The Empirics of Hidden Labor Force Dynamics in Germany

Sandro Provenzano 1
  • 1 Department of Economics Vienna University of Economics and Business, Vienna, Austria
Sandro Provenzano
  • Corresponding author
  • Department of Economics Vienna University of Economics and Business, Vienna, Austria
  • Email
  • Search for other articles:
  • degruyter.comGoogle Scholar

Abstract

The unemployment rate is the core indicator when researchers and policy-makers assess the level of underemployment in an economy. However, accumulating evidence suggests that the unemployment rate is biased and underestimates the true level of underemployment. Closing this gap is especially important because the distortion systematically changes along the business cycle and affects the various subgroups of the population differently. Neglecting these effects when setting up policies might flaw its effectiveness and result in unexpected outcomes. Although the existence of these effects is widely agreed upon only little is known about the magnitude of these effects across various subgroups. Using a highly disaggregated dataset from Germany, this study examines the dynamics in labor force participation that go beyond the unemployment rate. Ample evidence is found that the discouraged and the added worker effect significantly affect particular subgroups in the German labor market. In addition, the discouraged and the added worker effect are generally found to be very symmetric in economic upturns and downturns. Moreover, the labor market reforms in Germany between 2003 and 2005 are found to have reduced the discouraged worker effect on average by 25%, leaving the added worker effect unchanged.

1 Theoretical background

1.1 Introduction

The unemployment rate (UR) is commonly known and a core indicator in the field of economics and social science. It is often used by researchers and policy-makers when assessing the level of underemployment in an economy. However, its accuracy has often been subject to controversial discussions. Its critics argue that the unemployment rate does neither reflect the whole picture of movements in the labor market nor indicates the true level of underemployment. On the one hand, there is the dispute about how to define unemployment for example whether people in temporal or permanent labor market measures should be counted as employed or unemployed. On the other hand, the labor force is subject to cyclical variations that additionally distort the information content of the UR. The latter will be addressed in this study.

The cyclical variation in the labor force participation rate (LFPR) has been subject to economic research for a long time and can be traced back to Long (1953). The positive relationship between the LFPR and the business cycle has also been empirically examined for a while (Tella 1964) (Strand and Dernburg 1964). The question where the additional workers come from during upturns and go to during downturns can be explained with the discouraged worker effect (DWE). This concept implies that unless the economy is at full capacity, measuring the unemployed workforce underestimates the true level of underemployment. Hereby, the discouraged workers (DW) represent the fraction of the population that, depending on the economic situation, is part of the workforce i.e. employed or actively looking for a job or part of the inactive population.

In the literature, a lot of different terms like ‘discouraged workers’, ‘discouraged worker effect’, ‘labor reserve’ and ‘hidden unemployment’ are used to refer to similar phenomena. These terms are sometimes used as synonyms, even though they partly represent different concepts and differ in terms of measurement (Fuchs and Weber 2010: 8). In this study the DWE is defined as the variation in the LFPR that is the result of a systematic retreat of workers from the labor market in economic downturns and an entering in economic upturns. At this point, it has to be made clear that the term might be misleading, since there are various reasons for retreat or entering that go beyond what is commonly understood as ‘discouragement’ and may additionally differ between individuals. Rather, it is meant in the sense that “the costs of of active job search increase and the benefits go down” (Lee and Parasnis 2014: 90).

In the case of Germany, there are three different estimates of the DW. Eurostat regularly estimates the DW for all OECD countries based on the Labor Force Survey (Department of International Economic and Social Affairs 1984: 126). The German Institute for Economic Research (DIW) also investigated the DWE in the past, using a survey called “German Socio-Economic Panel” (SOEP) (Holst and Schupp 2000).

Both approaches have in common that they ask currently inactive people about their willingness to work under different scenarios, which allows them to directly identify members of the DW. Further, since the subjects can be directly asked, this approach allows the reseracher to collect more information on the socio-economic background and the circumstances of economic inactivity. However, the major drawback of the survey approach is that people are interviewed about a hypothetical situation for example whether they would be willing to work if the economic situation significantly improved. The responses might be strongly biased in two ways. First, it is socially desirable to work and therefore people will tend to answer that they would work, even if they would not. Second, due to phenomena such as ‘cognitive dissonance’ they might answer that they would not work even if they would (Fuchs and Weber 2010: 14). While both effects have different signs, it is not clear which effect outweighs the other.

Since the 1980s, the German Institute for Employment Research (IAB) has regularly investigated and estimated the DWE in (West-)Germany. Instead of using a survey to identify subjects to the DWE, this method makes use of the positive correlation between the visible and the hidden unemployment (Fuchs and Weber 2010: 15). It has the advantage that it is based on the actual labor market behavior of its subjects instead of relying on potentially biased personal statements about hypothetical scenarios. However, depending on the assumptions made about the full employment indicator and the econometric model specification, estimates of the DWE vary significantly.

In addition to the negative relationship between the economic condition and the labor force participation, there is a second phenomenon that works in the reverse direction and is commonly known as the added worker effect (AWE). It follows the logic that economically inactive people enter the labor market under deteriorating economic conditions to compensate for the reduced household income that is induced by the job loss of other household members. As a result, this phenomenon causes increased labor force participation in deteriorating economic conditions. While the AWE has been known for a long time as well, nowadays, it is often assumed to be negligible and therefore omitted in research on the German labor market (Fuchs and Weber 2005b: 9). However, since strong evidence for the AWE is found this effect is also taken into account when estimating the true level of underemployment. The net of the added workers (AW) and the DW will be referred to as the hidden labor force (HLF).

The purpose of this paper is to contribute to the literature of the DWE and the AWE by introducing some variations to the IAB-approach and offering an alternative estimation of the HLF in Germany. Doing so, my main focus will be the DWE, but the AWE will also be investigated. Further, an emphasis is put on the sociodemographic composition of the HLF. Last but not least, it is investigated whether or not the dynamics of the HLF are subject to asymmetry in economic upturns and downturns, as well as before and after the labor market reforms that came into force in 2005.

The remainder of this paper is organized into five sections. In the remaining part of this section, the related literature and its findings are analyzed. Secondly, the basic model is introduced, the HLF is operationalized and the data is delineated. Thirdly, the empirical results are presented and put into context. In the fourth section, potential limitations will be discussed. The fifth section summarizes the results and concludes the paper.

1.2 The business cycle indicator

In order to estimate the DWE (and the AWE), a business cycle indicator has to be chosen. In early research, the employment to population ratio was used (Tella 1964) (Strand and Dernburg 1964). However, this indicator was criticized as including autonomous factors and being upwardly biased (Mincer 1966). Benati (2001) proposed the UR of prime aged males (2554 years old), real GDP or capacity utilization in the manufacturing sector. The Swiss Economic Institute (KOF) uses a principal component of the job vacancy rate, the UR, the growth rate of the wage index and the changes in full-time-equivalent employment (Graff, Mannino, and Siegenthaler 2013: 52). The IAB, in turn, among others uses the general UR, specific URs for example for women, foreigners or the youth and the job vacancy rate (Fuchs and Weber 2010: 18). In the next step, it is necessary to make an assumption about the business cycle indicator under full employment conditions. The KOF uses the Bry-Boschan algorithm to estimate temporal full employment values of their business cycle indicator. They argue that the last peak is used as an orientation to make a decision about labor market participation (Graff, Mannino, and Siegenthaler 2013: 52). While this is certainly a very conservative assumption, most research uses constant and relatively ambitious values. In terms of the general UR, most full employment value assumptions lie between 2% and 4% (Fuchs and Weber 2010: 20).

Based on the idea of Armstrong (1999), who used benchmarks on the regional level in order to estimate the HLF in Northern Ireland, the IAB has come up with a full employment value of 2.5% for Germany. Following Armstrong, they argue that a UR of 2.5% has been persistently observed at the regional level in Upper-Bavaria, thus, is achieveable in Germany on average given the legal, social and tariff framework conditions. They argue that even URs of 2% have been observed, but they assume 2.5% as a cautious estimate (Fuchs and Weber 2010: 21). This procedure is reasonable as the top values only represent rare individual cases of downward deviation from the generally achieved full-employment value.

1.3 Findings on the size, structure and the dynamics

Strand and Dernburg find for the U.S. that the DWE is “strongest among the female population and the very young and very old males” (Strand and Dernburg 1964). The labor market economist Jacob Mincer found for the U.S. that while an increase in the general UR by one percentage point decreased the LFPR of young Caucasians by two percentage points, it decreased the LFPR of young Afroamericans by five percentage points (Mincer 1966). A more recent study even found that for a panel of European countries, there is only a significant DWE for women (Ozerkek 2013). In general, there is a consensus that different subgroups of the population are not equally affected by the DWE. Past research has in particular pointed out that women react much more sensitively to cyclical fluctuations as men. Further, it seems that there is a tendency that generally marginalized groups are more likely to be subject to the DWE than others.

An emerging strand of research related to the DWE is whether or not it is subject to asymmetry. For Australia, Dixon (1996) found evidence that participation is more sensitive during economic upturns as opposed to downturns, while O‘Brien (2011) finds the opposite for the old male population. Lenten (2001), in contrast, does not find asymmetric behavior for Australia at all. Other research for France, Japan, Sweden and the U.S. has found that while Sweden shows symmetric responses, in general, there is a significantly larger reaction among older females to a deteriorating as compared to an improving economic situation (Darby, Hart, and Vecchi 1998). As of now, no clear pattern for asymmetric reactions has been found.

Research from the 1960s in the US found significant evidence that women and younger people are affected by the AWE (Strand and Dernburg 1964). Presently, the AWE is generally neglected in research on OECD countries. This is mainly due to the fact that the DWE outweighs the AWE and it can only be detected for particular subgroups of the population, which makes it necessary to have highly disaggregated data. Moreover, current research “confirm(s) the hypothesis that the added worker effect is dominant in developing countries while the discouraged worker effect is dominant in the OECD countries” (Lee and Parasnis 2014: 96). A more recent study of Fuchs and Weber (2013) using a trend-cycle decomposition finds evidence for both, the DWE and the AWE in Germany which is in contrast to Fuchs and Weber (2005b).

Using the survey approach, the DIW finds that in 1998, the DW in Germany counted 1.8 million people of which 1.1 million were Western German women. Overall they estimated that women account for 70% of the DW (Holst and Schupp 2000). According to the IAB, the DW in Germany in the year 2000 amounted to almost 1.5 million of which 30.7% were Eastern German and 54.4% were women (Fuchs and Weber 2005a: 34) (Fuchs and Weber 2005b: 60). In terms of total DW-estimates for the year 2014, the Federal Statistical Office in Germany estimated the DW to be 990,000 people. Yet, only 510,000 of them were available on short notice, which is more comparable to other estimates. Of these 510,000, 57% were women (Statistisches Bundesamt 2015). The IAB estimated the DW in Germany in 2014 to account for 255,000 people (Fuchs et al. 2015).

2 Conceptual and methodological issues

2.1 The segments of the labor market

Depending upon the economic situation, members of the HLF are either actively participating in the labor market or not. Since the economic situation varies over time, so does the HLF. An economically inactive person is not necessarily part of the DW, and can hardly be directly identified as such. Rather, we have to think of it that we know that a person was part of the DW, if we observe this person systematically entering the labor force under improved economic conditions. The same applies to the AW, who for example have to systematically enter into the labor market under deteriorating economic conditions in order to be identified as such.

Figures 1 and 2 illustrate the DWE and the AWE. The figures are snapshots of the different sections in labor market under good and bad economic conditions. In order to be consistent, the definitions provided by the ILO (International Labor Organization 1982) as implemented in the German Microcensus (Statistisches Bundesamt 2013) will be used. The whole blue ovals represent the part of the population that is above 14 and below 65 years old and can therefore be considered the population of working age. The labor force is represented by the orange ovals. It consists of the employed (paid employed and self-employed) and the unemployed (without work, actively looking for work and available for work within two weeks) population. The whole golden ovals that are partly covered by the others portray the discouraged worker population (DW-Pop). The visible parts of the golden ovals represent the DW under a good (Figure 1) and bad (Figure 2) economic condition. The invisible parts are potentially part of the DW under deteriorated economic conditions. The whole grey ovals depict the whole added worker population (AW-Pop). The invisible parts of the grey ovals represent the AW in the respective situation. The visible parts of the grey ovals represent those who would enter the labor market under deteriorating economic conditions. The fact that the DW-Pop and the AW-Pop embrace only a part of the whole population implies that not every subgroup of the population is affected by the two phenomena (see Section 3.1).

Figure 1:
Figure 1:

Good economic condition.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 2:
Figure 2:

Bad economic condition.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

As the economic situation deteriorates, the DW leave and the AW enter the workforce as illustrated in the transition from Figure 1 to Figure 2. As can be seen, the size of the DW as well as the size of the AW are larger under bad as compared to good economic conditions. Further, the labor force (the orange oval) is larger under the good as compared to the bad economic condition. This is always the case if the DWE outweighs the AWE.

2.2 Estimating the hidden labor force

The two important ratios in order to estimate the HLF are the UR and the LFPR as shown in eqs. (1) and (2) below.

UR=registered unemployed Popregistered unemployed Pop+employed Pop
[30pt]LFPR=employed Pop+unemployed Poptotal Pop

At first, the LFPR of each subgroup i in a given year t is regressed on the lagged UR (see eq. (3)). The reason for using the lagged UR instead of the UR is twofold. Firstly, it is assumed that it takes some time for people to adapt their labor market behavior to the altering economic condition. Secondly, using the lagged UR is in line with the concept of ‘Granger causality’ (see Subsection 2.4 for more information). In order to account for the highly significant persistent pattern in the data, an AR(1)-process in form of the lagged LFPR is included into the model.

LFPRi,t=consi+β1iURt1+β2iLFPRi,t1+ui,t
[15pt]β1_lri=β1i1β2i

The coefficient β1i indicates how the LFPR at time t is affected by shocks in the UR at time t-1. The use of an AR(1)-process in eq. (3) makes it necessary to transform the short-run effect β1i of URi,t1 on LFPRi,t in order to obtain the long-run effect β1_lri (see eq. (4)). The coefficient β1_lri indicates the average strength of reaction in the labor market participation of a particular subgroup i that is associated with changes in the UR. A coefficient of 0.01 would for example indicate that if the UR increases by one percentage point, on average, the LFPR would decrease by one percentage point. In other words, β1_lri shows how people in a particular subgroup i react to changes in the economic situation. Subgroups with a significantly negative β1_lri (at the 10% significance level) are subject to the DWE. If β1_lri is significantly positive, the subgroup i is subject to the AWE.

In the next step, the potential labor force participation rate of subgroup i at time t (PLFPRi,t) is calculated. In order to do that, the potential unemployment rate (PUR) has to be subtracted from the UR to compute the distance from an economy at full capacity. After that, the result has to be multiplied with the expected reaction per unit change which is given by β1_lri (groups with an insignificant β1_lri are omitted). The result has to be subtracted from the LFPR (note that β1_lri is negative if the subgroup i is subjected to the DWE which makes the latter part positive). The PLFPR would equal the LFPR under full employment.

PLFPRi,t=LFPRi,t(URtPURt)β1_lri

Now we can calculate the HLF in a given year by multiplying the difference between the PLFPR and LFPR of group i at time t with its population size at time t and summing these products up over all groups. Notice that if HLFt is positive the DW outweighs the AW and vice versa.

HLFt=i=1n(PLFPRi,tLFPRi,t)Populationi,t

2.3 Data

The LFPRs of the various subgroups for the considered period from 1991 to 2014 are calculated with data that comes from the German Microcensus and is provided upon request by the Federal Statistical Office (Destatis). The Microcensus is a population and labor market survey in the framework of the EU Labor Force Survey and is published annually by the Federal Statistical Office. With the help of this survey, disaggregated LFPRs for 80 subgroups of the population of working age (from 15 to 64 years of age) have been computed. The German population is divided into 10 age groups, into German citizens and non-German citizens, males and females and Western and Eastern Germans (including West-Berlin). The age groups are separated in five year steps from 15 to 64. The reason for distinguishing between Western and Eastern Germans is that both groups were shaped differently during the division of Germany in terms of labor market policy and participation. For example, Eastern German women still exhibit a much higher LFPR as their Western German counterparts as they had a different position in society and were strongly included into the workforce of the German Democratic Republic (see Figure A.1).

Having LFPRs on the disaggregated level for different subgroups instead of just one LFPR for the whole population has several advantages. First, it allows us to analyze if a particular subgroup is even subject to one of the two HLF-phenomena. This might be the case since different groups with different sociodemographic backgrounds have a different labor market behavior and might react differently to shocks like a deteriorating economic condition. Second, a much more detailed picture of the HLF can be drawn. Third, it can be controlled for group-specific time trends.

Unfortunately, there are several missing values and high relative standard-errors for the very small subgroup of Eastern German, non-German females between 60 and 64 years of age. Moreover, due to policy shifts of the early retirement scheme in Eastern Germany during the early 1990s, there are strong distortions in the LFPRs among the Eastern German, Germans between 55 and 59 (Fuchs and Weber 2005a: 11). As a result, no reliable conclusion regarding the effects within these groups can be drawn. In order to avoid flawed and unprecise results, these three groups are excluded from the dataset, leaving us with 77 groups. Even though particular minorities of the population are excluded, the remaining 77 subgroups still represent approximately 98% of the population between 15–64 years of age. The dataset can therefore still be considered representative of the population. Moreover, it has to be noted that a procedural modification of the Microcensus occurred in 2005. Prior to 2005, the data was collected once a year in a fixed reference week in late April. Starting in 2005, the data for the Microcensus has been collected throughout the year. As the former procedure was clearly not representative for the whole year there might be a trend break in the LFPRs in 2005 which will be discussed in Subsection 4.2.

The official UR from the Federal Ministry of Labor and Social Affairs is used as the business cycle indicator. The official UR is subject to changes in procedure based on labor market policy measures in 2005. The major change regarding the UR was to combine members of unemployment benefits and social welfare. The potential implications for the analysis and how it has been taken into account are set out in more detail in Subsection 3.2 and Subsection 4.2.

2.4 Endogeneity and causality issues

Using the UR as the explanatory variable is due to the fact that it is a good proxy for the economic condition and the business cycle in a country due to its coincident countercyclical nature. In this context, the UR is superior to other indicators like the capacity utilization as it has a special emphasis on the employment dimension of the business cycle which is of central interest. In addition, since the UR is omnipresent in the media for example in newspapers, on the internet and on television it is a good proxy for how people perceive the economic condition of their country. Furthermore, using the UR as the explanatory variable in order to estimate the HLF is state of the art in the literature (O‘Brien 2011) (Lee and Parasnis 2014).

In order to avoid spurious regressions, the LFPRs are detrended. In addition, this captures and controls for other, non-cyclical impacts on the LFPRs that are induced by social and socio-economic changes such as an increasing LFPRs of females due to the altering role of women in society. Other examples are decreasing LFPRs of young people which are associated with longer education and training periods. Further, increasing LFPRs of older people can be explained by improved health and stepwise increased retirement ages. Linear time trends are assumed for each group due to the fact that the underlying socio-economic processes as described above affect the LFPRs gradually and persistently. Moreover, plotting the LFPRs over time additionally suggests linear time trends (see Appendix A.1). As can be seen in Figures 11, 12, and A.1 the LFPRs of most groups (especially the 55-64 year old and female groups) show consistent increasing linear time trends. On the contrary, the LFPRs of the young (15–29 year old) exhibit slightly negative linear time trends. Only the LFPR of the group of medium-aged (30–54) males seems relatively constant over time. Interestingly, the LFPRs of the subgroups exhibit signs of convergence as those with lower initial levels increase faster and the range between the lowest and highest value decreases sharply from 1991 to 2014. Consequently, the LFPRs do not represent rates in the proper meaning of the word, but the variation of the LFPRs around the time trend. Performing unit root tests ensures that the detrended LFPRs are stationary.

Since eq. (3) is based on eqs. (1) and (2), which mutually share (related) variables, the model might suffer from endogeneity and reverse causality bias. When conducting Granger causality estimations, patterns of reverse causality are found. The hypothesis that the LFPRs do not cause UR could be rejected for the AW-groups, however, it could not be rejected for the DW-groups. In order to curb endogeneity, the lagged UR is used. Consequently, a Granger-causal impact of the UR on the LFPRs is captured. This means that it is unlikely that the UR is affected by the LFPRs since the UR precedes the LFPRs (Granger 1969). Especially, given that the results on reverse causality are ambiguous, we consider this to be a minor issue.

As a result, there are two phenomena left that have an impact on the LFPRs. Firstly, an increase\decrease in the UR reduces\increases the expected returns of searching employment and thus leads to a retreat\entry from\into the labor market. Secondly, an increase\decrease in the lagged UR implies a reduced\increased household income and increases\decreases the expected benefits of searching employment of other household members. Therefore, other economically inactive\active household members enter\retreat into\from the labor market. While the first refers to the DWE, the second describes the AWE. The coefficient β1_lri captures the net of both effects. Thus, we underestimate the true magnitude of the DWE and the AWE if a particular subgroup is subject to both.

As opposed to Fuchs and Weber (2005b), who regularly publish figures of the DW for the IAB, this approach recognizes that only a net effect can be estimated. Consequently, estimates of the DW and the AW actually represent lower limit estimations since the pure effects cannot be isolated. Furthermore, this study represents a more systematic approach since a general model and business cycle indicator is used consistently for all subgroups. Fuchs and Weber (2005b), in contrast, model the labor market behavior differently for each subgroup. While different models are reasonable and necessary in case of different dynamics and causal chains, the authors leave open why this would be the case here. Fuchs and Weber (2005b: 51–54) illustrates the wide variety of models and (explanatory) variables (even between subsequent cohorts of for example Western German, German males) the authors use to estimate the DWE. This proceeding might be the reason why the authors do not find evidence for the AWE at all.

To summarize, the correlation between the UR and the LFPR is driven by two phenomena, the DWE and the AWE. Both of which take effect in the opposite direction and β1_lri indicates their net effect on group i. Distinguishing between the two effects makes it necessary to have good instruments in an IV-setting. The challenge would be to find data on events where the economic condition deteriorated without affecting the household income at all (for example due to fully compensating unemployment benefits that are granted at least in the medium-term). This would allow to observe the DWE in the absence of an AWE. Unfortunately, unemployment benefits are usually rather short-lived and normally represent only a fraction of the previous income. In order to isolate the AWE a situation of decreased household income under unchanged economic conditions would have to be observed. However, wage cuts are usually due to a changing economic environment. Moreover, for example the necessity to guarantee childcare might become problematic when both parents actively participate in the labor market. Restrictions like this might additionally distort the results and complicate the feasibility of such an approach. Future research will have to tackle these challenges.

In this study, it is made use of the fact that all negative\positive correlation can be attributed to the DWE\AWE. As a result DW-groups are groups for which the DWE significantly outweighs the AWE, while AW-groups are groups for which the AWE significantly outweighs the DWE.

3 Empirical analysis

3.1 The hidden labor force groups

Table 1 shows the results of four different fixed effects regression models with robust standard errors. In the first model all 77 groups are included. As can be seen, the β1_lr-coefficient that is presented in column 1 row 4 (‘long-run effect lagged UR’) indicates a significantly negative relationship between the lagged UR and the LFPRs. However, as we can see in the second model, once we take into account which share of the population each group represents the relationship turns insignificant. It can be concluded that there is no evidence for a significant DWE or AWE at the aggregate. Further, some of the groups that were overrepresented in the unweighted model are subject to the DWE.

Table 1:

Panel regressions.

(1)(2)(3)(4)
LFPRLFPRLFPRLFPR
lagged UR0.002710.000000.001860.00602
(0.00056)(0.00018)(0.00031)(0.00072)
lagged LFPR0.4490.7560.6200.654
(0.0369)(0.0266)(0.0683)(0.0425)
Intercept0.02370.0005710.01680.0535
(0.00517)(0.00169)(0.00281)(0.00683)
long-run effect0.004910.000020.004890.01740
lagged UR(0.00097)(0.00075)(0.00082)(0.00202)
Observations17711771207552
No. of groups7777924
Obs per group23232323
WeightedNOYESYESYES
GroupsALLALLAWDW
F-statistic116.8404.2122.7316.5
(R2)0.2770.6480.6500.790
Adjusted (R2)0.2760.6320.6320.780

Robust standard errors in parentheses. (p<0.05), (p<0.01), (p<0.001)

Note: Models are fixed effects regression models. See eq. (4) for information on how the ‘long-run effect lagged UR’ is derived. The line ‘weighted’ means whether or not the regression is weighted by the population shares of the subgroups. ‘LFPR’ - ‘Labor Force Participation Rate’. ‘UR’ - ‘Unemployment Rate’. ‘AW’ - ‘Added Workers’. ‘DW’ - ‘Discouraged Workers’.

In order to find out which groups are subject to the DWE and the AWE, OLS regressions with robust standard errors are conducted for each group separately (see eq. (3)). A group qualifies as being a DW-group\AW-group if it has a significantly (at the 10% significance level) negative\positive β1_lr-coefficient (see eq. (4)). As a result there are 9 AW-groups and 24 DW-groups left in the dataset that on average represent 30.6% and 16.2% of the population at working age. Table 2 illustrates which groups are subject to the DWE and AWE.

In the third model, the sample is reduced to the 9 AW-groups and in the fourth model to the 24 DW-groups. As expected, their LFPRs exhibit highly significant positive and negative (long-run) relationships with the lagged UR. The β1_lr-value of 0.00489 in the third model indicates that an increase in the UR by one percentage point, on average, increases the LFPR of the AW-Pop by almost half a percentage point. The β1_lr-value of -0.0174 in model four indicates that an increase in the UR by one percentage point, on average, decreases the LFPR of the DW-Pop by 1.7 percentage points.

3.2 The size of the hidden labor force

As mentioned in Section 1.2, it is necessary to make an assumption about the UR under full capacity conditions in order to be able to estimate the absolute size of the HLF. While a UR of 0% seems to naturally correspond to an economy under full employment, it is clearly not feasible, for example due to transition constraints or mismatch. While basically any assumption about an exact level of the UR at full capacity seems arbitrary, Armstrong (1999) offers a reasonable solution as discussed in Section 1.2. This is why I follow Fuchs and Weber (2010) and take the value of 2.5% for the period under scrutiny.

As mentioned in Subsection 2.3 the unemployment rate is subject to a major change in calculation procedure due to the labor market reforms that were finalized in 2005. I follow Fuchs and Weber (2010) who argue that these reforms have directly increased the full employment UR from 2.5% to 3.5%. Based on a study of the German Federal Employment Agency that emphasizes that the labor market reforms directly relieve the UR gradually (Bundesagentur für Arbei 2006: 67), Fuchs and Weber (2010) leave open whether or not this value has to be readjusted downwards at a later point in time.

Following eqs. (3)–(6) the HLF can be estimated. A table with figures for the HLF for all years can be found in Appendix 7. Figure 3 illustrates the UR, the DW, the AW and the net of the latter two, the HLF, in Germany in the considered period from 1991 to 2014. Since the changes in the DW, the AW and the HLF directly depend on the changes in the UR, the lines follow a very similar pattern. The deviations between them can be explained with demographic shifts.

Figure 3:
Figure 3:

The size of the hidden labor force over time.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

As can be seen in the figure, the HLF strongly increased during the 1990s due to deteriorating economic conditions. From around 2003, the HLF started to sharply decrease due to improving economic conditions. The significance of the AWE as a factor of compensation for the negative relationship between the business cycle and the LFPR can be seen quite well in the figure. Depending on the year the AWE even more than halves the impact of the DWE on the labor force participation.

As compared to Fuchs et al. (2015) who estimate the DW to be 255,000 in 2014, the estimate of 500,000 is rather large. This is surprising since according to the IAB all groups are subject to the DWE but might be the result of neglecting long-run effects (Fuchs and Weber 2005b: 51–54). However, the estimate is very close to the estimate of the Federal Statistical Office with 510,000.

3.3 The structure of the discouraged workers

The structure of the DW depends on two variables: the share that the subgroups have in the DW-Pop and their respective β1_lr-coefficients that indicate their sensitivity to changes in the UR. Table 2 illustrates a broad picture about which parts of the population are affected by the DW- and AW-phenomenon including their respective β1_lr-value and significance level. Empty cells represent insignificant β1_lr-values and black cells the excluded groups. As can be seen in the table, the DWE affects primarily young, Non-German and old groups in the population. The AWE, in contrast, seems to primarily impact Western German females of medium age. Further, the magnitude of the effect is strongest among the old.

Table 2:

Disaggregated β1_lr-coefficients.

WGMWGFWNMWNFEGMEGFENMENF
15–190.01070.00970.01840.0071
20–240.00250.00760.00710.02210.0229
25–290.00360.00840.0092
30–340.00210.00740.00100.0017
35–390.00270.00680.0072
40–440.00390.0041
45–490.00670.0080
50–540.00600.00630.0108
55–590.0124
60–640.02470.02730.03140.01460.01820.0309

(p<0.1), (p<0.05), (p<0.01)

Note: In the first line, the first letter is either ‘E’ for ‘Eastern German’ or ‘W’ for ‘Western German’. The second letter refers to either ‘N’ for ‘non-German’ or ‘G’ for ‘German’. The third letter refers to ‘F’ for ‘female’ or ‘M’ for ‘male’. Empty cells represent insignificant coefficients. Black cells represent the excluded groups (see Subsection 2.3). P-values are based on robust standard errors.

While the β1_lr-coefficients are assumed to be constant in the period considered, the shares vary over time. Consequently, the structure of the DW varies over time. Yet, unless there are strong demographic shocks, the structure is relatively stable and rather follows long- or medium-term trends for example due to an aging population. The evolution over time of the structure of the DW broken down by sociodemographic characteristics can be seen in Table 6 in Appendix A.2.

Figures 4, 5, 6, and 7 display the structure of the DW in 2014 broken down by sociodemographic characteristics. As can be seen, some groups are over- or underrepresented in the DW-Pop or the DW as compared to the population or the DW-Pop. The first step (from population to DW-Pop) is based on whether these groups react significantly negative to the UR in their labor market participation behavior. The second step (from DW-Pop to DW) is based on how strong these groups react to changes in the UR.

Figure 4:
Figure 4:

Discouraged workers in 2014 by sex.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 5:
Figure 5:

Discouraged workers in 2014 by region.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 6:
Figure 6:

Discouraged orkers in 2014 by citizenship.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 7:
Figure 7:

Discouraged workers in 2014 by age.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

As can be seen in Figure 4, women are overrepresented in the DW-Pop. Taking their sensitivity to changes in the UR into account hardly compensates for this as can be seen by the third bar. The share of 56.6% again is very close to the findings of the Federal Staticial Office (57%) for 2014 (see Section 1.3).

Figure 5 displays that Eastern Germans are overrepresented in the DW-Pop, but react weaker to changes in the economic condition which partly compensates for that. The DW-share of the Eastern Germans in 2000 and 2014 was 24.2% and thus lower than the findings of IAB of 30.7% for the year 2000 (see Section 1.3). This result is in contrast to the findings of the DIW, where the Eastern Germans play a “minor role” (Holst and Schupp 2000) among the DW.

Figure 6 shows that non-German citizens, which account for 11% of the population, almost triple their share in the first step and account for 31% in the DW-Pop. Their relatively weak sensitivity to changing economic conditions almost halves their share among the DW to 16.6%. As a result, they are still overrepresented by 5.6 percentage points among the DW as compared to the population.

In Figure 7, the shares of three aggregated age groups (1529, 30-54, 55-64) are presented. The young age groups, which account for 25.5% of the population, are underrepresented in both, the DW-Pop (22.7%) and then again among the DW (10.7%). The medium age group represents more than half of the population (54%), but plays a small role in the DW-Pop (16.6%) and an even minor role among the DW (4.4%). Therefore, the medium age groups are extremely underrepresented among the DW. This circumstance can be explained with their role in society. Due to the fact that these groups tend to take care of the young and the old cohort, they are strongly dependent on labor market participation. Additionally, they hardly have an alternative source of income other than employment or unemployment benefits and therefore tend to actively participate in the labor market under good and bad economic conditions. These circumstances also explain why these groups have the highest LFPRs (see Figure 11). The old age groups, on the other hand, demonstrate the highest overrepresentation with a share of 84.9% as compared to a share of 20.5% in the population. This result is mainly due to the fact that they nearly triple their share in the first step from population to DW-Pop (60.7%). In addition, their strong sensitivity to changes in the UR adds another 24.2 percentage points.

The results on the structure of the DW are in line with the literature where women, young and old age groups play the major role in the DW (see Section 1.3). This is in contrast to the estimates of the IAB who counterintuitively find that the medium age groups (2549) play a major role (Fuchs and Weber 2010: 28). Non-German Citizens play an even more important role according to the estimates of the IAB. For the year 2007 they find that German (199,000) and Non-German men (164,000) almost equal each other in absolute numbers (Fuchs and Weber 2010: 26). This is probably associated with an opaque and unrealistic full-employment assumption and estimation method for Non-German males from 25–44 (Fuchs and Weber 2005b: 57). However, the results are generally in line with a more recent study of Fuchs and Weber (2013) where no significant DWE was found for females and males between 15 and 49 years of age while a significant DWE was found for older females and males.

3.4 The structure of the added workers

The structure of the AW also depends on whether particular subgroups are even subject to the AWE and their sensitivity to changes in the economic condition. Table 6 in Appendix A.2 contains figures over time on the structure of the AW broken down by sociodemocrafic characteristics. As can be seen in the table, Non-Germans, Eastern Germans and old people are basically not affected by the AWE.

Figure 8 shows the structure of the AW by age groups. As can be seen, old people do not play a role in the AW-Pop, thus they do not play a role among the AW either. Medium age groups however, account for 89.2% in the AW-Pop and 92.1% of the AW as compared to a share of 54% in the population. While medium age groups are thus strongly overrepresented, young age groups are underrepresented.

A closer look at the structure of the AW by sex clearly indicates that women are extremely overrepresented in the AW-Pop (75.6%) and among the AW (89.2%) as compared to the population (49.6%) (see Figure 9). The core groups of the AW are Western German, German females of medium age (66.7%). These results are in line with a recent study of Fuchs and Weber (2013) who find that the AWE is much stronger for young and medium-aged females as for their male counterparts. Moreover, they do not find evidence that the AWE affects older females or males.

Figure 8:
Figure 8:

Added workers in 2014 by age.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 9:
Figure 9:

Added orkers in 2014 by sex.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

4 Robustness and limitations

4.1 Dynamics of the hidden labor force

An important aspect to examine is whether the dynamics of the HLF are subject to asymmetry in economic upturns and downturns. In order to do that, two dummies are defined. Dummy DDOWN,t refers to an economic downturn and equals one if the UR increases from period t-2 to period t-1 and is zero otherwise. In contrast, dummy DUP,t corresponds to an economic upturn and equals one if the UR decreases from period t-2 to period t-1 and is zero otherwise. These dummies are interacted with the lagged UR and included into the model.

In the first model, all DW-groups are included and in the other three models one aggregated age group of the DW-groups each (see Table 3). The Null hypothesis that states that the β1_lr-coefficients during economic upturns and during economic downturns equal each other can be tested. As can be seen in the ‘Asymmetry test’-line, only the medium age groups react significantly different in both regimes. The same operation was also conducted for men and women, Eastern and Western Germans, Germans and non-Germans as well as for the AW, but the coefficients did not differ significantly (the regression outputs are not included in the paper).

Table 3:

Asymmetry regressions for discouraged workers.

(1)(2)(3)(4)
LFPRLFPRLFPRLFPR
URDDOWN0.005930.006170.003040.00853
(0.000730)(0.000802)(0.000849)(0.00128)
URDUP0.005750.006040.003470.00846
(0.000791)(0.000801)(0.000869)(0.00146)
lagged LFPR0.6650.2880.2240.654
(0.0451)(0.0568)(0.0962)(0.0629)
Intercept0.05180.05570.02940.0755
(0.00710)(0.00740)(0.00785)(0.0125)
long-run effect0.017690.008670.003910.02464
URDDOWN(0.00212)(0.00093)(0.00104)(0.00296)
long-run effect0.017160.008490.004480.02443
URDUP(0.00208)(0.00091)(0.00104)(0.00294)
Asymmetry0.000540.000180.000560.00021
test(0.00060)(0.00033)(0.00032)(0.00091)
Observations552253138161
No. of groups241167
Obs per group23232323
Age groupsAllYoungMediumOld
WeightedYESYESYESYES
F-statistic212.165.912.3199.2
\(R^2\)0.7910.4250.2870.889
Adjusted \(R^2\)0.7800.3940.2430.883

Robust standard errors in parentheses (p<0.1), (p<0.05), (p<0.01)

Note: Models are fixed effects regression models with regime dependent effects. See eq. (4) for information on how the long-run effects are derived. The ‘Asymmetry test’ is performed using the Delta-method (Phillips and Park 1988). The line ‘weighted’ means whether or not the regression is weighted by the population shares of the subgroups. ‘LFPR’ - ‘Labor Force Participation Rate’. ‘UR’ - ‘Unemployment Rate’. ‘Young’ - ‘15–29’. ‘Medium - ‘30–54’. ‘Old’ - ‘55–64’.

On average, the medium-aged part of the DW increased its LFPR by 0.056 percentage points more under improving as compared to deteriorating economic conditions. This implies that on average, it is easier to ‘motivate’ as to ‘discourage’ medium age groups among the DW-Pop. This might be connected to the fact that this age group is strongly dependent on active labor market participation (see Subsection 3.3). However, the absolute magnitude of this effect is very small. On average, the labor force of the medium-aged DW-Pop increases by 127 people more per percentage point decrease in the UR as it decreases per percentage point increase in the UR. Therefore, it can be concluded that the DW- and the AW-groups demonstrate a strongly symmetric reaction to economic upturns and downturns. This result provides strong evidence for the robustness of the estimates and the methods used in this study.

4.2 Policy shifts in 2005

Despite the encouraging results from Subsection 4.1 that underline the robustness of the approach with regard to upturns and downturns, the analysis raises further points of concern that need to be addressed. As party discussed in Subsection 2.3 and Subsection 3.2, there are policy shifts between 2003 and 2005, known as “Agenda 2010”, that have directly affected the German labor market. Furthermore, the way the LFPRs and UR are computed has changed since 2005. Even if, for the UR this change has been taken into account (see Subsection 3.2), the changes in indicators might still distort the analysis and therefore need to be examined in more depth. Moreover, the transformed labor market conditions after 2005 might alter or even dissolve the underlying mechanisms of the DWE and the AWE.

One way of coping with the potential trend break in 2005 would be using data starting in 2005. However, due to the short time horizon since then and the necessity of a reasonable amount of observations, at the moment, this can only be realized using quarterly instead of yearly data. However, when conducting the analysis using quarterly data it has to kept in mind that different dynamics are potentially being captured. Even if it seems reasonable to assume that economic actors on the labor market need some time to adapt to altering situations there might be dynamics that happen during the year. These short-run dynamics would not be detected using yearly data. This is why an analysis using quarterly data rather constitutes a complement than a substitute to the analysis with yearly data. Both of which reveal information about the DWE- and AWE-dynamics for different time frames. Moreover, even if the effects might differ significantly with respect to the magnitude, in general, they should exhibit the same signs as they follow the same general phenomena. Therefore, a sign test for quarterly data constitutes a robustness test for the applied approach in this study. Table 4 illustrates the results of such an analysis using quarterly data. This table directly corresponds to Table 1 in Subsection 3.1.

Table 4:

Panel regressions using quarterly data.

(1)(2)(3)(4)
LFPRLFPRLFPRLFPR
lagged UR0.000580.000960.002260.00287
(0.00116)(0.00035)(0.00054)(0.00103)
1st lag LFPR0.09720.2060.2420.280
(0.0363)(0.0189)(0.0579)(0.0323)
2nd lag LFPR0.1190.1390.1200.190
(0.0367)(0.0202)(0.0541)(0.0339)
3rd lag LFPR0.1160.1570.1020.165
(0.0367)(0.0183)(0.0513)(0.0334)
4th lag LFPR0.3240.4510.3910.336
(0.0339)(0.019)(0.047)(0.0315)
Intercept0.2510.02030.04780.124
(0.0806)(0.0136)(0.0285)(0.036)
long-run effect0.001690.020480.015450.09842
lagged UR(0.0033)(0.00868)(0.00437)(0.06462)
Observations32003200360960
No. of groups8080924
Obs per group40404040
WeightedNOYESYESYES
GroupsALLALLAWDW
F-statistic16.99 782.6 114.5 721.2
\(R^2}\)0.2170.9950.9890.988
Adjusted \(R^2\)0.2140.9950.9880.988

Robust standard errors in parentheses. (p<0.05), (p<0.01), (p<0.001)

Note: This table corresponds to Table 1 using quarterly data from 2005-2015 based on the Microcensus and own calculations instead of yearly data starting in 1991. Models are fixed effects regression models. Four lags are included to control for the very high persistence in the quarterly LFPRs. See eq. (4) for information on how the ‘long-run effect lagged UR’ is derived. The line ‘weighted’ means whether or not the regression is weighted by the population shares of the subgroups. ‘LFPR’ - ‘Labor Force Participation Rate’. ‘UR’ - ‘Unemployment Rate’. ‘AW’ - ‘Added Workers’. ‘DW’ - ‘Discouraged Workers’.

As can be seen in the table, the results differ from the results in Table 1. While the short-run and long-run effects of UR in the unweighted regression over all groups (column 1) are now insignificant, they are significant but close to zero for the weighted regression (column 2). More important are column 3 and 4. Generally, the coefficients of the respective variables have the same signs. However, in column 3 (AW-groups only), the magnitude of the coefficients associated with the unemployment rate are a bit greater in the short-run and much greater in long-run while being equally significant. In column 4 (DW-groups only), the short-run effect of the unemployment rate is much smaller while the long-run effect is much greater than in Table 1. Moreover, the long-run effect is now insignificant. Consequently, it can be concluded that, as expected, the direction of the effects corresponds to those observed for yearly data. This finding supports the robustness of the results in this study. Further, the magnitude of the effects differs significantly. Yet, this might be the result of methodological issues related to a very short time period. Further, it might come from the, relative to the yearly data, high variance of the quarterly LFPR point estimates that is associated with the continuous estimation using smaller samples respectively since 2005. Therefore, these results have to be taken with caution. But there are several other reason that might explain the difference in the estimated magnitude. First of all, it might be that the short-run dynamics estimated using quarterly data differ from the more gradual ones estimated with yearly data because they refer to different effects within the same phenomenon. Secondly, it might be that the changes in the computations of the indicators, UR and LFPR, since 2005 contributed to some bias in the estimation of the original results. Last but not least, it might be that the labor market reforms that fully came into force in 2005 have altered the DWE and AWE. While the former two cannot be disentangled and directly tested for, the latter can be examined using regression models with regime dependent effects similar to those used in Subsection 4.1 for before and after 2005.

Table 5:

Asymmetry regressions before and after 2005.

(1)(2)(3)(4)
LFPRLFPRLFPRLFPR
URD91040.002280.0002220.001750.00496
(0.00054)(0.00018)(0.00036)(0.00063)
URD05140.001530.000620.001610.00370
(0.00051)(0.00019)(0.00043)(0.00067)
lagged LFPR0.4420.7510.6290.669
(0.0372)(0.0260)(0.0722)(0.0375)
Intercept0.01730.003940.01540.0399
(0.00481)(0.00171)(0.00350)(0.00601)
long-run effect0.004080.000890.004720.01497
URD9104(0.00092)(0.00073)(0.00081)(0.00166)
long-run effect0.002750.002490.004350.01117
URD0514(0.00088)(0.00082)(0.00088)(0.00167)
Asymmetry0.001340.001600.000370.0038
test(0.00020)(0.00029)(0.00033)(0.00072)
Observations17711771207552
No. of groups7777924
Obs per group23232323
WeightedNOYESYESYES
GroupsALLALLAWDW
F-statistic88.13297.5985.73285.36
\(R^2\)0.2870.6570.6540.813
Adjusted \(R^2\)0.2860.6410.6340.804

Robust standard errors in parentheses (p<0.1), (p<0.05), (p<0.01)

Note: Models are fixed effects regression models with regime dependent effects. See eq. (4) for information on how the long-run effects are derived. The ‘Asymmetry test’ is performed using the Delta-method (Phillips and Park 1988). The line ‘weighted’ means whether or not the regression is weighted by the population shares of the subgroups. ‘LFPR’ - ‘Labor Force Participation Rate’. ‘UR’ - ‘Unemployment Rate’. D9104 and D0514 are dummies that take on the value of one for the period 1991-2004 and 2005-2014 respectively and are equal to zero otherwise.

Table 5 contains the results examining whether or not we have a trend break in the data in 2005. The table reveals that overall the effect became significantly less negative since 2005 for both, the weighted and the unweighted model. More specifically, the DWE on average decreased by 25%, thus leading to a significantly lower reaction to altering economic conditions since 2005 among the DW. Keeping in mind the results from Subsection 4.1, this finding is unlikely to be driven by the fact that the period after 2005 coincides with an overall improving labor market situation (as measured by a diminishing UR). While the change in the indicator’s computation might have contributed (or even counteracted) to this change, it seems likely that the altering labor market framework conditions are the main cause of this effect. In fact, a part of the labor market reforms consisted of reducing the unemployment benefits which, in turn, has reduced the opportunity cost of employment. Consequently, the reforms tend to leave people more attached to the labor market and dependent on revenues from employment, reducing the fraction of those who can afford to get discouraged in bad and encouraged in good economic times.

In contrast to the finding for the DWE, column 3 in Table 5 reveals that the AWE is not significantly different before and after 2005. This result is surprising, as it has been assumed that both phenomena generally follow similar incentives, just in reversed directions. Under this narrative we would expect the AW to react stronger to variations in the business cycle as due to reduced unemployment benefits households depend more on the secondary worker under economic hardship. Therefore, this finding raises doubt on the narrative from Subsection 2.4 that the sole underlying fundamental of the AWE is the compensation for reduced household income in economically difficult times. It seems likely that other mechanism are at force that cause people entering the labor market in economically difficult times and retreating under improving conditions. At least, it can be argued that the AWE is no very sensitive to variations in the opportunity cost of employment - the unemployment benefits. It might be that other aspects of the labor market reforms like the more active employment agencies or the transition into a more flexible framework for subcontracted employment provide answers to the puzzle. Moreover, the structure of employment in Germany has changed since 2005. First of all, real wages have been more or less stagnant for most of the period. Furthermore, a boom of employment subject to social security has been observed that is to large extent due to increased secondary, temporary and part-time employment (Goecke et al. 2013). It might therefore be the case that an increased reaction to business cycle variations due to more flexible employment opportunities as well as lower unemployment benefits has increased the reaction among the AW. Further, it seems plausible that the overall structural shift towards stronger labor market dependency and great employment opportunities have engaged a part of the AW permanently, at large, leaving the average response to business cycle variations unchanged. Furthermore, it might be that the AW are heterogeneous with respect to the type of employment. This could imply heterogeneous incentives that, in turn, lead to unexpected outcomes as observed in this context. However, it has to be noted that the framework of this study does not allow to make a conclusive assessment with regard to these questions. It is therefore necessary that follow-up studies focus on the type and circumstances of employment among the DW and the AW and their implications on altering economic conditions.

As set out in this section, the study is not free of limitations. It became clear that the DWE and the AWE exhibit different magnitudes for different time frames. Neither exclusively using yearly nor quarterly data allows to draw a complete picture including all interdependencies at force. Moreover, transforming economic framework conditions continuously alter the response of the DW towards the business cycle. It should be kept in mind that the DW have reacted on average 25% weaker towards variations in the economic condition after the finalization of the labor market reforms in Germany in 2005 as compared to before. On the contrary, the methodology used in this study has proven to be robust with respect to economic upturns and downturns. Further, the response of the AW was left unchanged following the reforms. Moreover, it could be shown that the direction of the effects is robust for different groups of the population irrespective of the time frame of the dynamics under observation. Nevertheless, the finding of non-responding AW to the reforms of the first half of the 2000s constitutes an unresolved puzzle that needs to be addressed in subsequent research.

5 Conclusion

This paper underlines the importance of going beyond the UR when assessing the true level of underemployment in an economy. Further, this paper shows that the inaccuracy of the UR increases the further away the economy moves from full employment. Depending on the subgroup, there is not only strong evidence for the DWE, but also for the AWE which is neglected in current research on the German labor market. Since the extent to which a particular subgroup is affected by the DWE and the AWE strongly varies across the population, it is crucial to work with disaggregated data when investigating the two. The analysis shows that given the current structure of the population, the absolute DWE exceeds the AWE which implies that the hidden labor force is always positive. Furthermore, the AWE is an important stabilizer of the workforce, as the AW on average comprise 51% of the size of the DW. This means that on average the AWE halves the impact of the DWE on the labor force participation. The magnitude of the hidden labor force in the considered period from 1991 to 2014 is considerable and varies between 243,000 in 2012 and 777,000 in 2003.

In 2014, females, Eastern-Germans and Non-Germans were overrepresented by 57 percentage points among the DW. While the young (1529) were underrepresented by around 15 percentage points, the medium-aged workforce (3054) was extremely underrepresented by almost 50 percentage points. In contrast, the old workforce (5564) more than quadruples its share, accounting for 84.9% of the DW. These differences across sociodemographic characteristics can be explained with different levels of labor market dependencies which are associated with each groups’ role in society. Moreover, the extend to which a particular subgroup is subject to the DWE strongly depends on the alternatives to gainful employment given to this group in terms of social policy measures for example retirement policy. The AW are dominated by medium-aged, Western German females. This indicates that these women tend to be the ‘secondary’ workers and are under the pressure of compensating for the reduced household income during economically difficult times.

Except for the DW of medium age who react slightly stronger to economic upturns as compared to downturns, there is no evidence for asymmetric responses. The asymmetry of medium-aged workers might be due to their strong labor market dependency which tends to not allow them to retreat under deteriorating economic conditions. However, the magnitude of this effect is negligible in absolute terms.

Information on the structure and the dynamic of the DW and the AW is crucial for policy-makers. As it turned out, the policy reforms, “Agenda 2010”, have reduced the response of the DW to the business cycle on average by 25%, leaving the response of the AW unchanged. Neglecting these effects when setting up policies might flaw their effectiveness and result in unexpected outcomes. Future research on the DWE and the AWE should focus on further disentangling the two. Moreover, working with data from panel surveys which is generally possible but limited due to legal obstacles would be another way to shed further light into the issue (Böhm 2011). A great advantage was that it would allow to analyze the circumstances of cyclical changes in the labor market participation in more detail. Further, it would give us deeper insights into the the type of employment of the DW and the AW which is crucial for understanding and predicting their response to policies. Last but not least, subsequent research should focus on obtaining a conclusive picture regarding the time dimension of the DWE and the AWE. As a result, this would allow policy-makers to design and implement much more effective labor market measures.

Acknowledgment

I would like to express my gratitude to Jesus Crespo Cuaresma for invaluable and constructive feedback and support. Further, I wish to thank the ifo Center for Business Cycle Analysis and Surveys for the collaboration during the early stages of the study, and the German Federal Statistical Office for the provision of the data. Last but not least, I want to thank Peter Winker and two anonymous referees for insightful comments.

References

  • Armstrong, D. (1999), Hidden Male Unemployment in Northern Ireland. Regional Studies 33.6:499–511.

  • Benati, L. (2001), Some Empirical Evidence on the Discouraged Woker Effect. Economic Letters 70:387–395.

    • Crossref
    • Export Citation
  • Böhm, K. (2011), Schätzung der Stillen Reserve mit dem Mikrozensuspanel 2001-2004. Eine Machbarkeitsstudie. IAB Forschungsbericht 2.

  • Bundesagentur für Arbeit (2006), Amtliche Nachrichten der Bundesagentur für Arbeit. 54 Jg. Sondernummer August.

  • Darby, J., Hart R., and M. Vecchi (1998), Labour Force Participation and the Business Cycle: A Comparative Analysis of Europe, Japan and the United States. Working Papers from Business School - Economics, University of Glasgow, can be downloaded at: http://www.gla.ac.uk/media/media_219041_en.pdf (last retrieved 23. 07.2016).

  • Department of International Economic and Social Affairs (1984), Handbook of Household Surveys. United Nations Publication, Series F(31).

  • Dixon, R. (1996), Apparent Asymmetries in the Relation Between the Participation Rate and the Employment Rate in Australia. Department of Economics - Working Papers Series from The University of Melbourne, 522.

  • Fuchs, J., Hummer M., Hutter C., Klinger S., Wanger S., Weber E., Weigand R., and G. Zika (2015), Der Arbeitsmarkt bleibt auf Erfolgskurs. IAB-Kurzbericht 7/2015.

  • Fuchs, J. and B. Weber (2005a), Neuschätzung der Stillen Reserve und des Erwerbspersonenpotenzials für Ostdeutschland (einschl. Berlin-Ost) IAB Forschungsbericht 18/2015.

  • Fuchs, J. and B. Weber (2005b), Neuschätzung der Stillen Reserve und des Erwerbspersonenpotenzials für Westdeutschland (inkl. Berlin-West). IAB Forschungsbericht 15/2005.

  • Fuchs, J. and B. Weber (2010), Umfang und Struktur der westdeutschen Stillen Reserve. IAB Forschungsbericht 11/2010.

  • Fuchs, J. and E. Weber (2013), A New Look at the Discouragement and the Addded Worker Hypothesis: Applying a Trend-Cycle Decomposition to Unemployment. Applied Economics Letters 20:1374–1378.

    • Crossref
    • Export Citation
  • Goecke, H., Pimpertz J., Schäfer H., and C. Schröder (2013), Zehn Jahre Agenda 2010 Eine empirische Bestandsaufnahme ihrer Wirkungen. Institut der deutschen Wirtschaft Köln policy paper 7.

  • Graff, M., Mannino M., and M. Siegenthaler (2013), Die neue Schätzung des schweizerischen Erwerbspersonenpotenzials an der KOF. KOF Analysen 4:45–56.

  • Granger, C. W. J. (1969), Investigating Causal Relations by Econometric Models and Cross-spectral Methods. Econometrica 37:424–438.

    • Crossref
    • Export Citation
  • Holst, E. and J. Schupp (2000), Stille Reserve wichtig für die Arbeitsmarktflexibilität. DIW-Wochenbericht 29/2000:55–84.

  • International Labor Organization (1982), International Labour Organization (ILO) Resolutions Concerning Economically Active Population, Employment, Unemployment and Underemployment. in: 13th International Conference of Labour Statisticians.

  • Lee, G. and J. Parasnis (2014), Discouraged Workers in Developed Countries and Added Workers in Developing Countries? Unemployment Rate and Labour Force Participation. Economic Modelling 41:90–98.

    • Crossref
    • Export Citation
  • Lenten, L. (2001), Does Encouragement Differ from Discouragement? A Study of Asymmetry in the Australian Labour Force. Presented to the School of Economics Seminar University Adelaide. Retrieved on 2016-07-23: https://www.researchgate.net/publication/228397800_Does_Encouragement_Differ_ From_Discouragement_A_Study_of_Asymmetry_in_the_Australian_Labour_Force.

  • Long, C. (1953), Impact of Effective Demand on the Labor Supply. The American Economic Review 43 (2):458–467.

  • Mincer, J. (1966), Labor Force Participation and Unemployment: A Review of Recent Evidence 73–112 in: R.A. Gordon, M.S. Gordon (Eds.), Prosperity and Unemployment.

  • O‘Brien, M. (2011), Discouraged Older Male Workers and the Discouraged Worker Effect. Australian Journal of Labour Economics 14 (3):217–235.

  • Ozerkek, Y. (2013), Unemployment and Labor Force Participation: A Panel Cointegration Analysis for European Countries. Applied Econometrics and International Development 13:67–76.

  • Phillips, P. C. B. and J. Y. Park (1988), On the Formulation of Wald Tests of Nonlinear Restrictions. Econometrica 56:1065–1083.

    • Crossref
    • Export Citation
  • Bundesamt Statistisches (2013), Mikrozensus 2012 Stand und Entwicklung der Erwerbstätigkeit in Deutschland. Fachserie 1, 4.1.1.

  • Bundesamt Statistisches (2015), Ungenutztes Arbeitskräftepotenzial im Jahr 2014: 6 Millionen Menschen wollen (mehr) Arbeit. Pressemitteilung 21.Mai 2015-186/15.

  • Strand, K. and T. Dernburg (1964), Cyclical Variation in Civilian Labor-Force Participation. Review of Economics and Statistics 46:378–391.

    • Crossref
    • Export Citation
  • Tella, A. (1964), The Relation of Labor Force to Employment. Industrial and Labor Relations Review 17 (3):63–89.

Footnotes

Code and Datasets

The author(s) published code and data associated with this article in the ZBW Journal Data Archive, a storage platform for datasets. See: https://doi.org/10.15456/jbnst.2017231.110104.

Appendix

A.1 The labor force participation rate over time

Figure 10:
Figure 10:

The labor force participation rate and the unemployment rate over time.

Note: The labor force participation rate (LFPR) relates to the total population of working age. UR is the unemployment rate.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 11:
Figure 11:

The labor force participation rate by age and sex.

Note: The linear time trends are indicated in black.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure 12:
Figure 12:

The labor force participation rate by citizenship and sex.

Note: The linear time trends are indicated in black.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

Figure A.1:
Figure A.1:

The labor force participation rate by region and sex.

Note: The linear time trends are indicated in black.

Citation: Jahrbücher für Nationalökonomie und Statistik 237, 5; 10.1515/jbnst-2017-0110

A.2 The hidden labor force disaggregated

Table 6:

Hidden labor force shares over time.

Discouraged workersAdded workers
19912000201020141991200020102014
Men0.410.440.440.440.130.140.110.11
Women0.590.560.560.560.870.860.890.89
Germans0.870.860.810.831111
Non-Germans0.130.140.190.170000
Western Germans0.740.760.780.760.990.990.990.99
Eastern Germans0.260.240.220.240.010.010.010.01
Young0.140.100.130.110.100.070.080.08
Medium0.040.030.050.040.900.930.920.92
Old0.820.860.830.850000

Note: This table contains the shares over time of aggregated sociodemographic groups among the discouraged and the added workers.

A.3 Hidden labor force estimates

Table 7:

Estimates over time.

YearPopDWAWHLFURLoUΔLFPRPLFPRΔ
199154,7396383502886.97.60.771.371.90.5
199254,9967584183417.78.50.871.071.60.6
199355,4019305184128.99.91.070.771.40.7
199455,33810215784429.610.61.070.971.70.8
199555,28210085624469.410.41.070.571.30.8
199655,507118864953910.411.61.270.471.41.0
199755,639139273166111.412.91.570.771.91.2
199855,637141370970311.112.71.670.772.01.3
199955,604138366072310.512.11.671.172.41.3
200055,42612755866899.611.11.571.072.21.2
200155,30512525726809.410.91.571.572.81.2
200255,22313406047369.811.41.671.773.01.3
200355,051143365677710.512.11.672.273.61.4
200454,763139364474910.512.11.672.373.71.4
200555,116127266660711.713.01.373.774.81.1
200654,844104858446410.811.81.074.975.70.8
200754,53677343433899.70.775.576.10.6
200854,3765943372577.88.40.675.876.30.5
200954,0886343552798.18.70.676.276.70.5
201053,8845953222737.78.30.676.577.00.5
201152,6055202742477.17.60.577.177.60.5
201252,7234932502436.87.30.577.077.50.5
201352,8315242562686.97.50.677.477.90.5
201452,9555002412596.77.30.677.578.00.5

Note: Column 2–5 are in thousands. Column 6–11 are in percent\percentage points. ‘Pop’ - ‘Total Population between 15 and 64’. ‘DW’ - ‘Discouraged Workers’. ‘AW’ - ‘Added Workers’. ‘HLF’ - ‘Hidden Labor Force’. ‘UR’ - ‘Unemployment Rate’. ‘LoU’ - ‘Level of Underemployment’ (=registered unemployed Pop.+HLFregistered unemployed Pop.+HLF+employed Pop.). ‘LFPR’ - ‘Labor Force Participation Rate’. ‘PLFPR’ - ‘Potential Labor Force Participation Rate’.

If the inline PDF is not rendering correctly, you can download the PDF file here.

  • Armstrong, D. (1999), Hidden Male Unemployment in Northern Ireland. Regional Studies 33.6:499–511.

  • Benati, L. (2001), Some Empirical Evidence on the Discouraged Woker Effect. Economic Letters 70:387–395.

    • Crossref
    • Export Citation
  • Böhm, K. (2011), Schätzung der Stillen Reserve mit dem Mikrozensuspanel 2001-2004. Eine Machbarkeitsstudie. IAB Forschungsbericht 2.

  • Bundesagentur für Arbeit (2006), Amtliche Nachrichten der Bundesagentur für Arbeit. 54 Jg. Sondernummer August.

  • Darby, J., Hart R., and M. Vecchi (1998), Labour Force Participation and the Business Cycle: A Comparative Analysis of Europe, Japan and the United States. Working Papers from Business School - Economics, University of Glasgow, can be downloaded at: http://www.gla.ac.uk/media/media_219041_en.pdf (last retrieved 23. 07.2016).

  • Department of International Economic and Social Affairs (1984), Handbook of Household Surveys. United Nations Publication, Series F(31).

  • Dixon, R. (1996), Apparent Asymmetries in the Relation Between the Participation Rate and the Employment Rate in Australia. Department of Economics - Working Papers Series from The University of Melbourne, 522.

  • Fuchs, J., Hummer M., Hutter C., Klinger S., Wanger S., Weber E., Weigand R., and G. Zika (2015), Der Arbeitsmarkt bleibt auf Erfolgskurs. IAB-Kurzbericht 7/2015.

  • Fuchs, J. and B. Weber (2005a), Neuschätzung der Stillen Reserve und des Erwerbspersonenpotenzials für Ostdeutschland (einschl. Berlin-Ost) IAB Forschungsbericht 18/2015.

  • Fuchs, J. and B. Weber (2005b), Neuschätzung der Stillen Reserve und des Erwerbspersonenpotenzials für Westdeutschland (inkl. Berlin-West). IAB Forschungsbericht 15/2005.

  • Fuchs, J. and B. Weber (2010), Umfang und Struktur der westdeutschen Stillen Reserve. IAB Forschungsbericht 11/2010.

  • Fuchs, J. and E. Weber (2013), A New Look at the Discouragement and the Addded Worker Hypothesis: Applying a Trend-Cycle Decomposition to Unemployment. Applied Economics Letters 20:1374–1378.

    • Crossref
    • Export Citation
  • Goecke, H., Pimpertz J., Schäfer H., and C. Schröder (2013), Zehn Jahre Agenda 2010 Eine empirische Bestandsaufnahme ihrer Wirkungen. Institut der deutschen Wirtschaft Köln policy paper 7.

  • Graff, M., Mannino M., and M. Siegenthaler (2013), Die neue Schätzung des schweizerischen Erwerbspersonenpotenzials an der KOF. KOF Analysen 4:45–56.

  • Granger, C. W. J. (1969), Investigating Causal Relations by Econometric Models and Cross-spectral Methods. Econometrica 37:424–438.

    • Crossref
    • Export Citation
  • Holst, E. and J. Schupp (2000), Stille Reserve wichtig für die Arbeitsmarktflexibilität. DIW-Wochenbericht 29/2000:55–84.

  • International Labor Organization (1982), International Labour Organization (ILO) Resolutions Concerning Economically Active Population, Employment, Unemployment and Underemployment. in: 13th International Conference of Labour Statisticians.

  • Lee, G. and J. Parasnis (2014), Discouraged Workers in Developed Countries and Added Workers in Developing Countries? Unemployment Rate and Labour Force Participation. Economic Modelling 41:90–98.

    • Crossref
    • Export Citation
  • Lenten, L. (2001), Does Encouragement Differ from Discouragement? A Study of Asymmetry in the Australian Labour Force. Presented to the School of Economics Seminar University Adelaide. Retrieved on 2016-07-23: https://www.researchgate.net/publication/228397800_Does_Encouragement_Differ_ From_Discouragement_A_Study_of_Asymmetry_in_the_Australian_Labour_Force.

  • Long, C. (1953), Impact of Effective Demand on the Labor Supply. The American Economic Review 43 (2):458–467.

  • Mincer, J. (1966), Labor Force Participation and Unemployment: A Review of Recent Evidence 73–112 in: R.A. Gordon, M.S. Gordon (Eds.), Prosperity and Unemployment.

  • O‘Brien, M. (2011), Discouraged Older Male Workers and the Discouraged Worker Effect. Australian Journal of Labour Economics 14 (3):217–235.

  • Ozerkek, Y. (2013), Unemployment and Labor Force Participation: A Panel Cointegration Analysis for European Countries. Applied Econometrics and International Development 13:67–76.

  • Phillips, P. C. B. and J. Y. Park (1988), On the Formulation of Wald Tests of Nonlinear Restrictions. Econometrica 56:1065–1083.

    • Crossref
    • Export Citation
  • Bundesamt Statistisches (2013), Mikrozensus 2012 Stand und Entwicklung der Erwerbstätigkeit in Deutschland. Fachserie 1, 4.1.1.

  • Bundesamt Statistisches (2015), Ungenutztes Arbeitskräftepotenzial im Jahr 2014: 6 Millionen Menschen wollen (mehr) Arbeit. Pressemitteilung 21.Mai 2015-186/15.

  • Strand, K. and T. Dernburg (1964), Cyclical Variation in Civilian Labor-Force Participation. Review of Economics and Statistics 46:378–391.

    • Crossref
    • Export Citation
  • Tella, A. (1964), The Relation of Labor Force to Employment. Industrial and Labor Relations Review 17 (3):63–89.

FULL ACCESS

Journal + Issues

The Journal of Economics and Statistics is a scientific journal published in Germany since 1863. The Journal publishes papers in all fields of economics and applied statistics. A specific focus is on papers combining theory with empirical analyses.

Search