Institutional details and theory

Unemployment Benefits and Welfare Financial support for the unemployed is organized in a two-tier system in Germany. The first tier, i.e. unemployment benefits (UB), is a contribution-based insurance system with a replacement rate of 60 to 70, depending on whether individuals have children when they become unemployed or not. The maximum benefit duration is generally 12 months, except for individuals over 50 who may receive UB for up to two years, depending on their age when entering unemployment. The second tier, referred to as unemployment benefits II, is a tax-financed means-tested benefit system. To be eligible, individuals have to be able to work at least three hours a day and their household income must be below the legally defined minimum income threshold. As individuals in employment can still be eligible provided their household income is low enough, UB II is henceforth simply referred to as welfare. Welfare amounted to roughly EUR 360 per month in 2009 for a single adult. Over time, rates were gradually increased to EUR 420 per month in 2018. In addition to the welfare payment itself, recipients are eligible for reimbursement of the cost of accommodation and heating in a reasonable range as determined by the local job centre administering the welfare system.

In comparison to UB recipients, unemployed welfare recipients are much less likely to be re-integrated into the labour market. This is partly due to the fact that welfare recipients are more likely to be afflicted by employment impediments such as health issues or a lack of education than UB recipients. Moreover, unemployed welfare recipients often face a greater risk of poverty and social exclusion due to relatively low benefits compared to UB recipients. Re-employment wages tend to be rather low among previously unemployed welfare recipients and thus, they are more likely to be demoralised and hesitant to take up employment due to the risk of in-work-poverty. All in all, unemployed welfare recipients who find a job are likely to be affected by MW and thus, the focus of this study.

The German National Minimum Wage Following a long and intense debate, the federal government of Germany decided to introduce a statutory national MW. The associated MW bill was first drafted in April 2014 and finally passed by parliament in July 2014. On January 1st 2015, the MW came into effect with a legal minimum gross wage of EUR 8.5 per hour. At the time of the introduction, the German MW ranked among the highest in Europe when accounting for purchasing power (Caliendo et al., Reference Caliendo, Wittbrodt and Schro¨der2019), only surpassed by MW in France and Belgium. Up-ratings are decided upon by the minimum wage commission, whose members mostly comprise representatives of employer federations and unions. In 2017, the MW was increased to EUR 8.84 and further afterwards. As of January 2024, the MW stands at EUR 12.41, the highest MW in Europe when adjusting for purchasing power.

Two exceptions complicate the evaluation of the MW effects to some degree. First, initially there was an exception for long-term unemployed individuals, who mostly receive welfare benefits. As Vom Berge et al. (Reference Vom Berge, Klingert, Becker, Lenhart, Trenkle and Umkehrer2016) show, however, such exceptions were very rarely used by employers, potentially because they required additional paperwork that had to be filed to the Federal Employment Agency. Second, MW were allowed to be under-cut for some time if collectively-bargained agreements were below the MW. Both these cases might potentially lower the bite of the minimum wage below the level measured in the empirical analysis, which may lead to more conservative estimates of MW effects.

Some Theoretical Considerations In a perfectly competitive labour market, the wage equals the marginal product of labour in equilibrium and all unemployment is voluntary. In this – probably overly simplistic – textbook world, the introduction of a binding MW reduces the quantity of labour demanded and increases the quantity of labour supplied, leading to lower employment and higher unemployment among those whose productivity lies below the MW. In a non-competitive labour market, where employers have some market power, theoretical predictions on the employment effects of a minimum wage are less clear-cut (Manning, 2003). Such market power of firms can arise for multiple reasons, for example job differentiation akin to product differentiation (Bhaskar et al., Reference Bhaskar, Manning and To2002; Brueckner et al., Reference Brueckner, Thisse and Zenou2002), search frictions, i.e. the phenomenon that finding a new job takes time or money due to moving cost (Albrecht and Axell, Reference Albrecht and Axell1984; Burdett and Mortensen, Reference Burdett and Mortensen1998) or monopsony power due to a limited number of firms demanding certain types of labour (Manning, Reference Manning2003). In such a non-competitive labour market, a moderate MW can theoretically increase employment and lower unemployment (Flinn, Reference Flinn2006). This may be because minimum wages restrict the market power of firms in the labour market, which may lead to an increase in the quantity of labour demanded when increasing minimum wages from the non-competitive level towards the competitive wage (Boal and Ransom, Reference Boal and Ransom1997). Evidence that such labour market power of firms is relevant is growing (e.g, see Hirsch et al., Reference Hirsch, Jahn and Schnabel2018; Kölling, Reference Kölling2022). That being said, the size of potential employment gains is determined by the wage elasticity of labour supply, unless the MW is set too high, which leads to employment losses.

Thus, for unemployed welfare recipients considered here, the most important aspects are whether the MW is set above their potential productivity in the labour market and how their supply of labour responds to the MW. Economy-wide estimates of employment effects of the MW introduction suggest at most mildly negative employment effects (Caliendo et al., Reference Caliendo, Wittbrodt and Schro¨der2019), although these may be concentrated among particularly disadvantaged groups such as welfare recipients. Regarding labour supply, evidence from other countries points towards an increase in job search efforts among the unemployed (Adams et al., Reference Adams, Meer and Sloan2022). Whether or not this finding holds for case considered here, is uncertain, however. Hence, the overall sign and magnitude of effects of the MW on unemployed welfare recipients in Germany remain an empirical question.

To document the impact of minimum wages, both contributory employment, i.e. employment subject to social security contributions as well as minor employment, i.e. contribution-exempt jobs with monthly earnings below EUR 450 per month, need to be studied. This is because, on the one hand, MW increase the earnings potential in contributory employment much more than in minor employment as rising wages likely lead to a reduction in hours in minor employment in order to stay below the earnings threshold. On the other hand, minor employment is highly prevalent among welfare recipients and some minor employment may be converted into contributory employment (Caliendo et al., Reference Caliendo, Fedorets, Preuss, Schro¨der and Wittbrodt2018). Again, which of these employment dimensions is most affected is an empirical question. In the case of minor employment, however, one should expect negative effects, if any, both based on results of previous research as well as the possibility of converting minor to contributory employment.

Empirical strategy and data

To estimate the effect of the MW legislation on employment outcomes, this paper uses a Difference-in-Differences (DiD) approach based on quarterly panel data for a sample of unemployed welfare recipients as of December 31, 2013, i.e. way before the introduction of the minimum wage. As all unemployed welfare recipients are potentially affected by the MW legislation, a regional approach to generating a comparison group based on the bite of the minimum wage is used by following Card (Reference Card1992). Similar strategies have been used in the evaluation of the MW in Germany by others (Bonin et al., Reference Bonin, Isphording, Krause-Pilatus, Lichter, Pestel and Rinne2020; Caliendo et al., Reference Caliendo, Fedorets, Preuss, Schro¨der and Wittbrodt2018; Schmitz, Reference Schmitz2019).

To be precise, the sample is divided into two groups according to the median of the bite of the MW measured at the level of the 258 local labour markets identified by the Federal Office for Construction and Regional Planning. To measure the bite, this paper uses the so-called Kaitz Index defined as the MW divided by the median hourly wage in the labour market region, where the median hourly wage is computed using the Structure of Earnings Survey (SES) from 2014. The SES provides nationally representative earnings data from payroll accounts of roughly one million workers as of April 2014, i.e. around three months before the MW legislation was passed. These data are taken from Bonin et al. (Reference Bonin, Isphording, Krause-Pilatus, Lichter, Pestel and Rinne2020) and displayed in Figure 1. Figure 1 shows the spatial distribution of the Kaitz index in quartiles, where the two darker shaded areas represent the treatment group and the two lighter shaded areas represent the comparison group. Alternative bite measures are also employed as part of the sensitivity checks, results do not hinge on the use of the Kaitz-index based on the median.

Figure 1. Bite of the MW (Kaitz-Index)– Minimum wage in percent of median hourly wage.

Note: This figure shows the spatial distribution of the Kaitz index in quartiles across local labour markets, where the two darker shaded areas represent the treatment group and the two lighter shaded areas represent the comparison group.

Source: Bonin et al. (Reference Bonin, Isphording, Krause-Pilatus, Lichter, Pestel and Rinne2020).

To deliver unbiased estimates, the DiD approach requires that, in the absence of the MW introduction, both groups of unemployed welfare recipients would have followed the same employment trajectory (common-trends assumption, or CTA). As is common in the literature, the validity of the CTA is assessed by inspecting trends in outcomes before the MW legislation was passed. These analyses provide no indication regarding potential violations of the CTA, see section Descriptive Analysis for details. Nonetheless, the sensitivity of the results in this regard is assessed by allowing for differential trends based on a variety of additional personal and regional characteristics, leading to the same conclusions as the baseline empirical approach.Footnote ¹ Moreover, the bite needs to be measured accurately. If exceptions for the long-term unemployed and certain collectively-bargained sectoral wages distort the measurement of the bite, the DiD approach will deliver conservative estimates due to attenuation bias.

Estimation For our baseline estimation approach, a fixed-effects panel regression is used with standard errors clustered at individual levelFootnote ² . The corresponding regression model is

$${Y_{it}} = {\alpha _i} + \mathop \sum \limits_{t\; \ne \;Q42013} {\lambda _t} + \mathop \sum \limits_{t\; \ne \;Q42013} {\delta _t}\;{D_r}\;{\lambda _t} + \gamma '{X_r}{\lambda _t} + + {\varepsilon _{it}},$$

where α _i is the individual fixed-effect, λ _t are quarter dummies, D _r is an indicator taking on the value of one for individuals in a high bite local labour market region r, and zero otherwise. The vector X _r contains two indicators on the type of local labour market region, i.e. sparsely or moderately populated local labour markets, with (mostly) urban as the reference category. Hence, this approach already allows for differential trends regarding the type of local labour market. As the treatment indicator is interacted with quarter-dummies, δ _t gives the DiD estimate of the dynamic causal effect of the MW legislation relative to the 4th quarter of 2013. As the end of 2013 is used as a base period, this naturally allows for anticipation effects of the reform without risk of bias. This is important because even the drafted MW bill in April 2014 may already influence behaviour of our sample members.

Data and Sample As the data source, this paper used a two percent sample of the Integrated Employment Biographies (IEB) from the Statistics department of the Federal Employment Agency of Germany (Schneider, Reference Schneider2020). These data provide information on all spells in employment, registered unemployment, unemployment benefit and welfare benefit receipt as well as participation in active labour market policies, measured with daily precision. From these data, a sample of individuals that is registered as unemployed and received welfare as of December 31st 2013 is drawn. This yields a sample of 12,334 individuals in high-bite regions and 23,703 individuals in low-bite regions. For this sample, a quarterly panel is generated with outcomes being contributory employment, i.e. employment subject to social security contributions and minor employment, i.e. employment with earnings below the social security threshold of EUR 450.Footnote ³ Outcomes are measured five years prior and five years after sampling.

Table 1 shows some selected sample statistics for our treatment and comparison group. As can be seen, individuals living in high-bite regions are on average 39.7 years old and slightly older than individuals in low-bite regions with 39.4 years. No statistically significant differences can be found in terms of the share of women across high- and low-bite regions. Women make up 51 percent of both samples. Regarding the share of individuals without a professional education, high bite regions display a substantially lower share of individuals without a vocational or university degree. While the distribution of individual characteristics tends to be somewhat in favour of individuals’ labour market prospects in high-bite regions, indicators on regional labour market characteristics point in the opposite direction. The mean unemployment rate with 11 percent is considerably higher in high-bite regions than in regions with a relatively low bite with about 9.8 percent. Moreover, the vacancy-to-unemployment ratio is lower in high-bite regions than in low-bite regions. Thus, local macroeconomic conditions are more favourable for individuals in low-bite regions than for individuals in high-bite regions. Hence, overall, it is unclear how individuals in high-bite regions are selected compared to low-bite regions. For the analysis, this is secondary, however, as the CTA may hold even if treatment and comparison groups differ in terms of underlying characteristics. This is assessed in the next Section.

Table 1. Descriptive statistics

Note: This tables shows descriptive statistics on background characteristics of the treatment (high bite) and comparison (low bite) group. P-values stem from two-sample t-tests of equal means.

Table 2. Sensitivity – additional parametric trends specification

Note: This tables shows estimated static effects on the likelihood of being in contributory and minor employment after sampling. Baseline control variables include individual, time-period and time-period by local labour market type fixed effects. Additional trends are allowed the quadratic and differ between before and after sampling. Standard errors are clustered at the individual level.

Empirical analysis

This section presents results from our main empirical analysis. First, descriptive results are presented, mainly to convince the reader that the CTA is a plausible assumption in this setting. Second, dynamic effects of the MW legislation are presented. Third, effect heterogeneity is inspected and lastly, the sensitivity of our estimates is checked.

Descriptive analysis

Figure 2 display the share of individuals in contributory employment (panel A) and minor employment (panel B) for both treatment groups. For both outcomes, we see that prior to Q4 2013, trends are very similar. The only difference visible is a slightly larger seasonality in employment rates in high-bite labour markets. This is probably due to the fact that high- bite regions tend have more employment in agriculture, manufacturing and construction, which typically show larger seasonal variation than other sectors. Differences are, however, relatively small. After the sampling date, contributory employment rates develop similarly until they start diverging more and more after the introduction of the MW at the beginning of 2015 in favour of the high-bite group. For minor employment, no large differences emerge.

Figure 2. Trends in Employment Outcomes. (a). Contributory Employment. (b). Minor Employment.

Main analysis

Figure 3 shows dynamic effects of the MW legislation as estimated via the DiD strategy. Results clearly show that, apart from a single quarter in 2009, trends in contributory employment are not statistically significantly different between the treatment groups before sampling. This provides additional evidence on the validity of the CTA. After sampling, there is a marginally significant positive effect in the 2nd quarter of 2014, i.e. when the MW bill was passed, which hints towards small anticipation effects. However, the majority of effects kick in when the MW was actually introduced in the first quarter of 2015. At this point, effects amount to a little more than one percentage point. Two years after the MW introduction, the effect grows to about 2.5 percentage points. This trend continues until the last quarter of 2018, i.e. the end of our observation period, with effects amounting to roughly four percentage points. This is a sizeable effect given that roughly 25 percent in the comparison group are in contributory employment at this point.

Figure 3. Estimated Effects on Employment Outcomes. (a). Contributory Employment. (b). Minor Employment.

Hence, the MW legislation seems to have had a large positive and long-lasting impact on employment rates of unemployed welfare recipients. Thus, it may be that positive effects on labour supply of increased wages dominate a potentially negative effect of the MW on labour demand. Alternatively, labour demand may have actually increased due to the MW legislation by means of reducing labour market power of firms. At least two factors may explain the rather long-term effects of the MW introduction. First, the MW may have prevented scarring effects that occur during unemployment (Clark et al., Reference Clark, Georgellis and Sanfey2001), leading to better employability of unemployed welfare recipients even in the long run. Second, the MW legislation may have paved the way for unemployed welfare recipients to better quality jobs, potentially reducing the risk of often-occurring cycles of employment and re-lapse into unemployment.

Regarding minor employment, there are no significant differences in trends prior to sampling and there are no statistically significant effects of the MW introduction. This finding is not surprising given that the MW introduction did not increase the earnings potential of minor jobs as an hours reduction was common in those jobs in order to stay under the social security contribution threshold (Burauel et al., Reference Burauel, Caliendo, Grabka, Obst, Preuss and Schro¨der2020). The absence of negative effects on minor employment imply that a large-scale conversion of minor to contributory employment did not materialize in this population.

Overall, the results imply that employment of (previously) unemployed welfare recipients is quite reactive to changes in wages at the bottom on the wage distribution. From the literature, we know that among those who remain in employment, the MW legislation led to an increase in hourly wages of about 6.5 percent (Bachmann et al., Reference Bachmann, Boockmann, Gonschor, Kalweit, Klauser, Laub, Rulff and Vonnahme2022) until 2018 for those directly affected, while welfare benefits increased by 3.3 percent during the same frame. That is, over the four years studied, we obtain a net-of-benefits wage semi-elasticity of around 4/(6.5 − 3.3) ≈ 1.25. Hence, for every percentage point that hourly wages grow faster than welfare benefits, employment rates of (previously) unemployed welfare recipients rise by 1.25 percentage points.

Effect heterogeneity

This part explores whether there are substantial differences in the impact of MW on employment across sub-groups defined by standard background characteristics. To analyse such effect heterogeneity, the sample is split in two and effects are estimated separately for each sub-sample. To keep things simple, a static DiD regression to obtain average effects after the reform is used by interacting the high-exposure-dummy with a post-reform dummy. The sample is split according to gender, age at sampling (over 40 or younger), having a professional degree or not as well as above or below median regional labour market tightness. Results are shown in Figure 4.

Figure 4. Heterogeneous Effects. (a). Contributory Employment. (b). Minor Employment.

Source: Integrated Employment Biographies, own calculations.

Note: Minimum wage bill passed by parliament in July 2014. Minimum wage intro- duced at EUR 8.50 on January 1 2015. Minimum wage increased to EUR 8.84 on January 1 2017.

As panel A shows, effects on contributory employment are very similar across sub- groups according to personal characteristics. The only significant difference in effects is found for regarding regional labour market tightness. The results show that when vacancies are relatively abundant in comparison to the number of unemployed individuals in the region, effects are even more positive in comparison to regions where vacancies are relatively scarce. This suggests an interaction effect of local labour market conditions and the effects of minimum wages. It appears that a (potential) reduction in labour demand due to MW is less likely to have a strong impact on jobseekers if vacancies are abundant, making it more likely that positive supply-side effects dominate the resulting impact estimate. This also implies that effects of MW on employment prospects of unemployed welfare recipients may change over the business cycle. Effects on minor employment are statistically insignificant across all sub-groups (see panel B).

Sensitivity analysis

Having presented the main empirical results of this study, the sensitivity of the results is assessed. First, we test for the sensitivity of the estimates by allowing for differential trends between individuals and regions based on a variety of observed characteristics. Second, we re-estimate effects using other bite measures common in the literature.

Allow for additional trend-differences This paragraph inspects the robustness of our results by allowing for differential trends across individuals or regions according to some observed background variables. Column one of Table 2 shows our baseline estimates of static effects of the MW in Germany. These amount to a two percentage point increase in contributory employment rates and a null effect on minor employment. In columns two to four, additional flexible parametric trends are gradually added to the baseline specification. These trends are allowed to be quadratic and different prior to and after sampling.

Column two adds such trends for professional education groups, genders and age groups. Column three adds trends for regions with a high share of individuals employed in agriculture, manufacturing or construction. Column four additionally allows for differen-tial trends for regions with a high welfare recipiency rate, (long-term) unemployment rate as well as vacancy-to-unemployment ratio. Expanding the specification from our base- line approach to column four leads to somewhat of an attenuation of estimated effects for contributory employment, essentially nothing changes regarding minor employment. The most expansive specification in column four suggests an increase in the likelihood of being in contributory employment after sampling by 1.5 percentage points. Hence, our conclusion on positive employment effects in this regard is robust to such changes in the specification.

Different bite measures Our baseline empirical approach generates treatment and comparison groups based on the Kaitz index using the local median hourly wages. Resulting static estimates are shown in the first column of Table 3. Next, we inspect whether results are robust to the use of alternative treatment indicators based on other bite measures. In the second column, treatment and comparison groups are generated via a median split according to the Kaitz index based on the mean, i.e. the ratio of the minimum wage of EUR 8.5 introduced in 2015 to the mean hourly wage in the respective labour market region. In the third column, treatment and comparison groups are based on a median split according to the mean gap in hourly wages among individuals in the respective labour market region. In the fourth column, the treatment group consists of individuals in regional labour markets that have a share of workers paid below the minimum wage above the median of the distribution.

Table 3. Sensitivity – bite measures treatment/comparison group using different bite measures

Note: This tables shows estimated static effects on the likelihood of being in contributory and minor employment after sampling using treatment indicators based on different bite measures. Control variables include individual, time-period and time-period by local labour market type fixed effects as in the baseline empirical approach using the Kaitz index based on the median of the wage distribution. Standard errors are clustered at the individual level.

As one can see, there is some variation in effect sizes across specifications based on the different bite measures. However, estimates effects on contributory employment are consistently significantly positive. Estimates on minor employment switch signs and become significant at the 5 percent level when using the share of employees paid below the MW to generate the treatment and comparison group. As effects on minor employment are not robust, this should not be taken as definitive evidence on negative effects in this regard. Moreover, the variation in effect sizes does not come as a surprise: when deciding which bite measure to employ for our main approach, we compared average bite measures across the different definitions of treatment and comparison groups and found that the distinction based on the Kaitz index in combination with the median tends to yield the largest group differences in exposure to the MW reform in terms of all bite measures. Hence, we should expect to see the pattern in estimates for the different bite measures as groups based on the mean wage gap and the share of individuals paid below the minimum wage are more similar in terms of exposure than groups generated based on the Kaitz index.

Moreover, defining treatment and comparison group based on the Kaitz index implicitly does so using a ranking in median or mean hourly wages across local labour markets. This ranking is much more likely to stay constant over time, and thus it is more likely to be applicable to the up-rating of the minimum wage in 2017, than a ranking based on the wage gap or the share affected (Caliendo et al., Reference Caliendo, Pestel and Olthaus2025). The latter two might be quite different for the up-rating, which ought to lead to smaller estimates in absolute terms compared to estimates based on the Kaitz index as well.

All in all, the differences in effect sizes across different bite measures used should not be over-interpreted. However, it is reassuring the find that our conclusions on positive employment effects are robust to the use of different bite measures to generate the treatment variable.