years. It increases linearly from 0 to 1 from a payroll of 1.5 million to a payroll of 24 million euros, and it is 0 for smaller ˝rms and 1 for larger ˝rms. Column 7 reports di˙erence-in-di˙erences estimates for worker and ˝rm characteristics, i.e. the di˙erence between columns 6 and 3. The associated p-value in column 8 is obtained by regressing an outcome variable (at worker of ˝rm level) on the time period, large-˝rm dummy and their interaction, using a variance-covariance matrix robust to individualand ˝rm-level clustering (Panels A and C) or ˝rm-level clustering (Panel B).
ability pension in˛ow rate decreases by 0.2 percentage points between the two periods, compared to a signi˝cantly smaller decline of 0.06 percentage points for the small ˝rms.
The numbers reported in columns 7 and 8 provide evidence that the 2007 reform reduced the sickness bene˝t in˛ow by 0.84 percentage points and the disability pension in˛ow by 0.14 percentage points and increased participation in vocational rehabilitation programs by 0.16 percentage points. In the subsequent analysis, we examine the time pattern of these e˙ects, assess the validity of the underlying parallel trend assumption, and estimate distinct e˙ects for experience rating in DI and coinsurance in UI.
It should be stressed that the worker group studied is characterized by relatively high disability and layo˙ risks. In the Supplementary Appendix, we compare the disability and unemployment outcomes of TEL-and LEL-insured workers in 19992006. The LELinsured workers have about a 30% higher risk of becoming disability bene˝t recipients and a 70% higher layo˙ probability than the TEL-insured workers. Among the LELinsured workers, the most common causes of disability are musculoskeletal and connective tissue diseases (45%), while mental health and behavioral disorders are relatively rare (9% compared to 21% among the TEL-insured workers).
6 Results Figure 2, panels A to E depict the average in˛ow rates to di˙erent bene˝ts by year and ˝rm size. Panel F shows the share of employees who remain employed with the same employer they worked for in the past two years. To assess the statistical signi˝cance of the changes between the ˝rm-size groups, we estimate linear probability models of the following form:  Year Small firms Large firms 12 Except for the layo˙ rate in 2003, the di˙erences in the in˛ow rates between the small and large ˝rms remain roughly constant over the years 19992005, as the 95% con˝dence intervals contain 0 each year in Figure 3. Thus, the parallel trend assumption of the DID approach seems to be valid.

The e˙ect of the 2007 reform
In the large ˝rms, the in˛ow rates to sickness bene˝ts and disability pensions (

Separating the e˙ects of UI and DI costs
The DID estimates in Table 1 and Figure 3 describe the joint e˙ect of employers' DI and UI costs. These costs may have opposite e˙ects on some outcomes, potentially overriding each other within the UT-eligible worker group. To separate the two distinct e˙ects, we estimate the following linear probability model: where U T it a dummy variable for workers who are eligible for the UT scheme in year t, Born 1950+ i is a dummy variable for those born in 1950 or later, and X ijt is a vector of controls for age, gender, education level, marital status, tenure, living region, having a young child, home ownership and industry.

13
Time variation in the outcomes across the employees of small ˝rms identi˝es the year ˝xed e˙ects ϕ s . As illustrated in Figure 1, the age threshold of the UT scheme is 55 for individuals born before 1950, 57 for those born in 195154, and 58 for later cohorts. This variation across birth cohorts identi˝es the e˙ect of UT eligibility, α 1 , in the presence of age and year e˙ects. The interaction term Large j · U T it allows the e˙ect of UT eligibility · Born 1950+ to be di˙erent in large and small ˝rms. The interaction term U T it i equals 1 for those who qualify for extended UI bene˝ts after the entitlement period of regular UI bene˝ts but not for the unemployment pension, and 0 for others (see Figure 1). Thus, δ 1 is the e˙ect of replacing the unemployment pension with extra weeks of UI bene˝ts from age 60 upwards, and it is identi˝ed by comparing the UT-eligible employees of small ˝rms who were born before 1950 and those born in 1950 or later at the same age.
The parameters of primary interest are λ, the e˙ect of employer's UI costs through the lump sum coinsurance payments, and η s , the e˙ect of employer's DI costs through the experience rating system. These e˙ects are identi˝ed from di˙erences in the outcomes over time between workers employed by large and small ˝rms. The two e˙ects are separately identi˝ed, because the UI costs can only arise from workers born in 1950 or later who enter unemployment after reaching the age threshold of the UT scheme, while the DI costs can arise from all disability pension recipients irrespective of their birth year and age (see Figure 1). The e˙ect of DI costs is allowed to vary over time, starting from 2006, i.e. one year before the reform came into e˙ect, in order to allow for possible anticipatory behavior and to capture the gradual implementation of experience rating. However, to increase statistical power, we assume that the e˙ect is constant in the years when the experience rating system was fully implemented and, therefore, we impose the restriction γ 2009 = γ 2010 = . . . = γ 2014 . As we did not ˝nd signi˝cant di˙erences between large and small ˝rms in 19992005 except for a di˙erence in the layo˙ rate in 2003 (see Figure 3), we assume that the reform had no e˙ect prior to 2006.
The estimated e˙ects of the key variables, in percentage points, are reported in Table   2. From column 1 we see that the likelihood of starting a sickness bene˝t period is 1.6 percentage points higher for individuals who are employed in large ˝rms than for those working in small ˝rms. The in˛ow rates to other bene˝ts do not di˙er notably between large and small ˝rms. As seen in column 5, eligibility for the UT scheme increases the layo˙ risk by 0.8 percentage points in small ˝rms and by 2.3 percentage points (= 0.843+ 1.438) in large ˝rms. It seems that ˝rms often target layo˙s at older workers whose income level is secured up to old-age retirement, and this practice is particularly common among large ˝rms. Replacing the unemployment pension with extra weeks of UI bene˝ts has no e˙ect on the layo˙ risk, as the coe°cient on the interaction term between UT eligibility and those born in 1950 or later does not di˙er from 0 at the conventional risk levels.
Coinsurance UI costs reduce the layo˙ rate of older workers by 1.4 percentage points (column 5). This e˙ect o˙sets the di˙erence of 1.4 percentage points in the e˙ect of UT eligibility between large and small ˝rms, so that the layo˙ risk does not di˙er between UTeligible workers in small and large ˝rms when the large ˝rms are subject to coinsurance UI costs. It is noteworthy that UI costs do not increase the take-up of other bene˝ts, given that the coe°cient on the triple interaction term is negative, yet insigni˝cant, for all health-related bene˝ts. These results imply that coinsurance in UI eliminates excess layo˙s of older workers in large ˝rms that would otherwise occur due to the UT scheme, without having adverse spillover e˙ects on in˛ows to other types of bene˝ts.
The introduction of experience rating in 2007 started to reduce sickness bene˝t claims from the ˝rst year onward: in large ˝rms, the in˛ow to sickness bene˝ts dropped by 0.71.0 percentage points in 20072014 compared to small ˝rms (column 1). This suggests that experience rating induces employers to invest in preventive measures that reduce the onset   Notes: All models include year dummies and controls for age, gender, education, marital status, tenure, region of residence, having a child under age 7, home ownership and industry. Standard errors clustered at individual and ˝rm levels in parenthesis. Signi˝cance levels: *** 1%, ** 5% and * 10%.
of health problems at the workplace and, therefore, the in˛ow to sickness bene˝ts. The average duration of sickness bene˝ts increased by around 20% in large ˝rms compared to small ˝rms after the reform (see Supplementary Appendix). That is, large employers have mainly succeeded in eliminating short sickness spells, which explains much of the strong e˙ect on the sickness bene˝t in˛ow in the years 20072014.
In 20092014, the in˛ow to vocational rehabilitation programs increased by 0.2 percentage points more in large ˝rms than in small ˝rms (column 2), which is a large relative e˙ect given that the average annual in˛ow is only 0.2%. This ˝nding suggests that experience rating induces employers to direct workers who have di°culties in performing their current job tasks due to health problems to vocational rehabilitation programs. These programs may reduce disability pension claims by enabling workers with disabilities to switch to more suitable jobs that allow them to continue working.
We ˝nd no evidence of experience rating in DI having signi˝cant e˙ects on the in˛ow to rehabilitation bene˝ts (column 3) or on layo˙s ( this estimate is less likely to be a˙ected by anticipatory behavior, it seems evident that experience rating reduces disability pension claims. It is worth noting that the e˙ects of experience rating on health-related outcomes come without adverse spillover e˙ects on the layo˙ risk. There is also a positive e˙ect on the probability of remaining employed with the same ˝rm (column 6).
In summary, the results imply that the introduction of experience rating in DI in 2007 reduced the in˛ow to sickness and disability pension bene˝ts, and increased participation in vocational rehabilitation programs. In addition, the adoption of coinsurance in UI reduced the layo˙ rate of the oldest workers with a particularly high risk of unemployment due to their eligibility for the UT scheme. There is no evidence of adverse spillover e˙ects: experience rating in DI does not increase layo˙s, nor does coinsurance in UI increase the in˛ow to health-related programs.

Selective hiring
The aim of experience rating in DI is to encourage employers to invest in preventive measures that reduce the onset of health problems, as well as to accommodate workers with a medical condition that reduces their working capacity, so that they can continue working.
However, experience rating may also discourage employers from hiring workers with high disability risks. That way, contrary to its primary goal, the experience rating system may worsen the employment prospects of job seekers with poor health. To minimize the risk of future disability pension claims, hiring ˝rms may require physical examinations of job applicants or favor younger job applicants over older ones. Selective hiring can reduce health problems in large ˝rms subject to experience rating and, therefore, explain part of the negative e˙ects of experience rating in DI on the in˛ow rates to health-related programs. Table 3 reports the sample means of background characteristics for recently hired employees. Unfortunately, our data does not include good measures of health. The best measure is the time spent on sickness bene˝ts before recruitment. Average days on sickness bene˝ts among recently hired workers remain stable over time in large ˝rms but increase slightly in small ˝rms. The di˙erence between these changes in the large and small ˝rms is small, 0.9 days, but statistically signi˝cant at the 5% level (columns 7 and 8), suggesting that large employers may have been more cautious in hiring after the reform. The average age of recently hired workers increases by 0.2 years in large ˝rms, which does not di˙er signi˝cantly from the drop of 0.8 years in small ˝rms. Likewise, the share of new workers aged 50 and older increases more in large ˝rms than in small ˝rms, but the di˙erence is not statistically signi˝cant. Also in most other cases, the composition of recently hired employees changes in the same direction by roughly the same amount in the large and small ˝rms. One exception is the share of those with tertiary education, which increases in large ˝rms but decreases in small ˝rms. This group is very small, accounting for less than 7% of new hires as well as of all employees in both ˝rm-size groups, so it cannot have a notable impact on the outcomes. In the light of these di˙erences, our estimates of the in˛ow e˙ects are unlikely to be signi˝cantly a˙ected by selective hiring after the reform, although the small di˙erence in the sickness bene˝t days between large and small ˝rms is a little worrisome.
As an another check, we re-estimate linear probability models using data only on workers who have been continuously working for the same employer for at least eight years. In that sample, all workers in all years were hired before the 2007 reform, so that possible changes in hiring practices due to the reform cannot a˙ect the results. The results for the e˙ects of DI and UI costs from this subsample are shown in Table 4. For all outcomes the e˙ects of DI costs are similar to those reported in  Tables 2 and 4. 20 Notes: The ˝gures are for recently hired individuals who remained in the ˝rm for at least for two years. A ˝rm is classi˝ed as large if its average payroll was more than 1.5 million in 2004 euros in the prereform years. Column 7 reports the di˙erence-in-di˙erence estimates for worker characteristics, i.e. the di˙erence between columns 6 and 3. The associated p-value in column 8 is obtained by regressing an outcome variable on the time period, the large-˝rm dummy and their interaction, using the variancecovariance matrix robust to ˝rm-level clustering.

Robustness and heterogeneity
We checked the robustness of our results for the e˙ects of experience rating in DI and coinsurance in UI in various ways. We discuss these ˝ndings brie˛y here, but the detailed results are available in the Supplementary Appendix. All the results in Section 6.2 are based on the models that include controls for background characteristics. Our results are not sensitive with respect to the exclusion of these control variables (apart from age) or to the inclusion of ˝rm ˝xed e˙ects. In Section 5, we found some di˙erential changes in the composition of the workforce of large and small ˝rms over time. In particular, the average education level has been increasing slightly faster in large ˝rms than in small ˝rms. This may a˙ect the results as education may be correlated with health risks.
Therefore, we estimated models separately for workers with upper secondary or higher education and for those with lower education. The results are rather similar for both education groups, implying that our main results are not driven by di˙erential education trends. In addition, we estimated models separately for construction workers and for other workers. The results are otherwise similar but experience rating in DI increases the in˛ow to vocational rehabilitation programs only among construction workers.
We treated large ˝rms as a single group, ignoring the fact that the degree of experience 21 7 Notes: All models include year dummies and controls for age, gender, education, marital status, tenure, region of residence, having a child under age 7, home ownership and industry. The standard errors clustered at the ˝rm level in parenthesis. Signi˝cance levels: *** 1%, ** 5% and * 10%.
rating and coinsurance varies across these ˝rms. Firms whose payroll is less than 24 million but more than 1.5 million in 2004 euros can be only marginally subject to the DI and UI costs of their employees. If we exclude these ˝rms and only include the 21 largest ˝rms that are fully covered by experience rating in DI and coinsurance in UI in the group of large ˝rms, we will ˝nd somewhat stronger e˙ects for both types of costs than those reported in Table 2. In that respect, the reported results are conservative.

Concluding remarks
The aim of experience rating in DI is to promote preventive health and safety practices at the workplaces, and to encourage employers to accommodate their employees with health-related work limitations. This study provides evidence of such e˙ects. We ˝nd declines in the sickness bene˝t and disability pension in˛ow rates in large ˝rms when they became exposed to experience rating. We also ˝nd that the employees of these ˝rms are more likely to participate in vocational rehabilitation programs. We show that none of these e˙ects is driven by selective hiring. We ˝nd no evidence of negative spillover e˙ects on layo˙s and, therefore, experience rating also reduces the overall exit rate from employment.

22
Our ˝ndings are at odds with the previous results of Kyyrä and Paukkeri (2018), who did not ˝nd signi˝cant e˙ects for other worker groups in Finland. In this study, we analyze a rather special group of blue-collar workers in construction, dock work, agriculture and forestry that is characterized by a higher-than-average disability risk. For this group musculoskeletal and connective tissue diseases are the most common causes of disability, while mental health and behavioral disorders play a relatively small role. Thus, our results suggest that experience rating in DI is a useful tool in reducing the disability in˛ow at least for worker groups with relatively high disability risks due to musculoskeletal and connective tissue diseases.
Moreover, our results con˝rm the previous ˝nding that the layo˙ risk is much higher for older workers who can qualify for extended UI bene˝ts. We ˝nd that large ˝rms in particular exploit this scheme as a soft way to reduce their workforce. We show that coinsurance UI costs mitigate this problem by reducing the excess layo˙s of older employees, without having negative spillover e˙ects on the in˛ow rates to other bene˝ts.