The unintended consequences of IMF programs: Women left behind in the labor market

Abstract

Despite societal progress toward achieving gender equality, women remain disadvantaged across all walks of life. We examine how pressures by the International Monetary Fund (IMF) for austerity and liberalization policies affect women’s economic empowerment in developing countries. Focusing on labor market outcomes, we expect that IMF programs increase the gender unemployment gap. This effect will hold regardless of existing societal gender norms, reflecting that women work in less ‘crisis-proof’ jobs. In contrast, we believe that prevailing societal gender norms will adversely impact women’s ability to remain in the labor force, resulting in widening gender labor force participation gaps. Relying on a dataset covering 128 developing countries between 1992 and 2018, we confirm these expectations. Our findings underscore how, during economically turbulent times, policy demands of the Fund can amplify existing gender inequalities.

1 Introduction

Despite growing political awareness and progress toward greater gender equality, women remain disadvantaged across all walks of life. Nowhere else are gender disparities as visible as in women’s labor market outcomes. Women’s labor force participation rates—measuring women’s participation in labor markets—have historically lagged behind their male peers (Iversen & Rosenbluth, 2006; Elias & Rai, 2019; Iversen et al., 2020). When women are gainfully employed outside the household, they earn systematically less than men (Sloane et al., 2021) tend to work in less secure jobs (Blanton et al., 2019) and face significant hurdles when trying to advance their careers (Bertrand, 2011; Armstrong et al., 2023). Whereas the literature seems to agree that the increase in unemployment—attributed to an IMF program—disproportionally hurts women, evidence from numerous country cases is at best mixed on the question of whether IMF programs have undone years of progress in women’s labor force participation. For instance, Assaad and Arntz (2005) document for the case of Egypt that the IMF-induced austerity program—signing off on drastic cuts in public sector employment—resulted in a sharp decline in female labor force participation rate, reversing a thirty-year positive trend of increasing female employment rates. In sharp contrast, Rono (2002) documents the opposite effect for the case of Kenya during the 1990s. Whereas Kenya experienced a sharp spike in female unemployment at the outset of the IMF program, the implemented measures lifted female employment onto a positive growth trajectory.

These observations provide our motivating empirical puzzle. We believe that the uniform effect of women’s rising unemployment during IMF programs can be attributed to the fact that women are primarily employed in more vulnerable jobs (Detraz & Peksen, 2016; Seguino, 2019; Blanton et al., 2019). Although IMF programs seem to produce these uniform effects on women’s unemployment, there are wide disparities concerning women’s labor force participation rates.¹ We argue that differences in IMF program design cannot explain these differences in the labor market outcomes for women but are a result of societal gender norms shaping women’s ability to (re-)enter the labor force. Whereas changes in unemployment rates reflect short-term labor market fluctuations, women’s labor force participation rates reflect long-term structural patterns shaping women’s willingness and, more importantly, ability to actively participate in a nation’s labor force. And indeed, substantial research points out that asymmetric household bargaining dynamics, the availability of social services like childcare, legal provisions protecting women’s economic rights, and women’s education outcomes reflect deeply rooted societal gender norms and biases (Afshar & Dennis, 2016; Goldin, 2014; Donald & Lusiani, 2017; Fruttero et al., 2020). Meanwhile, these factors shape women’s incentives and ability to enter the labor force (Iversen et al., 2020). For this reason, we hypothesize that declines in labor force participation of women relative to men in the course of IMF programs are most pronounced in countries where existing societal norms put women at a disadvantage. Put differently, the IMF’s economic policy prescriptions work ‘unintentionally’ as a powerful catalyst for these underlying societal gender norms, with adverse implications for women’s economic empowerment and gender equality more broadly.

Using a dataset comprising 128 countries between 1992 and 2018, we find support for our hypotheses. To capture gender differences in labor market outcomes, we rely on two measures. To capture the effect of IMF programs on unemployment, we calculate a gender ‘unemployment gap,’ which is the difference in unemployment rates between men and women, as a ratio of the total unemployment rate. We also calculate a gender ‘labor force participation gap’ following a similar approach. The rationale for this modeling choice is that these outcomes capture a differential impact of an IMF program on women.² We find that IMF programs have an unconditional negative effect on the unemployment gap. Women are more adversely affected by unemployment by IMF programs than men. In substantive terms, entering an IMF program increases the unemployment gap by 1.19 points (or 11.5% of a standard deviation) in the short run, which accumulates to 12.63 points (or more than a standard deviation) over the long run. We cannot detect such a pattern concerning the gender gap in labor force participation unless we account for the degree of gender discrimination in societies.³ Importantly, we can detect a negative gender bias in labor force participation rates in countries where existing societal norms put women at disadvantage. We can also verify that our results are not alone driven by IMF program design features, nor do they result from a country’s crisis experience.

To mitigate concerns about competing mechanisms and further explore these results qualitatively, we provide evidence from the selected country cases of Egypt, South Korea, and Colombia during IMF programs (see Section A1). We demonstrate that, although Colombia and South Korea faced similar crises that prompted IMF intervention, the labor force participation of women relative to men deteriorated only in South Korea but not in Colombia—reflecting differences in societal gender norms. The case of Egypt equally demonstrates how discriminating societal gender norms drove women out of the labor force under IMF programs, while unemployment rate movements showed no discernible differences between men and women.

We contribute to several strands of the literature. First, we complement existing work on the adverse impacts of IMF programs (Forster et al., 2019; Stubbs et al., 2020; Peksen et al., 2017). Our work relates to the literature that focuses on IMF programs and labor market outcomes (for a recent survey, see Reinsberg (2019)). Whereas scholars concentrate on the impact of IMF programs on the reform of labor market institutions and its implications for aggregate labor market outcomes (Vreeland, 2002; Lee & Woo, 2021; Chletsos & Sintos, 2023), we focus on differences in employment outcomes between women and men. Complementing earlier research (for a survey, see Afshar and Dennis (2016)),⁴ we demonstrate that the disproportional displacement of women happens independently of existing gender norms. This aligns with substantial literature in labor economics that concludes that women work in positions and professions that make them more susceptible to displacement during economic downturns (Bertrand, 2011; Duflo, 2012; Goldin, 2014; Klasen, 2019; Kabeer, 2020). In addition to studying unemployment gaps between men and women, we analyze differences in labor force participation that likely reflect deeply-rooted gender norms keeping women out of the labor force (Iversen & Rosenbluth, 2006; Bertrand, 2011; Peksen, 2019; Klasen, 2019). In line with prior literature, our findings underscore the importance of societal gender norms for women’s labor force participation rates.

Second, we contribute to the literature studying the gender dimension of international political economy (Sen, 1996; Hutchinson et al., 2018; Elias & Rai, 2019; Iversen et al., 2020; Betz et al., 2021; Betz et al., 2023; Armstrong et al., 2023). Our work is closely related to recent scholarship emphasizing the importance of studying the distributional consequences of economic policies on women’s livelihoods (for a survey, see Demir and Tabrizy (2022)). We follow in this tradition and expand on existing work within the subfield of studying international financial markets (Prügl, 2012), monetary institutions (Capie & Wood, 2019; Bodea et al., 2021; Masciandaro et al., 2023), and structural adjustments (Detraz & Peksen, 2016; Bruff & Wöhl, 2016; Afshar & Dennis, 2016). Here, our focus is on the IMF. Although the IMF has been paying increasing attention to gender-related challenges in adjustment programs (Çağatay & Özler, 1995; Elson, 2013; Afshar & Dennis, 2016; Peksen, 2019),⁵ quantitative evidence capturing the gender footprint of IMF-sponsored programs is scant (for a recent survey, see Reinsberg et al. (2024)). We expand this literature and uncover the mediating role of societal norms shaping the effect of IMF programs on women’s labor market outcomes.

Finally, we contribute to an important policy debate on the distributional consequences of IMF programs. Despite advances in incorporating concepts such as gender budgeting into lending frameworks (Coburn, 2019), these innovations seem to fail to protect women. Built on institutional foundations forming the breeding ground for widespread gender disparities, existing labor market structures and work realities—independent of existing societal gender norms—are prone to disadvantage women during economic hardship. At the same time, none of the IMF’s current instruments is in a position to uproot these existing labor market realities and mitigate the adverse amplifying effects of its programs.

2 Theoretical considerations

Since IMF lending operations started, the IMF has, to a varying degree, attached conditions when it provided a helping hand (Bird, 2007; Breen, 2013; Dreher et al., 2015). Built around general balance-of-payments considerations, a uniting feature of IMF programs is their contractionary nature, which can disproportionally impact marginalized and vulnerable groups in society (Çağatay & Özler, 1995; Aslanbeigui & Summerfield, 2000; Detraz & Peksen, 2016; Li et al., 2015; Kentikelenis & Babb, 2019).⁶ Despite a ‘short-run-pain-long-run-gain’ narrative, an abundance of evidence underscores the potential scaring effects of austerity measures on a borrowing country’s labor market—a side effect of the IMF treatment (Vreeland, 2002; Agnello et al., 2014; Rickard & Caraway, 2019; Chletsos & Sintos, 2023). An implicit assumption in this literature is that women and men are affected similarly. Relaxing this assumption, numerous studies find that the adverse labor market effects are more pronounced for women than men (Elson, 2013; Detraz & Peksen, 2016; Donald & Lusiani, 2017). Besides contractionary macroeconomic policies aimed at restoring the balance of payments, researchers have assigned liberal market reforms in the course of IMF programs a prominent role in hurting women’s labor market outcomes (Detraz & Peksen, 2016; Berik, 2017; Seguino, 2019). For example, Detraz and Peksen (2016) provide systematic cross-country evidence for increased women’s unemployment rates during IMF programs.⁷ Nevertheless, it remains unclear to what extent IMF programs are driving these patterns.

We believe that the increase of women’s unemployment reflects an uncomfortable reality in global labor markets: women work in less stable and ‘crisis-proof’ jobs. Despite advances concerning a greater inclusion of women in labor markets, numerous studies document that women are more prone to lose their jobs during times of economic hardship (Elson, 2010; Blanton et al., 2019; Blundell et al., 2020; Coibion et al., 2020). In fact, a substantially larger share of women is employed in part-time jobs and depends on precarious forms of employment to sustain their livelihoods (Fruttero et al., 2020). An abundance of evidence concerning women’s labor market status in developing countries documents that women constitute the lion’s share of workers in low-paying and vulnerable jobs in the informal sector (Duflo, 2012; Klasen, 2019; Armstrong et al., 2023). Even in the most progressive countries, women are more frequently employed in less secure jobs (Bertrand, 2011; Klugman et al., 2014; Goldin, 2014; Blanton et al., 2019). For instance, Blanton et al. (2019) provide evidence that women are first in line to lose their jobs during an economic crisis. Recent evidence on the impact of the pandemic on local labor markets confirms an existing gender bias in layoffs (Blundell et al., 2020; Coibion et al., 2020; Bednarzik et al., 2021). Furthermore, women more frequently find employment in the public sector or the services sector which are closely linked to the viability of national budgets (Çağatay & Özler, 1995; Rubery, 2015; Donald & Lusiani, 2017; Fruttero et al., 2020). The ILO arrives at the sobering conclusion that “when women are employed, they tend to work in low-quality jobs in vulnerable conditions.”⁸

Built on contractionary macroeconomic policies and structural adjustment measures that lead to a short-run contraction in economic activity and a tightening of public budgets, IMF programs have the potential to amplify these initial adverse effects on women (Detraz & Peksen, 2016; Berik, 2017; Seguino, 2019). For example, Donald and Lusiani (2017) report for the case of Ukraine that the government let go of 165,000 women who were performing lower-level public sector jobs to fulfill IMF loan requirements. Besides austerity measures, structural adjustment measures such as privatizing state-owned enterprises can lead to mass layoffs (Brainerd, 2000; Çağatay & Erturk, 2004). For instance, analyzing the privatization of public sector firms in Egypt under the auspices of the IMF, Çağatay and Erturk (2004) illustrate these asymmetric displacement patterns. We expect these effects to be uniform across countries. We argue that this selection of women into these jobs leads to widening unemployment gaps between men and women during an IMF program. We synthesize these insights into our first hypothesis.

Hypothesis 1

An IMF program will increase the gender unemployment gap, irrespective of existing societal gender norms.

Despite this uniform distribution of IMF program effects on the gender unemployment gap, we expect heterogeneous program effects on the gender gap in labor force participation. Several studies hint at diverging impacts of IMF programs on women’s labor force participation rates (Rono, 2002; Assaad & Arntz, 2005; Donald & Lusiani, 2017). For instance, Assaad and Arntz (2005) document for the case of Egypt that the IMF-induced austerity program—signing off on drastic cuts in public sector employment—resulted in a decline in the women’s labor force participation rate. In sharp contrast, Rono (2002) documents an opposite effect in the case of Kenya during the 1990s, where IMF-induced reforms led to an increase in women’s labor force participation. We argue that there are multiple reasons for these divergent outcomes.⁹

First, a combination of austerity measures and structural reforms towards international (financial and trade) openness can lead to shifts in sectoral employment opportunities, adversely impacting women (for a discussion, see Afshar and Dennis (2016)). For instance, several authors document in the case of South Korea how increased financial and trade openness—which was an essential component of the IMF-supported adjustment program—drove firms to shift part of their value chains to low-wage countries in South Asia to cut costs and retain their international competitiveness (Jones, 2006; Peng, 2011; Chae, 2016). As a result, women employed in firms that decided to outsource labor-intensive production processes lost their employment permanently, forcing them either to exit the labor force or to pursue precarious forms of employment (Chae, 2016).¹⁰ We detect a similar pattern in the case of Egypt, where women were disproportionally employed in low-profile jobs in the public sector, which fell victim to budgetary cuts and privatizations during the IMF adjustment program, leading to a ‘defiminization’ of economic activity (Moghadam, 2005). Although women could have theoretically returned to the labor force, low wages (and high opportunity costs), substantial gender wage gaps, and traditional family models, in combination with little legal workplace protection, squeezed women out of the labor force. Due to discriminatory societal gender norms, women had little or no legal workplace protection from gender-based violence (Moghadam, 2003; Moghadam, 2005; Hammad, 2017). Hammad (2017, 52), providing a comprehensive statistical account for the prevalence of gender-based violence in Egypt, reports that based on a representative survey “52% of the women recognized that women are likely to be harassed in the workplace,” deterring women from entering the labor force.

Second, it is well established that women are disproportionally impacted by budgetary austerity measures (Brah et al., 2015; Rubery, 2015; Afshar & Dennis, 2016; Bruff & Wöhl, 2016). Women frequently secure paid positions in the public sector, which are more likely to vanish as a result of IMF austerity measures that demand cuts in public sector employment (Afshar & Dennis, 2016). Societal gender norms play an important role in implementing budgetary cuts during IMF-supported adjustment programs (Rubery, 2015; Reinsberg et al., 2024). In this respect, Reinsberg et al. (2024) show that progressive societal gender norms can have a first-order impact on keeping women in public sector employment, cushioning the adverse impact of IMF-supported austerity measures and underscoring the importance of societal gender norms during the implementation phase of IMF programs. Besides this direct effect, austerity measures frequently impact social services, driving up women’s work and time commitment for household chores and limiting their ability to seek employment outside the household (Stewart, 2016).¹¹ Prime among these public services is childcare, which is often publicly funded (Brah et al., 2015; Stewart, 2016). For instance, in the case of Romania’s IMF adjustment program in 2008, the government cut financial support for childcare facilities, leading to a sudden drop in women’s labor force participation rates (Popescu et al., 2016). Furthermore, austerity measures frequently imply that public financial support for women’s social and political initiatives is reduced or cut entirely, putting a stop-gap on women’s advocacy efforts and limiting their ability to advocate against the gendered impacts of budgetary austerity measures (Brah et al., 2015). These adverse effects on women’s labor market outcomes can be expected to be even more severe when discriminatory gender norms are deeply ingrained in society.

Finally, austerity measures frequently imply a reduction in price subsidies of food and other household items (Afshar & Dennis, 2016; Stewart, 2016; Rickard & Caraway, 2019). In some instances, governments introduce value-added taxes on consumption goods. These measures have direct and indirect negative repercussions on women and their ability to enter the labor force. For instance, Stewart (2016) argues that women spend more time preparing food and running errands as households shift from processed foods to meals that need longer preparation. In addition, given that price increases might not be implemented uniformly, women spend more time searching for the most efficient ways to spend household budgets (Stewart, 2016). As discriminatory societal gender norms infuse a stark asymmetry into household-level bargaining dynamics (Becker, 1993; Klasen, 2019; Bue et al., 2022),¹² these additional tasks will be assigned to women, who will have less time at their disposal to look for work outside the household while managing diminishing household budgets, placing a disproportionate burden on women’s shoulders (Çağatay & Özler, 1995; Elson, 2010; Donald & Lusiani, 2017; Forster et al., 2020).

Thus, the IMF’s adjustment measures can unintentionally function as a powerful catalyst for discriminating societal gender norms, keeping women out of the labor force. Against this background, we argue that in countries where societal gender norms disadvantage women, an IMF program will have a more pronounced negative effect on women’s labor force participation. Summarizing these insights, we formulate our second hypothesis.

Hypothesis 2

An IMF program increases the gender labor force participation gap, especially under circumstances of adverse societal gender norms and legal provisions putting women at a disadvantage.

While our attention is directed toward labor market outcomes, we acknowledge that IMF programs can adversely affect women’s livelihoods in other ways. These adverse effects can be plausibly linked to women’s job loss as a result of these programs. Beyond the emotionally and psychologically challenging aspect of job loss (Bednarzik et al., 2021), the consequences of IMF programs are anticipated to elevate stress levels (Daoud & Reinsberg, 2019), impose economic hardship on families (Stewart, 2016), and place an additional burden on women (Brah et al., 2015; Detraz & Peksen, 2016). Nevertheless, the magnitude and, thus, the interpretation of our findings depend on the alternative paths available to women.

In an optimistic scenario, unemployed women may pursue higher education, while a pessimistic scenario may lead them into informal job settings with limited labor protections or prompt temporary withdrawal from the labor market for unpaid care work at home. These precarious forms of employment expose women to higher income and health risks (Kentikelenis, 2017; Forster et al., 2020). In addition, in times of economic decline and male job losses, there is a potential risk of heightened domestic violence against women (Elias & Rai, 2019). In addition to experiencing direct or indirect job loss and precarious working conditions when women find re-employment in the informal sector, these changes in social and household dynamics are likely to infuse substantial instability and uncertainty into households. In addition, access to health services might be constrained due to cuts in health budgets. Taken together, we expect these developments to translate into elevated stress levels among women and declining fertility rates (Kubrin et al., 2022).”¹³ This heightened stress can potentially lead to negative mental health outcomes, such as depression, contributing to a rise in the incidence of women’s suicides (Goulas & Zervoyianni, 2016; Kubrin et al., 2022). Consistent with these arguments, for the case of South Korea, Hong et al. (2010, 1626) argue that the IMF-adjustment program “resulted in income disparities and a major breakdown of the social fabric, which consequently exacerbated social problems, such as crime, suicide, marital discord, and wife battering.” Rather than formulating explicit hypotheses, we empirically verify the adverse consequences of IMF programs on women’s livelihoods and wellbeing in further analyses.

3 Empirical analysis

3.1 Data and methods

We use multivariate regression analysis on panel data to test our hypotheses. Given our substantive interest in the gendered impacts of adjustment programs in developing countries, we include countries below the threshold for high-income countries at $12,696.¹⁴ Due to missing data, our effective sample includes up to 128 countries between 1992 and 2018. We focus on labor force participation and unemployment as our two key outcome variables. They capture important dimensions in the economic empowerment of women (World Economic Forum, 2023). Our preferred specifications directly model the ‘gender gap’ in these outcomes such that lower values imply a lack of women’s empowerment (or more unequal outcomes).

First, we calculate the ‘unemployment gap’ as the difference in unemployment rates between men and women, as a ratio of the total unemployment rate. Formally, we calculate $y^U=100 \frac{U_M-U_F}{U}$. Note that we subtract the women’s value from the men’s value, given that unemployment is a negative outcome. The unemployment rate indicates the number of unemployed people as a percentage of all economically active people in the labor force. Data are available from the World Development Indicators (WDI, 2021).¹⁵

Second, we calculate the labor force participation gap (‘LFP gap’) as the difference in labor force participation between women and men as a ratio of the total labor force participation rate. Formally, we calculate $y^P=100 \frac{P_F-P_M}{P}$. The labor force participation rate indicates the percentage of all people of working age who are employed or are actively seeking work. Data are available from the World Development Indicators (WDI, 2021).

The benefit of modeling the gap is that it is a close representation of our theoretical concept of gender inequality. The drawback is that it does not allow us to infer if a worsening gap is due to better treatment of men or worse treatment of women. To parse out these differential effects, we also show results on key outcomes separately for both genders in the supplemental appendix. Furthermore, following the above methodology, we calculate gender gaps concerning shares of employment in the informal sector of the economy, enrollment in tertiary education, and the incidence of suicides. All measures are available through the World Development Indicators (WDI, 2021).

3.2 Key predictors

Our main predictor is a binary variable indicating whether a country is under an IMF program. To account for program heterogeneity, we also draw on a count of the total number of binding conditions, which includes prior actions, quantitative performance criteria, and structural performance criteria. Both pieces of information are available from the IMF Monitor Database (Kentikelenis et al., 2016).

Two contextual measures allow us to account for cross-country variation in the effect of IMF programs on the gender labor force participation gap. The first captures pre-existing societal gender norms, which we measure through the Public Gender Egalitarianism index (Woo et al., 2023). This measure is based on latent variable analysis of a host of cross-national survey questions on gender equality norms in over 100 societies over 50 years. Surveys elicit the extent to which respondents agree that “both husband and wife should contribute to household income” and disagree that “men make better business executives than women do,” among other issues.

Second, we use the ‘Women, Business, and the Law’ (WBL) index, constructed by the World Bank (WBL, 2018), to identify contexts in which the level of de jure protection of women’s economic rights is low.¹⁶ Using the WBL index, we can test whether the relationship between IMF programs and the labor force participation gap is significantly different when women lack legal, economic empowerment rights.

3.3 Control variables

We include several control variables that are most likely to confound the relationship between IMF programs and gender outcomes. We contend that economic crises and weak institutions are the most likely confounders because they may make it more difficult for countries to promote gender equality while increasing the probability of IMF assistance at the same time.

First, we include the (logged) GDP per capita, given that richer countries have a greater capacity to withstand economic shocks and promote gender equality. Second, we include the (log-transformed) rate of inflation, as well as the rate of economic growth. Both variables capture deteriorating economic circumstances. These variables are drawn from the World Development Indicators (WDI, 2021). We also include a binary indicator for financial crisis (Laeven & Valencia, 2013). Financial crises often trigger IMF programs. Finally, we include the Polity index, capturing institutional democracy (Marshall et al., 2015). Democratic countries are more likely to promote women’s rights (Beer, 2009), but the process of democratization often triggers instability that may provide incentives for countries to turn to international organizations and undergo international commitments to enhance their credibility (Mansfield & Pevehouse, 2006).

In robustness checks, we employ enlarged sets of control variables to ensure that our results are not driven by differences in IMF program design, changes in de jure labor institutions, omitted changes in government preferences toward gender equality, and regime breakdown and inter-state war. We will introduce the corresponding variables in the results section. We present detailed definitions and descriptive statistics for key variables in Table A1 and a correlation matrix in Table A2. For brevity, we relegated this table and subsequent robustness checks to the supplemental appendix.

3.4 Empirical models

As our outcome variables are normally distributed, linear models are most appropriate for inference. We initially proceed under the assumption that IMF programs are exogenous once we control for the confounding effect of underlying economic fundamentals and crisis variables. There is little reason to believe that policymakers would undergo an IMF program with the goal of affecting gender-related outcomes. Neither should decisions about IMF program lending be dependent on gender-related outcomes. To remove country-specific heterogeneity, we include country-fixed effects. We also include year effects to account for common global shocks.

An important methodological issue that is often overlooked concerns the appropriate specification of our model. For example, non-stationarity of the dependent variable could induce bias unless there were non-stationary predictors with which the dependent variable had a co-integrating relationship. Political methodologists often refer to the need for ‘equation balance’ as a criterion for valid inference (Enns & Wlezien, 2017). To guide model choice, we follow the auto-regressive distributed lag bounds testing procedure, detailed in the appendix (Pesaran et al., 2001). Our diagnostic tests suggest that the dependent variables are non-stationary, suggesting the need to use the outcome variables in the first differences. Moreover, we find evidence of co-integration, suggesting that an error-correction model is appropriate for our analysis (Appendix A3).

Diagnostic tests aside, we also believe that an error-correction model is the most conservative choice as it contains the simpler models obtained by imposing parameter restrictions. In fact, political methodologists have long advised the use of general models like the error-correction model, as they are more robust against stationarity problems and serial correlation than more restrictive specifications (De Boef & Keele, 2008; Beck & Katz, 2011; Strobl et al., 2023; Asai et al., 2023). An important substantive benefit of using the error-correction model is that it can distinguish between short-term effects and long-term effects. This versatility comes at the cost of a less straightforward interpretation of effects.

For our main analysis, we opt for error-correction models as suggested by diagnostic tests and our interest in capturing the dynamics of adjustments in our outcomes. Formally, our estimating equation—here for the example of the unemployment gap—reads as follows:

$$\begin{aligned} \Delta y^U_{it} = \beta _0 y^U_{i,t-1} + \beta _{11} \Delta IMF_{it} + \beta _{12} IMF_{i,t-1} + \Delta {\textbf {X}}_{it} \mathbf {\beta }_{21} + {\textbf {X}}_{i,t-1} \mathbf {\beta }_{22} + u_i +v_t +\varepsilon _{it} \end{aligned}$$

(1)

In additional tests, we show the results from simpler models that only use the lagged levels of the predictors and first-differenced outcomes. For the example of the unemployment gap, the estimating equation reads as follows:

$$\begin{aligned} \Delta y^U_{it} = \beta ^*_0 y^U_{i,t-1} + \beta ^*_{11} IMF_{i,t-1} + {\textbf {X}}_{i,t-1} \mathbf {\beta }^*_{22} + u_i +v_t +\varepsilon _{it} \end{aligned}$$

(2)

Endogeneity would arise if variables that correlate with gender outcomes and IMF programs were omitted. To address such endogeneity, we employ an instrumental-variable design. We use a Bartik-style instrument that exploits plausibly exogenous time-varying common shocks and multiplies these shocks with (possibly endogenous) exposure shares varying across countries (Borusyak et al., 2022). We believe this assumption is plausible in our context, which differs from other contexts in which the use of shift-share instruments is based on exogenous exposure shares and endogenous common shocks (Goldsmith-Pinkham et al., 2020).

Our shift-share instrument is the multiplicative interaction between the number of countries under programs and the long-run probability of a country being under IMF programs (Forster et al., 2019; Stubbs et al., 2020; Nelson & Wallace, 2017). The instrument is relevant because when global economic circumstances are dire—as proxied by the number of countries under programs—more vulnerable countries with higher propensities of IMF programs will be more likely to go under IMF assistance again. The instrument is plausibly excludable because it only identifies the differential effect of changes in the global economy on the probability of undergoing an IMF program between regular borrowers of IMF loans and irregular borrowers of IMF loans. The exclusion restriction is not violated as long as there is no omitted variable affecting gender gaps in regular borrowers differently than in non-regular borrowers, which we consider unlikely (Lang, 2021; Nunn & Qian, 2014; Stubbs et al., 2020).¹⁷ Formally, we estimate the following system of equations using maximum likelihood:

$$\begin{aligned} \begin{array}{ll} \Delta y^U_{it} &{}= \beta _0 y^U_{i,t-1} + \beta _{11} \Delta IMF_{it} + \beta _{12} IMF_{i,t-1} + \Delta {\textbf {X}}_{it} \mathbf {\beta }_{21} + {\textbf {X}}_{i,t-1} \mathbf {\beta }_{22} + u_i +v_t +\varepsilon _{it} \\ IMF_{i,t-1}&{}= \alpha _0 y^U_{i,t-1} + \alpha _1 z_{i,t-1} + \Delta {\textbf {X}}_{it} \mathbf {\alpha }_{21} + {\textbf {X}}_{i,t-1} \mathbf {\alpha }_{22} + f_i + g_t + \epsilon _{it} \end{array} \end{aligned}$$

(3)

Table 1 Unemployment Gap refers to the gap between men and women concerning the unemployment rate. LFP gap is capturing the gap between men and women concerning the labor force participation rate. Linear regression with country-fixed effects and year-fixed effects. Outcome variables are in first differences. Country-clustered errors in parentheses. The third column for each outcome includes a selection equation for IMF programs using the interaction between the number of countries under programs and the probability of an IMF program as an instrumental variable. The F-statistic helps assess instrument relevance

4 Results

4.1 IMF programs and gendered labor market outcomes

Table 1 presents three models for each of our two gender outcomes. On the left-hand part of the table, we find that IMF programs have an unconditional negative effect on the unemployment gap. In other words, women are more adversely affected by unemployment by IMF programs than men.¹⁸ The effects are most pronounced in the second model with full controls. Substantively, considering the short-term coefficient, an IMF program is related to a deterioration in the unemployment gap by 1.188 (95%-CI: 0.024-2.352)—equivalent to 11.5% of its standard deviation.¹⁹ Considering the long-run multiplier, an IMF program is related to a cumulative deterioration in the unemployment gap by 12.631 (95%-CI: 1.928-23.334)—equivalent to more than a standard deviation.²⁰ If interpreted causally, these effects would be economically significant. The long-run effect of an IMF program on the unemployment gap would be similar to the unemployment gap between Argentina ($y^U_{ARG}=-25.1)$ and Portugal ($y^U_{PRT}=-12.4$) in the most recent sample year. In the last model, we confirm that such a causal interpretation is warranted. Controlling for selection into IMF programs, we find that entering an IMF program adversely affects the unemployment gap in the short term. The associated effect size is 1.174 (95%-CI: 0.008-2.340)—11.3% of a standard deviation. The cumulative long-run effect of an IMF program is 32.550, but not statistically significant.

On the right-hand part of the table, we find that IMF programs have no unconditional effect on the labor force participation gap. While the first model—with few controls—indicates that being under an IMF program tends to affect the gender participation gap adversely, this result is not robust to the inclusion of control variables and an instrumental-variable design. Hence, we find no consistent evidence of a gender difference in labor force participation rates as a result of IMF programs.²¹

In the supplemental appendix, we present additional analyses to demonstrate the robustness of our main results. Given that the calculation of the unemployment rate involves using the labor force in the denominator, we also perform analyses using the (more encompassing) working-age population, thereby purging any participation effects from the unemployment effect. Our results are similar (Table A4). We also perform diagnostic tests and additional analyses to rule out that our results are sensitive to outlier observations. In the appendix, we show residual plots and leverage plots for both outcomes and do not find any apparent problems with outliers (Fig. A1). Using Cook’s distance, we identify influential observations and drop them from the sample.²² Our results are qualitatively unaffected (Table A5).

While we believe that the unemployment and labor force participation gaps capture the key socioeconomic metrics of gender inequality, we might have missed other forms of economic discrimination. To that end, we probe broader measures of women’s economic empowerment. The first is the Gender Development Index, developed by the United Nations Development Program and extended by IMF staff (Stotsky et al., 2016).²³ In our fully-specified model, entering an IMF program leads to a reduction of the gender development index (p<0.05). The effect is not substantively large but statistically significant. Importantly, it does not vanish once controlling for potential endogenous selection into IMF programs (Table A6). Our second outcome is the Global Gender Gap Index, published by the World Economic Forum. This index measures gender parity across four key dimensions—economic opportunities, educational attainment, health and survival, and political empowerment—for up to 101 developing countries from 2006-23 (World Economic Forum, 2023). We do not find any significant relationship between IMF programs and the index (Table A7). This could mean that the adverse effects of IMF programs are indeed confined to labor market outcomes. Alternatively, the lack of significance could be due to limited data.

We also test whether our findings hold using simpler models. As previewed in our methods section, we use a specification that includes all predictors in levels only. We find that our results are qualitatively similar, especially as regards the unemployment gap. However, IMF programs now appear to predict an unconditionally greater labor force participation gap (Table A8). We believe this result is likely spurious, given that the model imposes restrictions that may not hold.

To mitigate concerns that we may see a deterioration in gender outcomes as a result of labor market reforms, we control for the labor freedom index from the Heritage Foundation (Teorell et al., 2021).²⁴. Controlling for this variable does not affect our main results (Table A9). Relatedly, the effect may be confounded by differences in the strength of labor relative to capital as a societal group. Including the labor share in national income from the Penn World Tables (Feenstra et al., 2015) to our model,²⁵ we find no changes in our main results (Table A9). Importantly, we also verify that including the percentage of women working in the tertiary education sector does not impact our results (Table A10).²⁶ Finally, we verify that our results are not driven by dynamics following regime breakdown and inter-state war (Table A11). Regime breakdown may offer a window of opportunity for women to break existing glass ceilings, whereas inter-state war and the related casualties of men soldiers often create shortages in the workforce that could boost the participation of women in the economy (Webster et al., 2019).

A remaining concern is omitted-variable bias. While our model specifications already include observable confounders, such as financial crises, we may not be able to measure them comprehensively. We have, therefore, presented models using an instrumental-variable design. To probe the robustness of our instrumental-variable analysis, we use an alternative instrument. Specifically, we interact the (logged) IMF liquidity ratio—the ratio of the IMF’s net uncommitted usable resources to its liquid liabilities—with the long-run probability of undergoing an IMF program (Lang, 2021). Following the same logic as our previous shift-share instrument, we assume that global shocks to IMF liquidity are plausibly exogenous with respect to country-specific outcomes. Exposure shares may be endogenous but are absorbed by fixed effects. Using this alternative instrument, we corroborate our main findings (Table A12).

Another strategy to isolate the impact of IMF programs is to drop all cases in which other crises occurred that could have plausibly affected our outcomes of interest. Financial crises are one type of crisis that is often resolved with the help of IMF programs. As it can be difficult to untangle their respective contributions to gendered outcomes, we drop observations in which financial crises occur. Our estimates then reflect the effect of IMF policies under non-crisis circumstances. Similarly, natural disasters can disrupt labor markets while often requiring IMF assistance.²⁷ In both cases, we find even stronger support for our main results (Table A13). This increases our confidence that we are not picking up initial crisis effects that we might have misleadingly attributed to the IMF program.

We may also be concerned that the results are driven by different kinds of IMF programs. We perform two additional tests to address this possibility. First, we control for the incidence of gender budgeting—a newly available indicator collected by IMF staff (IMF, 2021). An increasing number of countries have introduced gender-responsive budgeting, which allows governments to assess the gendered implications of their revenue collection and spending decisions. We recover our main effects even when controlling for this indicator (Table A14). Second, we add to the model a count of the number of binding conditions. Our results remain unchanged, even if they allow for the possibility of endogenous conditionality using a compound instrument strategy that is described in more detail in the appendix (Stubbs et al., 2020) (Table A15). Furthermore, we unpack IMF conditionality and consider specific policy conditions, notably austerity, privatization, price liberalization, and external sector reforms. We also consider labor conditions but note that endogeneity is harder to address. While none of the conditions appears to affect labor force participation gaps, IMF conditions mandating privatization of state-owned enterprises increase the unemployment gap for women. Labor conditions appear to adversely affect the labor force participation gap but ameliorate the unemployment gap (Table A16). These patterns make sense when considering that labor conditions remove protections for formal employment relationships, which are primarily held by men.

4.2 IMF programs, societal gender norms, and gendered labor market outcomes

While IMF programs have homogeneous effects on the gender unemployment gap, we expected IMF program effects on women’s labor force participation gap to be moderated by societal gender norms and the legal environment on women’s economic rights in a given country. Table 2 examines effect heterogeneity for these moderating variables concerning the gender participation gap and the unemployment gap.

Table 2 Unemployment Gap refers to the gap between men and women concerning the unemployment rate. LFP gap is capturing the gap between men and women concerning the labor force participation rate. Linear regression with country-fixed effects and year-fixed effects. Linear regression with country-fixed effects and year-fixed effects. The dependent variable is the (differenced) gender participation gap (models 1-2) and, respectively, the (differenced) unemployment gap (models 3-4). Country-clustered errors in parentheses

First, we find that societal gender biases adversely affect the impact of IMF programs on the participation gap. The negative participation gap under IMF pressure is larger in societies with backward gender norms, as measured by lower values of the Public Gender Egalitarianism index (p<0.01). Similarly, the effect of an IMF program on the participation gap is more detrimental in countries with low de jure economic rights for women (p<0.01).

Coefficient estimates corresponding to the conditional effects are sizeable. Moving societal gender norms from the minimum to its maximum is related to an improvement in the labor force participation gap by 3.261—about one-and-a-half standard deviations. IMF programs are significantly positively related to improvements in the labor force participation gap only in the most gender-equal societies, but under adverse social gender norms, their negative effects are amplified.²⁸

In contrast to the labor force participation gap, the relationship between IMF programs and the unemployment gap is unconditional. This result can be intuited by considering that decisions about (un)employment are made by firms with limited participation and effective influence from other societal actors and the government. The economic rationale would dictate that firms lay off the most vulnerable employees in the first place, which happen to be women.²⁹

In the appendix, we present two robustness tests for our main finding. First, we use an alternative indicator of societal gender norms based on relevant survey items. We take the country average of an additive index of discriminating societal gender norms based on three World Values Survey questions (Haerpfer et al., 2020). These questions elicit respondent support for statements including “women should be housewives”, “men should be prioritized when jobs are short”, and “a man can beat his wife if she burns food” (Roberts & Kwon, 2021).³⁰ Our results are similar compared to when using the latent index (Table A17).

In a second test, we consider that our results might hinge on extrapolation. Rather than using a linear interaction model, we, therefore, use a binning estimator that splits the moderator into terciles and evaluates the outcome at the median values in each tercile. Allowing for this more flexible approach, we confirm that the conditional effect of IMF programs and societal gender norms with respect to the labor force participation gap is indeed linear. For the unemployment gap, we do not find any significant moderating effects (Fig. A2).

4.3 Welfare implications for women

While our evidence suggests that women disproportionally lose out in the labor market as a result of IMF programs, how we ultimately interpret these findings should depend on the alternative options that women must pursue as a result. In a relatively optimistic scenario, newly unemployed women will enroll in higher education. In a pessimistic scenario, they will end up in informal job situations with limited labor protection. They may also temporarily leave the labor market and pursue unpaid care work at home. When the economic situation deteriorates and men lose their jobs, women may suffer from increased domestic violence, potentially leading to more women suicides. With only a few data series with good coverage available, we now probe some of these outcomes.

Table 3 Linear regression with country-fixed effects and year-fixed effects. Outcome variables are in first differences. Country-clustered errors in parentheses

Table 3 scrutinizes the evolution of informal employment, tertiary enrollment, fertility, and suicides in the context of IMF programs using two-way fixed-effects models. Except for fertility—which is measured only for women—we again define these outcomes in terms of gender gaps, whereby more negative values indicate more disadvantageous outcomes for women. We find that the proportion of women pushed into informal employment is comparably higher than the respective portion of men if a country is under an IMF program ($p<0.05$). This finding is in line with evidence from country cases that report an increase in women’s informal sector activities and precarious employment relationships to earn a living wage during IMF programs (Byun, 1999; Peksen et al., 2017; Elias & Rai, 2019). For instance, in an interview during the IMF program in Korea, a women’s beauty salon owner described how the financial crises and austerity measures forced her to seek income from working as a prostitute. During her interview, she stated that “because of the IMF crisis, work was slow, and I had to make up a part-time job [...] I thought about washing dishes at a restaurant, but that doesn’t pay well, and I got to thinking I could be making money while others were asleep” (Byun 1999, 211). To qualify this finding, we also verify that women do not enroll in higher education programs. In line with our prior findings, our results indicate that fertility rates decline with the onset of IMF programs. Finally, we find a weakly significant relationship between IMF programs and the gender suicide gap. Taken together, these results lend support to a rather pessimistic view. As women are driven out of formal labor markets and stripped of viable income sources, they seem to be forced to take up informal employment to make ends meet while cutting back on their ambitions to pursue higher education. Facing increased economic and domestic pressures, women are exposed to substantially higher stress levels, resulting in deteriorating women’s mental health outcomes that can increase suicide risks (Branas et al., 2015; Peksen et al., 2017; Kentikelenis, 2017).³¹

In sum, our findings point to an unconditionally negative effect of IMF programs on the gender unemployment gap that persists even in favorable environments. Furthermore, there is evidence of a larger adverse gender participation gap in societies under IMF pressure with poor gender equality norms and limited legal protections for the economic rights of women. When women become unemployed—or leave the labor force altogether—they often take up informal employment, unpaid care work, and absorb the burden of adjustment. Hence, IMF programs unintentionally work as a powerful catalyst to amplify the adverse effects of societal gender norms on women.

5 Discussion and conclusion

Despite a greater recognition of the adverse consequences of IMF programs on women, even the Fund recognizes that times of economic turmoil have the potential to aggravate existing gender inequities (IMF, 2018). To advance our understanding of the underlying mechanisms linking IMF programs to gender inequality, we study the impact of these programs on women through a labor market lens. Given the salience of labor markets for women’s livelihoods and gender equality, we study the impact of IMF programs through their impact on gender gaps in employment outcomes. Specifically, we ask the question: to what extent do IMF programs exacerbate existing gender inequities in employment outcomes?

We argue that the unequal increase in unemployment for women compared to men during IMF programs can be attributed to the fact that women are primarily employed in less ‘crisis-proof’ jobs (Detraz & Peksen, 2016; Seguino, 2019; Blanton et al., 2019), whereas observed declines in labor force participation, are most pronounced in countries where existing societal norms put women at a disadvantage. Relying on a dataset covering 128 countries between 1992 and 2018, our findings underscore the subtle nature of societal gender biases that are trumping markers of women’s lack of de jure economic empowerment. Importantly, we verify that our results are not driven by IMF program design features, nor do they result from a country’s crisis experience. Our results withstand a series of robustness checks, which increases our confidence that these are not spurious.

From a policy perspective, our findings underscore the importance of considering the specific vulnerability of women toward IMF-supported adjustment measures. First, none of the IMF’s existing instruments can effectively address prevailing labor market dynamics or alleviate the amplifying impacts of its programs. However, by consistently incorporating gender mainstreaming, the IMF has the potential to integrate women’s labor market outcomes into its surveillance mandate. This approach could encourage governments to address institutional barriers rooted in discriminatory gender norms, ultimately enhancing women’s employment opportunities and fostering inclusivity in labor markets (IMF, 2022). Second, recent advances in incorporating concepts such as gender budgeting into lending frameworks aside (Coburn, 2019), these innovations appear, in light of our findings, not sufficient to protect women during times of economic hardship as even the most progressive forms of gender budgeting do not account for women’s unpaid work burdens (Afshar & Dennis, 2016; Bruff & Wöhl, 2016). Thus, embracing a wider notion of gender budgeting already at the program design stage (Blanton et al., 2019; Stubbs et al., 2021), for instance, by incorporating tax breaks for families with children or accounting for women’s additional time commitment for household chores in IMF-supported adjustment programs (Stewart, 2016), might deliver positive employment outcomes for women and even boost prospects for a faster economic recovery. Finally, although we are sympathetic towards the IMF’s own notion that “political declarations mean little unless they are backed by measures that facilitate access to education, affordable childcare, and the labor market” (Andersen 2019, 2), we believe that the IMF’s new gender mainstreaming initiatives and the Fund’s efforts to level the playing field for women across the globe need to be anchored in the wider multilateral ecosystems through inter-institutional collaboration and cooperation to achieve its goals. After all, every single day provides governments, societies, and individuals with the opportunity to do their fair share to level the playing field for women so that they are not first in line to lose their jobs when a crisis hits.