Gender Segregation at Work over Business Cycle—Evidence from Selected EU Countries

Abstract

In this paper, we investigate whether gender employment rate responses to upward and downward fluctuations in the business cycle are symmetric and whether these responses differ depending on gender employment segregation in sectors and on different type of welfare states using the Esping-Andersen’s classification. We use the VAR model both in linear and non-linear (asymmetric) specification of GDP shocks and impulse response function. We find no convincing evidence of discouraged worker effect as it occurs in neither country fully, which suggests not hidden unemployment but rather the phenomenon of involuntary part-time workers becoming more common with the increase of precarious employment. Furthermore, we find that the pattern of gender employment adjustments to GDP fluctuations indicates that the gender sectoral segregation is a deeply entrenched feature within given economic sectors (construction, education, and accommodation) in all studied countries (Germany, Poland, and Portugal). Hence, this stagnation of gender segregation contributes to the preservation of gender pay differentials in spite of many years of equal pay legislation in the EU members.

1. Introduction

Changes in labour markets over the business cycle are not gender neutral. As the global financial and economic crisis showed, employment has more or less dropped in every country and unemployment rates have increased dramatically, but women and men were differently affected. Moreover, this gendered impact on labour markets depends on the phase of the business cycle (recession, recovery, or austerity). In the recession phase of the last crisis, female employment was generally impacted less than male employment, while during the recovery phase, male employment recovered faster than female employment. The austerity phase generally affects the sectors in which women are over-represented (social service activities, health, and education); as a result, female employment is expected to be impacted more strongly than male employment [1,2].

This gendered effect of the crisis is mainly explained by the sectoral gender segregation which is still at a high level in the European labour market. Moreover, it is viewed as one of the main factors underlying the gender pay gap across the sectors [3,4,5]. Recent research [6] using data from 2010 reports that, for every country, a significant part of the gender pay gap is due to the fact that women are over-represented in sectors with lower pay levels, e.g., education, health, and social work. On the other hand, men dominate high paying sectors, such as construction and chemical products, electric, and transport equipment. Although the causes underlying the gender pay gap are numerous and complex, reflecting discrimination on the grounds of gender as well as inequalities linked to education and the labour market, the gender segregation is considered as a main factor [7]. This is the reason why the gender pay gap can be viewed as a monetary ‘façade’ of gender segregation [8]. The effects of gender segregation are highly detrimental both to women’s and men’s chances in the labour market and in society in general (e.g., it narrows life choices, education and employment options, leads to unequal pay, further reinforces gender stereotypes, and limits access to certain jobs while also perpetuating unequal gender power relations in the public and private spheres) [3]. As a result, gender segregation is a major factor hindering the stimulation of more competitive, sustainable, and inclusive growth [3].

The central interest in this paper is to investigate the responses of the rate of feminisation and masculinisation (measured by female and male employment rate) to the business cycle shocks and to consider a potential impact on gender inequalities from the sectoral perspective. We are especially interested in testing whether feminisation/masculinisation rate responses to upward and downward fluctuations in the cycle are symmetric and to what extent these responses differ depending on gender employment segregation in chosen sectors, i.e., sectors over-represented by women (education), dominated by men (construction), or mixed (gender balanced), e.g., accommodation and food services [1]. Exploring asymmetric responses is fundamental to evaluate the relative importance of the “discouraged worker effect” when downward employment responses during downturns do not correspond with upward movements during upturns. This aspect of cyclical movements in participation rates is related to a “hidden” unemployment stemming from economic events that discourage workers from participating in the labour market during recessions.

Our first contribution is to examine the role of sectoral gender segregation in explaining gender differences in the employment response over the business cycle and its possible influence on gender inequalities. While a vast body of literature exists on gender differences in labour market variables (employment, unemployment, participation, and wages) over the business cycle [9], surprisingly little quantitative work has been done in the context of the evolution of sectoral gender employment and its response to the business cycle, as well as its linkage to gender wage gap [1,5]. Moreover, we take into account the possible asymmetric behaviour of gender employment over the business cycle, i.e., employment rate may contract more deeply and sharply during recessions than its expansion in booms. Understanding the role of sectoral gender segregation in the labour market is important from the policy perspective as it helps formulate the gender-aware rather than gender-blind policy responses to recession and recovery, and thereby supports activities aimed at reducing gender pay gap which is heavily influenced by sectoral segregation [3,10].

The second contribution of the paper is that it analyses the effect of the business cycle on the gender employment for countries, reflecting not only differences in countries’ macroeconomic situations and their institutional environments, but also different types of welfare states using the Esping-Andersen [11] classification. This led to our choice of countries—Germany, Portugal, and Poland. The Esping-Andersen continental model is represented by Germany. Portugal is known as the Mediterranean welfare state, while Poland represents the mixed model which combines elements of all three of Esping-Andersen’s classifications, i.e., democratic, continental and Mediterranean welfare states [12]. This approach allows for considering the country-specific impact of gender segregation and thus the ways to tackle them, e.g., empirical evidence demonstrates that women’s working hours depend very much on the specific country’s family policies, suggesting that women work more when there are easily available childcare places and less if family allowances are high [13].These findings point to a broad space for diverse public policy tools across countries, such as social security, labour market, and economic tools, to tackle stereotypical views on gender roles and gender segregation simultaneously.

The third contribution of the paper is that it employs a single-country analysis using time series techniques to the relatively longer time period (2008Q1-2020Q1), which is in contrast to most existing literature on sectoral segregation based on the cross-sectional analysis covering the smaller time period after the 2008 economic and financial crisis. The main reason for this approach is the diversity of countries under review with regard to welfare states regimes representing different degrees of both “de-familialisation” and “de-commodification”, social protection, public service and family policy provision, labour market regulation, and fiscal policy which may differently affect gender inequalities in the country-specific labour market. To the best of our knowledge, there is no such study that uses the time series approach to identify the dynamic linkages between gender employment and GDP from a sectoral perspective both in the context of welfare states regimes and to discuss possible impact of gender segregation at work on gender pay gap.

The paper is organised as follows. In Section 2, we discuss the gender segregation over the business cycle based on the selected literature. Data and methodology are presented in Section 3, then in Section 4 we concentrate on examining the responses of the gender employment rate to fluctuations in economic activity across the chosen sectors and countries. Section 5 concludes our discussion.

2. Review of Literature

It has been well documented that aggregate labour and output move asymmetrically over the business cycle [14,15,16,17,18,19], i.e., (un)employment moves more during recessions than booms. The consequences of cyclical movements in (un)employment may differ by gender. It is traditionally accepted that the labour supply curve is more elastic amongst women [20,21] and this evidence is often considered as the basis of existing differences in (un)employment across the business cycle [1,19,22,23,24]. Moreover, it is evidenced that during recessions male unemployment rises faster than female one, reducing the gender gap, while in recoveries, male unemployment falls faster, returning the gender gap to some trend. Various complex and interacting factors could be theoretically given as to why the effect of the business cycle on the labour market is different for women and men. These include: women’s commitment to participation as well as state policy and support for their employment; gender-related differences in individual economic agent’s responses both in the supply and demand of labour; the level of attachment to the labour market and exposure to macroeconomic shocks; levels of job tenure; and human capital accumulation and how households pool resources, income, and risk [25,26,27,28,29]. These factors might theoretically be manifested as the so-called added worker effect, i.e., the increase in the labour supply of married women when their husbands become unemployed. However, the evidence of the significance of this effect is mixed [25,30,31]. Since the European Employment Strategy in 1997, the female labour force participation has risen, entailing that women are no longer the secondary earner in families, as the added worker effect assumes.

Another reason for explaining the gender dimension of the business cycle is that men and women tend to be occupied in economic sectors that are affected differently by recessions and booms [1,27,32]. In spite of some encouraging trends towards gender equality in employment, the gender segregation of the labour market is commonly reported in many countries [33,34]. This gender segregation refers to the concentration of women or men (over- or underrepresentation of women or men) in different sectors and occupations (horizontal segregation) or/and to the concentration of women or men in different grades, levels of responsibility, or positions (vertical segregation)—see [3].

There are many theoretical explanations for gender segregation [3,4,33,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51]. As reviewing them in detail is beyond the scope of this paper, we will mention only a few concerning the sectoral gender segregation. According to neo-classical economics, gender segregation results from either different human capital investments of women and men or their preferences regarding household work and child-rearing [4,35]. This concept presumes that women are inclined to concentrate on home labour rather than pursuing careers and as a result they select sectors and occupations with lower human capital requirements, with more flexibility regarding intermittent labour force participation and part-time working which then is reflected in gender pay differences based on different marginal productivities [4]. As a result, the sectoral and occupational gender segregation emerges. However, women’s preferences have changed with the growing awareness of gender equality [44], making other explanations of gender segregation more meaningful. Overall, sectoral and occupational gender segregation is strongly related to educational gender segregation and factors influencing the choice of study fields or occupations that women and men make [41,42,45]. These choices are enforced by stereotypes and the long-lasting cultural beliefs regarding gender appropriate attributes, skills, and occupations [41,42]. For example, participation in the STEM (science, technology, engineering, and mathematics) sector is traditionally associated with various masculine identity traits and roles, while participation in the EHW (education, health, and welfare) sector is traditionally deemed feminine [10,46,47,48]. As the wages in male-dominated sectors (e.g., construction, ICT, or real estate) are usually higher than in female-dominated sectors (e.g., education, health care, or social services), the sectoral gender segregation has a fundamental impact on the gender pay gap. Other factors affecting sectoral and occupational gender segregation are also mentioned, such as an individual’s achievements, motivation, self-confidence and self-efficacy, characteristics of work (working conditions, risks for health and safety), characteristics of employment (working hours, wage, job security), and many others characteristics that affect the subjective and objective comfort of employment [33,42,49,50].

Here, we focus on three main hypothesis which, as Rubery [26] has shown, can theoretically explain the differences in gender employment responses to recessions and booms. These hypotheses are the substitution hypothesis, buffer hypothesis, and sex segregation hypothesis. The substitution hypothesis assumes that, during recessions, the employers wanting to minimise their production costs look for a cheaper labour force, i.e., women who are paid less than men. Hence, female employment evolves counter-cyclically (added worker effect). This hypothesis relies on the fact that women are treated as secondary workers, although the historical facts denied this vision of women.

The buffer hypothesis refers to workers trapped in a second labour market with low quality jobs, low social and labour protections, and weak bargaining power, but at the same time, this labour market is highly flexible with a high degree of competition between workers. Women are over-represented in this secondary labour market [51] and are in the front line of losing jobs during recessions. One gender interpretation of the buffer effect is the concentration of women in specific sectors, i.e., with lower pay than in male sectors. Another gender interpretation of the buffer hypothesis is that women are more sensitive to the discouraged effect due to the lesser importance attributed to their employment and social norms supported by the male breadwinner model. Hence, the female labour force constitutes a flexible reserve that can be expelled from the labour market during recessions while being encouraged to participate during booms. In conclusion then, female employment is pro-cyclical.

The sex segregation hypothesis assumes that labour markets are characterised by a rigid sex-typing of occupations and sectoral segregation. Hence, women can be more or less hit by the recession compared to men, depending on the sectoral nature of the crisis. If recession is concentrated in female-dominated sectors, then employment in these sectors falls and female employment is more affected by the recession. Therefore, female employment might be pro- or counter-cyclical according to the nature of the crisis.

A number of studies show that sex-based sectoral and occupational segregation is a major contributor to the pay gap [5,6,52,53,54,55]. Recent research [6] based on data from 2010 concludes that for all EU countries, a significant part of the gender pay gap is due to gender segregation. The gender pay gap remains remarkably resilient across all Member States, despite more than thirty years of equal pay legislation. It is a serious problem as it affects not only current earnings, but also lifetime earnings and pension entitlements [56]. Although the average pay gap at the occupational level in the EU is fairly small, the results across sectors in member countries are rather different [3].

Although there has been no clear trend towards the reduction of the gender pay gap [57], the pay gap responds differently to the business cycle. As empirical evidence shows [58], the gender pay gap expands during expansions and contracts during recessions, i.e., in the case of positive growth rate of GDP, female earnings decrease relative to men’s, and when the unemployment rate increases, female earnings increase relative to men’s. These results are in line with the nature of the labour supply curves of women and men, i.e., longer commitment to workforce puts men in a better position compared to women which enables them to gain higher earnings during expansions [55]. However, the main reason for this difference seems to be sectoral gender segregation [1,27,32].

3. Materials and Methods

3.1. Country Specific Characteristics and Data

We have chosen the following countries: Germany, Portugal, and Poland representing continental, Mediterranean, and mixed types of welfare states respectively using the Esping-Andersen classification [11]. However, the welfare states approach is not a precise description of welfare approaches in specific countries, but rather sets out ‘ideal-type’ models. Moreover, many of the core assumptions on which welfare states were constructed are no longer tenable and welfare states are now under threat from a number of directions: demographic change, pressures at national level from economic globalisation, the increasingly explicit EU-level requirements for national budget discipline, and the changing nature of work, climate change, or the financial crisis and Eurozone crisis more recently. The last two events in particular changed labour market regulation and social protection rules in the countries under review, but each country has been influenced differently.

Germany was strongly hit, first by the financial crisis, which brought strong decrease in GDP, and later by the Eurozone crisis. Yet in both crises, employment and unemployment were almost unaffected. This, sometimes referred to as the German “labour market miracle”, was made possible by the stabilisation of the labour market through a massive decrease in average working time. However, it is stressed that this German success has come at the price of a growing share of people with low wages and precarious working conditions, alongside a persistently large base of long-term unemployment. Both the prevention of job losses and the fiscal stimulus packages (increases in public investment, tax cuts and cuts in social security contribution rates, as well as the “cash for clunkers” programme in which the car industry was supported by subsidies for new car purchase) have stabilised Germany’s internal demand and allowed a relatively quick economic recovery after the crisis [59].

Domestic economic conditions in Portugal have worsened (fiscal deficit and debt to GDP ratios sharply increased) since 2009 and the Portuguese government had to implement a programme of fiscal consolidation and structural reforms. These austerity-driven reforms have introduced the following changes in the labour market: sizeable reduction of security in employment and unemployment for workers with permanent contracts (insiders) and decentralisation of wage bargaining shifted the balance of power towards employers, which were not offset by any significant unemployment compensation strategy or activation effort targeted at outsiders. As a result, these reforms have changed labour market regulation substantially and brought to the Portuguese labour market a liberalised dualisation [60].

In Poland, the global financial crisis had only a limited impact on the economy in its initial phase. At the first wave of the crisis, the GDP growth only slowed down (due to deteriorating environment in private enterprise sector damaged by losses on currency options), but the labour market appeared very robust at that time. However, the “green island in the sea of red” status of the Polish economy encouraged some migrants to return home (mainly from Ireland and the UK) which pushed the labour supply up and raised the unemployment rate and since 2010 made employment stagnate. In the second wave of the crisis, the Polish domestic demand showed much less resistance. It is only since 2013 that economic activity demonstrated some signs of acceleration and employment has started to pick up, with the unemployment rate starting to decline. Generally, the actions undertaken in Poland to tackle the consequences in the labour market were rather modest [61].

The data (available in Zenodo, see link at the end of paper) considered in the study consist of the quarterly female and male employment rates in the chosen economic sectors (construction, education, and accommodation and foodservices) and GDP per capita (in euro) across the countries under review. These economic sectors correspond to three specific cases of gender segregation: overrepresentation of men (construction), overrepresentation of women (education), and balanced representation of women and men (accommodation and food services). Moreover, the representation of women and men across these sectors is similar in three compared countries. The sample period runs from 2008:Q1 to 2020:Q1. Data have been gathered from Eurostat database and is seasonally adjusted using the TRAMO/SEATS procedure (see Figure 1, Figure 2 and Figure 3).

Although the total employment rates of female and male is growing (except Portugal in 2012 as a result of worsening economic conditions), sectoral employment rates are different across countries and sectors.

Not surprisingly, the widest gender employment gap is observed in the construction sector which is highly overrepresented by men (Figure 2). The proportion of male workers ranges from 85% in Germany to 93% in Portugal and Poland in the third quarter of 2017. In turn, the education sector in dominated by women, but cross-country differences are observable. The share of female workers in the education is at 71% in Germany, whereas in Portugal and Poland it is at 78% and 81% in the third quarter of 2017, respectively. The accommodation and food services sector is rather gender-balanced in Germany and Portugal as the share of female workers ranges from 58 to 60%, while in Poland a greater dominance of women is observed (70–72%).

The employment rates across countries and sectors reveal no significant change between 2008 and 2017. However, a slight upward trend in the female employment rate in the construction sector has been detectable for the last 5 years in all the countries. In contrast, the share of female employment in education sector has been increasing slightly, except in Portugal. Hence, sex segregation in education still appears to be strengthening as the representation of women is still dominating. The most important reason why men are not attracted into education (especially pre-primary and primary teaching) remains the low pay in this sector [62]. In fact, this extreme feminisation has made the pay gap grow.

Changes in the accommodation and food services sector seem to affect more men because the share of male workers has increased slightly more (except in Poland where both the female and male employment rate is rather stable). However, the proportion of women and men has not been affected considerably.

3.2. Methods

We consider the following VAR model to investigate the relationship between female (male) employment rate and GDP growth in different economic sectors (construction, education, and accommodation and food services) in the three chosen countries (Germany, Poland and Portugal):

$$Y_{j,t}=c+{\displaystyle \sum }_{i=1}^p{A}_{j,i}{Y}_{j,t-i}+{\eta }_t$$

(1)

where Y_t is a n × 1 vector of endogenous variables, c is an n × 1 vector of constants, A is a n × n matrix of coefficients, and i = 1,…, p is the number of lags, η_t is n × 1 vector of error terms with zero mean and the variance Ω. Subscript j refers to the particular gender group, i.e., females and males.

Endogenous variables include the female (male) employment rate ($E{R}_t$) as the main variable of interest and GDP per capita in euro ($GD{P}_t$). All variables are expressed in logarithms. The system of equations in the VAR model takes into account all the direct and indirect effects of GDP changes on the changes in employment rate through the estimation of interactions among endogenous variables.

Before studying the effects of GDP changes on the gender employment rate in different economic sectors, we will proceed to investigate the stochastic properties of the series considered in the VAR model by examining their order of integration on the basis of unit root tests. We will hence perform the ADF-GLS test and the ADF test. The results of these tests indicate that the first differences of all variables are stationary (reported in Zenodo, see link at the end of paper). Therefore, we define the vector Y_t in (1) as first-differences of all variables, i.e., $Y_t^{\prime }=\left[\begin{array}{cc}∆E{R}_t& ∆GD{P}_t\end{array}\right]$.

To find the suitable lag length of VAR model, we consider different tests; namely the likelihood ratio test, as well the Akaike, Schwarz, and Hanan-Quinn criteria. Whenever there is disagreement among them, the optimal lag length is chosen using the likelihood ratio test.

In estimating the VAR model, we follow the convention laid out by most of the literature to include both benchmark (linear) and non-linear (asymmetric) specifications of GDP shocks in our analysis. In the linear specification, we assume that the impact of changes in GDP on the gender employment rate is symmetrical, i.e., the response of employment rate to increases and decreases of GDP is similar and, as a consequence, we do not account for GDP shocks separately. In order to account for the existence of asymmetric responses where the effects of upward and downward changed in GDP may not be the same, we consider the non-linear (asymmetric) specification in which upward and downward changes of GDP are included into the model as separate variables $Y_t^{\prime }=\left[\begin{array}{ccc}∆E{R}_t& ∆GD{P}_t^{up}& ∆GD{P}_t^{down}\end{array}\right]$ which are defined as follows:

$$∆GD{P}_t^{up}=\left\{\begin{array}{c}∆GD{P}_t if ∆GD{P}_t>0\\ 0 if ∆GD{P}_t<0\end{array}\right. \mathrm{and} ∆GD{P}_t^{down}=\left\{\begin{array}{c}0 if ∆GD{P}_t>0\\ ∆GD{P}_t if ∆GD{P}_t<0\end{array}\right.$$

The estimation results are used to conduct two tests for asymmetric effects of GDP changes on changes in female (male) employment rate. First, the Granger causality test is used for the effects of each upward and downward shocks in GDP changes on female (male) employment rate. However, the differences between these effects may not provide strong evidence for asymmetry in the effects, because the differences may not be statistically significant. Therefore, in the second test the long-run multipliers estimated by the VAR model are used to test for the asymmetrical effects of upward and downward movements in GDP changes.

The null hypothesis is the following: $H_0:{{\displaystyle \sum }}_{i=1}^p{\delta }_i^{up}$=${{\displaystyle \sum }}_{i=1}^p{\delta }_i^{down}$, where =${\delta }_i^{up}$ and ${\delta }_i^{down}$ are upward and downward impact multipliers representing the effects of the upward and downward shocks in GDP in given sectors on female (male) employment rate. For each VAR model, the residual series are tested for serial correlation, heteroscedasticity, and normality using standard diagnostic tests.

Furthermore, the outcomes of tests for Granger causality and asymmetric effects are confronted with the generalized impulse responses functions (GIRF) applied to assess the impact of GDP shocks on gender employment rate (unlike standard approach the GIRF are not sensitive to the ordering of variables in the VAR system). Impulse responses show how the variable of interest responds along a specified time horizon to a one standard deviation shock in another variable in a given moment, and whether the effect of shock persists or dies out quickly. All calculations are performed in the EViews 11 software.

4. Results

4.1. Effects of Changes in GDP on Fluctuations in Gender Employment Rate

The estimated VAR models—for different countries and sectors—satisfy the stability condition (no root lies outside the unit circle) in all cases and passed most of the multivariate diagnostic tests, including the Breusch-Godfrey LM test for serial correlation, the White test for heteroscedasticity, and the Jarque-Bera normality test (reported in Zenodo, see link at the end). Therefore, the models are statistically reliable.

According to Wald test results reported in

Table 1. Causality test results—Wald chi-squared statistics (H₀: ∆GDP shocks do not impact female/male employment rate)—linear specification.

	Construction		Education		Accommodation
Countries	Females	Males	Females	Males	Females	Males
Germany	0.25 (0.99)	1.88 (0.93)	6.35 (0.38)	5.72 (0.46)	2.95 (0.57)	3.79 (0.44)
Poland	5.17 (0.52)	4.91 (0.3)	4.66 (0.32)	5.26 (0.51)	12.3 (0.02) *	12.1 (0.02) *
Portugal	8.5 (0.04) *	10.5 (0.02) *	2.74 (0.84)	3.93 (0.69)	6.5 (0.37)	8.83 (0.18)

Note: in brackets are p-values; * denotes the significance at least at the 10% level.

(linear specification of VAR model), the null hypothesis that changes in GDP do not impact changes in gender employment rate is rejected in the case of the construction sector in Portugal and Poland and the education sector in Germany and Portugal (only male), whereas, in the asymmetric specification of VAR model (

Table 2. Causality test results—Wald chi-squared statistics (H₀: $\mathsf{\Delta }GD{P}^{up}$ /$\mathsf{\Delta }GD{P}^{down}$ shocks do not impact female/male employment rate) and asymmetry test results—asymmetric specification.

Causality: $\mathsf{\Delta }\mathit{G}\mathit{D}\mathit{P}$$\text{→} \mathsf{\Delta }\mathit{E}\mathit{R}$	Construction		Education		Accommodation
	Females	Males	Females	Males	Females	Males
Germany
$\mathsf{\Delta }GD{P}^{up}$	6.73 (0.15)	4.04 (0.4)	9.2 (0.06) *	8.68 (0.07) *	15.8 (0.003) *	30.7 (0.00) *
$\mathsf{\Delta }GD{P}^{down}$	7.07 (0.13)	1.3 (0.86)	12.5 (0.01) *	10.7 (0.03) *	7.23 (0.1)	9.78 (0.04) *
Asymmetric effects	5.42 (0.03) *	0.19 (0.67)	4.16 (0.05) *	3.85 (0.06) *	5.86 (0.02) *	21.8 (0.000) *
Poland
$\mathsf{\Delta }GD{P}^{up}$	5.69 (0.22)	3.77 (0.29)	5.54 (0.14)	5.67 (0.13)	10.54 (0.02) *	8.04 (0.05) *
$\mathsf{\Delta }GD{P}^{down}$	2.4 (0.66)	1.02 (0.8)	4.5 (0.21)	5.28 (0.15)	0.37 (0.95)	3.42 (0.33)
Asymmetric effects	3.38 (0.07)	1.79 (0.19)	3.13 (0.08)	3.73 (0.06)	0.08 (0.78)	0.22 (0.64)
Portugal
$\mathsf{\Delta }GD{P}^{up}$	1.31 (0.52)	2.53 (0.28)	8.15 (0.02) *	8.8 (0.03) *	12.7 (0.012) *	10.2 (0.04) *
$\mathsf{\Delta }GD{P}^{down}$	6.23 (0.04) *	6.31 (0.04) *	5.62 (0.06) *	8.0 (0.05) *	11.8 (0.02) *	8.21 (0.08) *
Asymmetric effects	9.19 (0.004) *	7.75 (0.01) *	0.53 (0.47)	0.03 (0.86)	0.6 (0.44)	0.1 (0.76)

Note: in brackets are p-values; * denotes the significance at least at the 10% level.

) the significant impact of GDP on both female and male employment rates is found for upward and downward changes in GDP in education in Germany and the accommodation sector in Portugal. Additionally, upward changes in GDP affect significantly gender employment rates in accommodation in Germany and in construction in Portugal, while downward changes in GDP influence only female employment rate in construction in Poland. Moreover, asymmetric effects of upward and downward changes in GDP on gender employment rates are found in education in Germany and Poland and in accommodation in Germany (Table 2). The results for construction sector in Germany and accommodation sector in Poland indicate no causality between GDP shocks and gender employment rate. To further examine the results obtained above, the impulse response functions are also analysed below.

4.2. Responses of Employment Rate to GDP Shocks—Impulse Response Functions

Based on the estimates of Equation (1) for linear (symmetric) and non-linear (asymmetric) specifications, we can obtain the generalized impulse response functions (GIRF) of gender employment rate to upward and downward shocks in GDP (Figure 4, Figure 5 and Figure 6). These reflect the responses of gender employment rate to one standard error innovation of GDP shocks. The horizontal axis in all figures of impulse response function stands for a time horizon in quarters and the vertical axis is a plot of response impulse, i.e., it is a measurement of the magnitude of the response of variable in interest to a shock in another variable. We can thus graph the responses for each country and three sectors (construction, education, and accommodation and food services).

There are interesting general patterns in construction (Figure 4) which relate to employment rate and strength of the impact of GDP shocks. First, for all three countries the male employment responses to both upward and downward GDP shocks are very small and generally insignificant (only in Portugal are the employment adjustments to upward shocks significant, see Table 2). The male results do not fit with the existing narrative that men’s employment is more sensitive to GDP shocks and hence especially during recessions a fall in employment is expected to occur. A similar conclusion that men’s jobs are more likely to be protected in the recession and then job cuts focus on female occupations but not on male ones can be found in [63]. This may suggest that the initial “mancession” that have characterized the Great Recession has been becoming a “womancession” [64]. The main reason for that seems to be the shortage of skilled workers in all countries under review; due to the construction boom in Germany continuing since several last years and increased economic activity in the construction sector in Poland and Portugal it is hard to find the right workers. Hence, to operate to full potential these economies have been propped up by immigrants (mainly from Ukraine and South Asia in Poland, from Ukraine and Poland in Germany, and Lusophone countries in Portugal). As a result, there is still an increased demand for construction workers which impacts the wage growth. Moreover, this surprising result of insignificant responses both in female and male employment in Germany may be a manifestation of the specificity of German labour market which has demonstrated a robust performance after the 2008 Great recession, mainly due to large degree of working time flexibility and government interventions in the labour market by subsidizing labour hoarding through short-time work. This finding for Germany can be also treated as a manifestation of GDP-employment decoupling, i.e., when changes in GDP caused hardly any reaction in employment [65]. However, in explaining these surprising findings of male employment responses we should not overlook the undeclared work (not fully captured by official statistics) which is still the case in all countries but with a different intensity. The scale of undeclared work varies across countries under review, i.e., 4.4% in Germany, 6.6% in Portugal, and 20.8% in Poland in terms of labour input [66]. While undeclared work occurs in several economic sectors, it seems to be particularly widespread in the construction and also hotels and restaurants. Therefore, the potential adjustments of this part of construction workers to cyclical movements of GDP are hidden.

Second, the stronger responses to shocks in GDP occur in female employment for all three countries (Figure 4), which is in the marked contrast to very small male responses. Female employment exhibits a reduction (although insignificant) in response to upward GDP shocks, since first quarter in Germany and both since the third quarter in Portugal and Poland. Hence, the response pattern of female employment seems to be counter-cyclical, i.e., the female responses are negative in contrast to male ones exhibiting tiny and insignificant response in expansionary period, and positive (but significant only for Portuguese women, Table 2) in recessionary period. Such employment adjustments in expansions may indicate that men actively working on construction sites are in higher demand than women mainly working in sales and office positions, which is especially the case when a new build starts. Generally, this demonstrates that sex segregation is the main explanation of the gendered impact of the business cycle.

The response of German females might be also interpreted in terms of the sex segregation hypothesis as, in a sector like construction where women are under-represented, men’s jobs are more likely to be protected in the recession and then job cuts might have been focused on female occupations but not on male ones, all the more in the case of shortage of skilled workers. The female results in Germany might be treated as consistent with worker discouragement and with downward employment response to downward GDP shocks; however, from the fifth quarter onwards this response starts to be positive, as for Portuguese and Polish women. Interestingly, female employment participation in Germany exhibits a significantly asymmetric response, i.e., participation adjustments to negative shocks are larger than the adjustments to positive shocks, although the impact of positive and negative shocks in GDP changes is insignificant (Table 2).

The general pattern with regard to the education sector (Figure 5), which is over-represented by women in each country under review, is that the responsiveness of both female and male employment participation is significant in expansionary and recessionary periods in Germany and Portugal, but not significant in Poland. The female responses are much less than male ones (except in Portugal). This is not surprising, as the education in large part belongs to the public sector which is not strongly affected by the business cycle and is rather identified as “recession-proof”, ensuring job security and sheltering from some of the worst symptoms of economic downturn. This is the reason why it is still an attractive option for women in spite of relatively lower pay levels than in other sectors. This finding refers mainly to the public education sector; jobs in the private sector are more vulnerable during recessions as the market declines [67]. Nevertheless, the female and male participation responses differ across countries under review.

The German female participation responses (Figure 5) in education are positive and slowly declining, while German male participation reactions are negative and dying out on both upturns and downturns. The results for women indicate that there is considerable stability in employment (mainly as teachers) and economic downturns barely have an impact on female attrition. This means that the concentration of women in education acts as a protection to the recession. Male employment is more hit by job destructions which can be explained by the strong bargaining power of women resulting from their high dominance; as a consequence, the male workers are more likely to be in the front line to lose their jobs. These findings might suggest that sex sectoral segregation has shielded the women’s employment more than men’s [1,68]. However, there are other factors of key importance in addition to pay levels that may affect male decision to quit teaching, such as an increase in the weekly working hours and hence workload, and the growing stress connected to the increased number of students per classroom. These non-pecuniary factors linked to the austerity plan in times of recession may impel the men to leave teaching. The results for Germany (Table 2) provide some evidence of significantly asymmetric participation adjustments, i.e., upward female participation responses to upward GDP shocks are larger than to downward GDP shocks, while downward male participation responses to upward GDP shocks are larger than responses to downward GDGP shocks.

The response pattern of employment rate in education in Portugal displays major differences from those in Germany in the recessionary period (Figure 5). Females exhibit a significant fall in employment rate (since the third quarter) as a response to downward shock which contrasts with significant positive adjustment to the upward shock. For the equivalent males the employment participation response to downward shocks is significantly positive, while to upward shocks it is negative. These findings provide no evidence of significant asymmetry (Table 2). Both the female and male results, unlike those in Germany and Poland, do not fit well with the narrative that due to the dominance of women in education the women are protected more than men during recessions. It might seem that women in Portugal are more sensitive to the discouraged worker effect because of the lesser importance that has been historically attributed to their employment, and also the social norms supported by the breadwinner model. There is also possibility that women leave the labour force for family reasons in the times of recession. In addition, women seem more likely to be affected by the severe spending cuts in public sectors (e.g., “wage moderation” by freezing the minimum wage and reducing overtime pay) taken to reduce the government deficit and debt after the 2008 crisis hit the Portuguese economy much stronger than German and Polish economies. Hence, protection offered by the public sector to mainly female workers has been eroded by the austerity policy.

In the case of Poland, the participation rates of both females and males show no significant reactions to GDP shocks in contrast to their German and Portuguese equivalents (Table 2). In general, these responses are rather in line with notions of maintaining the dominance of women in education, however at the cost of being trapped in low paid jobs, as the changes in public pay sector are rather sluggish. What is puzzling is the apparent but insignificant response of male employment participation to upward GDP shocks which should be rather associated with the problems that education in Poland has been experiencing for several years: demographic ageing which is particularly acute in the case of teachers, and a big wave of teacher retirements (mainly women) resulting in serious shortage of staff that has been partly reduced by the increase of male teachers observed since 2017.

The gender employment adjustments to upward and downward GDP shocks the inaccommodation sector turn out to be significant in the case of all three countries (except Poland in recessionary period), but the responsiveness pattern differs across countries (Figure 6). First, the response of German female participation to upward GDP shocks is positive, while the response of German male participation is negative—Figure 6. In the recessionary period the responses of both female and male participation are negative but female adjustments to downward GDP shocks become positive from the fifth quarter onwards. The clearly positive reaction of women’s participation seems to be related to the sector-specific form of contracts due to high seasonality of this sector, i.e., first, the low-paid, part-time jobs are occupied, mostly by foreign workers (cleaning and serving food in hotels and restaurants), and then full-time jobs but rarely in managerial positions. Increasingly, tasks such as cooking—which are among the primary domestic duties of women—are performed by men who are classified as more skilled than women. This is clearly the manifestation of gender segregation in this sector: men and women are differentially valued, classified, and rewarded. One possible explanation for that may be the undeclared workers who are likely to be first fired during recessions. It is supposed that the share of undeclared work may widen [69] after the introduction of a nationwide minimum wage in 2015. However, further studies are needed to unambiguously confirm it. There is evidence of asymmetric responses in participation for both genders, i.e., upward female employment responses (and downward male responses) to upward GDP shocks are larger than to downward GDP shocks (Table 2).

Second, Portuguese female participation initially exhibits no response to upward GDP shocks, and from the fourth quarter onwards it shows positive reaction. The response of male participation is rather procyclical (after initial positive reaction a fall occurs in the fifth quarter after the upward GDP shocks). In the recessionary period the response of female participation is still positive and slightly declining, but the male employment exhibits a sharp fall. Although for both genders these responses to upward and downward GDP shocks are not significantly asymmetric (Table 2), this pattern of gender responses indicates that male employment is rather more protected in an expansionary period, and female employment—in a recessionary period. The accommodation sector is characterized by lower pay than in male-dominated sectors or even with lower pay for women than men within this sector [70] (e.g., the lower pay of female waiters than male waiters is observed in Portugal); hence, it is cheaper for employers to hire women than men during recessionary times. As a result, this sex segregation contributes to the increase of female employment during recessions but at the same time to the expansion of low paid jobs. Consequently, the pay gap is unlikely to be reduced noticeably within a sector and generally in the whole economy.

Third, the employment rates of both females and males in Poland show different reactions to GDP shocks, in sharp contrast to their German and Portuguese equivalents. The responses of both female and male participation to upward GDP shocks are negative, but male participation adjustments become positive in the sixth quarter. These findings seem surprising given growing demand for employees according to a positive long-term trend in this sector (new hotels and restaurants have been created) resulting from a good economic situation, improvement of both transport and conference infrastructure, development of touristic, changes of lifestyle, and ways of spending free time of Polish society. However, this booming time does not translate automatically to better pay conditions as typical features of this sector are the precariousness of employment and low pay. On the other hand, no positive employment adjustments to GDP shocks may be the effect of omitting the undeclared work in official employment statistics which is particularly high in Poland [66]. These undeclared workers are first in line to lose jobs in recessionary periods, although the magnitude of this adjustment to GDP changes is difficult to evaluate.

5. Conclusions

The situation of women and men in the labour market is diversified not only due to historical, institutional, political, and cultural conditions being considerably different across countries, but also over the business cycle, mainly in terms of labour participation, unemployment, working hours, and wages. Therefore, labour policy (e.g., austerity policy during recessions) should not be implemented in a gender-blind manner. The purpose of this paper is to propose an investigation of the responses of female and male employment rates to upward and downward fluctuations in economic activity across chosen sectors with different gender concentration (construction sector with dominance of men, education sector with concentration of women, accommodation and food services—gender-balanced sector) in three countries (Germany, Portugal, and Poland) representing different types of welfare states (continental, Mediterranean, and mixed respectively). Examining the potential asymmetric behaviour is crucial to evaluate the importance of the discouraged worker effect.

In all countries sex segregation seems to be the explanation of the gendered impact of fluctuations in GDP within all chosen sectors (construction, education, and accommodation and food services). However, men and women appear to be in “protected” or “buffer” jobs based on their job losses within sectors being proportionate or disproportionate to the level of gender representation in a sector.

The findings obtained for all three countries in the construction sector—with overrepresentation of men—show that male employment adjustments to the business cycle fluctuations are very small and generally insignificant (except Portuguese men during downturns in the cycle). The dominance of men and ensuring the bargaining power together with a shortage of construction workers due to the construction boom in all countries seem to protect male jobs during recessions. Hence job cuts are likely to be first focused on female occupations but not on male ones (especially in Germany). It could be that the undeclared work accounts for the lack of male employment response to GDP shocks to the greater extent in Poland and Portugal and to the lesser extent in Germany, but the scale of this phenomenon is difficult to evaluate. In spite of many EU initiatives to reduce the incidence of undeclared work it remains at too high a level having negative impacts on employment and productivity and distorting the level playing field [71].

On the other hand, in the education sector—with a concentration of women—the economic downturns have only barely impacted female employment (except in Portugal); hence, this high dominance of women and strong bargaining power act as a protection to recessions and shelter them from the worst symptoms of economic downturns. However, it is at the cost of being trapped in a lower paying sector. This time the male workers are in the front line to lose their jobs. The Portuguese results in education seem to suggest the lesser protection of female jobs than in the case of the German and Polish equivalents; however, this should be rather explained by severe public spending cuts taken after the 2008 crisis hit the Portuguese economy much stronger than German and Polish economies. Furthermore, it could be that structural factors (e.g., employment structure indicators, globalization, goods market structure, labour market policies, see [72]) are the major determinants of labour force participation in this case and then it may be more difficult to discern meaningful associations with the business cycle.

The results of gender employment responses to GDP shocks in accommodation and food services—with balanced representation of women and men—exhibit different patterns across countries. This seems to result from the country-specific characteristics of tourism industry and seasonality of employment in this sector reflecting customers’ seasonal and leisure demands. Female and male employment in Poland display negative responses to upward GDP movements, although the sector is still growing. The results for Portugal indicate that male employment is initially more preferred during expansions but from the fourth quarter after a shock females are more wanted, while during recessions female employment is in more demand. The results for German suggest that female workers are more advantaged in the expansionary period. Although countries vary with regard to the reaction of gender employment to cyclical movements in GDP, they display similar patterns in terms of pay differentials between women and men. It is commonly acknowledged that men are more valued, their work is classified as skilled and better paid, while women are associated with unskilled jobs (cleaning, serving food), their work is devaluated and hence lower paid [73].

Moreover, our findings indicate that the evidence of discouraged worker effect is rather weak as the asymmetric adjustments—with falls in participation during cyclical downturns being larger than the rises during upturns—occurs in neither country fully. Hence, these results do not point to the phenomenon of “hidden” unemployment, but rather the phenomenon of the involuntary part-time worker which has become more common with an increase in precarious employment [74].

In general, the findings on both female and male employment adjustments to cyclical GDP movements suggest rather that the gender employment segregation is a deeply entrenched feature within the given economic sector in all studied countries independently of the welfare states classification. This means that women still enter low paid occupations/sectors or are crowded into a limited number of occupations resulting in lower wages, while men work in better paid sectors. The stagnation of gender sectoral segregation is one of most significant contributing factors to the preservation of gender pay gap as is being emphasized in the EU reports on gender equality and gender segregation [3,75]. According to the Eurostat data, the average pay gaps in countries under review in 2016 are 22% (Germany), 17.5% (Portugal), and 7% (Poland). However, it should be emphasised that the results in economic sectors may differ considerably from the average. It is symptomatic that despite many years of equal pay legislation, the gender pay gap remains remarkably resilient across all EU countries. Higher growth or greater market integration alone do not appear to help eliminate country’s gender-based employment segregation [43]. Achieving equality in opportunities requires ensuring that we change the social norms and stereotypes that limit the set of choices available both to men and women.

Therefore, combating gender segregation, and hence the gender pay gap, is a burning issue as they affect not only current earnings, but also lifetime earnings and pension entitlements. Within the Strategic Engagement for Gender Equality 2016–2019, the European Commission has planned different actions [3] which, amongst others, include the following: support measures enhancing women’s and girls’ digital skills; promote women’s employment in the ICT sector; raise awareness on educational and vocational training choices; promote gender equality in all levels and types of education, including those connected to gendered study subject choices and careers, using existing policy cooperation tools and funding instruments; promote gender equality at all levels of education and training, as well as in working life; eliminate stereotypes about gender roles in the labour markets and within the household; and reduce gender bias in hiring, workplace cultures, promotions, and evaluations, in order to reduce gender segregation in the labour market.

However, it is difficult to assess the effectiveness of these actions, and the EIGE report highlights that progress is moving forward at a snail’s pace and in some domains is even going backwards. As a result, the EU is still a long way from being a gender-equal society.