Segregated Economies in an Integrated World

A persistent feature of national economies is occupational segregation by gender.Footnote ⁹ In industrialized economies, for example, women are disproportionately employed in the service sector and in white-collar jobs.Footnote ¹⁰ By contrast, in developing economies—this study's subject—it is well documented that women are disproportionately employed in industries that are simultaneously low-skilled and export-oriented.Footnote ¹¹ Among the explanations for the segregation are the discriminatory hiring practices that result from employers’ information problems.Footnote ¹² To elaborate, the jobs from which women in low- and middle-income economies are excluded are often characterized by sizable and increasing returns to human capital and specific skills. This matters because if productivity is increasing with human capital and employment tenure, employers will prefer hiring workers that have considerable human capital and who will remain at the firm for a long time. The problem is that ex ante, these attributes are imperfectly observed. For example, human capital is fully on display only after one is put to the task of production. In other cases, the imperfect observability is strategic. If job candidates know an employer prefers long-term employment contracts, all of them will say that they will remain at the job for a long time regardless of their true intentions.

Such information problems often lead to discriminatory hiring practices.Footnote ¹³ Consider: in many countries, patriarchal attitudes and structures are in place that result in girls receiving less education than boys. Likewise, patriarchal norms and structures mean that women are more likely than men to interrupt their careers for child rearing and other family responsibilities. Theories of employment discrimination argue that in light of the imperfect observability of job applicants’ productivity, employers use these statistical tendencies as informational shortcuts. Although employers do not know for certain the human capital and tenure preferences of any particular candidate in the labor pool, if women on average have less human capital than men do, then the employer with a demand for high human capital will prefer hiring men. Likewise, if a randomly drawn woman has a higher probability of interrupting her career than a randomly drawn man, then the employer who has imperfect information about job candidates’ productivity and who prefers long-term employment contracts will also discriminate against women. This “statistical discrimination” excludes women in developing countries from jobs that require considerable human capital and that are characterized by increasing returns to skill. Notably, though, in the low-skilled manufacturing industries that tend to employ more women, the human capital requirements are lower and the returns to specific skills not so steeply increasing. Accordingly, employers in these industries have smaller premiums on human capital and long-term employment contracts and are more amenable to hiring women.

The same information problems have another, related effect. It is common in many communities to stereotype women as possessing greater manual dexterity than men and as being more docile and honest than men. Such stereotypes are used by employers to infer that women are more productive than men in low-skilled, labor-intensive work.Footnote ¹⁴ In combination, these forces lead to the segregation of women in low-skilled occupations.

Notably, this concentration in low-skilled occupations is increasing for export-oriented firms.Footnote ¹⁵ To see why this is so, consider that when developing economies rushed to free trade in the 1980s and 1990s, one consequence was that exporters everywhere found themselves in an increasingly competitive environment characterized by a flatter demand curve for their products abroad.Footnote ¹⁶ This more elastic demand, in turn, obliged employers in exporting firms to place a high premium on reducing their costs—which, given the labor intensity of the production, often amounted to keeping the wage bill low. In a perverse twist, the labor market disadvantages noted in the previous paragraphs produce an overabundance of women applying for these low-skilled jobs. This oversupply keeps wages low, which in turn makes the export-oriented employers seeking to keep their labor costs low particularly amenable to hiring women.Footnote ¹⁷

Exchange Rates in a Segregated Economy

Given that women in developing economies are disproportionately employed in export-oriented industries (particularly when low-skilled), it can be demonstrated that REER movements have gendered implications. It is well known that an overvalued exchange rate increases export prices and reduces export volumes. Because labor-intensive, export-oriented industries tend to hire women disproportionately, the practical consequence of the overvaluation is a disproportionate reduction in the demand for female labor in the economy. Over time, this low demand will frustrate women seeking employment and cause some of them to exit the labor force altogether. An undervalued currency is expected to have the opposite effect, increasing export competitiveness and increasing the demand for female labor commensurately.

This baseline model is a useful starting point, but it simplifies relevant economic realities by ignoring the increasingly international nature of production. This matters because many exporters are also importers, and these supply chain linkages can reduce the pass-through of exchange rates to export prices and volumes.Footnote ¹⁸ We are therefore obliged to ask whether internationalized production eliminates the relationship between appreciation and FLFP sketched earlier. Here, I show that it need not. The crux of the matter regards factor intensity. To elaborate, consider a firm producing Q = F(K, L), where K is imported capital equipment, L is the firm's domestic labor force, and Q is produced entirely for the export market. Exchange rate movements alter the relative prices of capital and labor and necessitate using one factor more intensively and the other less so. Figure 1 illustrates the consequences. Let curve Q express the various ways the firm can combine K and L to produce Q. C is the firm's cost curve, and A is the cost-minimizing mix of capital and labor in the production of Q in the status quo situation. Because capital is imported, appreciation reduces the price of K relative to L, shifting the cost curve from C to C′. Now, cost minimization obliges the firm to produce at B, using capital more intensively in the production of Q and labor less so. Thus, even though export volumes are unchanged, appreciation can still affect employment by changing the relative prices of factor inputs.

FIGURE 1. Capital substitution in response to overvaluation

One other issue warrants discussion. To say that (low-skilled) export-oriented manufacturing is a particularly important source of female employment in developing countries is not to say that such jobs are the only source of female employment. Women also frequently work in nontraded services. Importantly, standard models suggest that by reducing import prices overvaluation increases purchasing power and thus constitutes a boon for nontraded services. This could offset the effects of overvaluation described before. However, Frenkel and Ros suggest that this account oversimplifies the dynamics at work.Footnote ¹⁹ In more detail, they study a two-sector economy—where one sector is export oriented and formal and the other is nontraded and informal—and identify several pathways by which overvaluation can reduce labor market activity, even if labor can easily move between sectors.

First, while acknowledging that appreciation increases purchasing power and thereby stimulates demand for services, Frenkel and Ros also show that overvaluation-induced unemployment in the traded sector reduces demand for services and thereby offsets the purchasing-power effect. Second, Frenkel and Ros argue that if movement between sectors is easy, then overvaluation-induced unemployment in the traded sector increases the labor supply in the nontraded sector as the unemployed in the former seek employment elsewhere. The increased labor supply reduces nontraded wages and causes some share of those workers to exit the labor force altogether. Finally, in a manner similar to Figure 1, Frenkel and Ros argue that, to the extent the nontraded services rely on imported intermediate goods, overvaluation encourages using labor less intensively and capital more intensively, further reducing employment and labor market activity. Thus, the net effect of overvaluation might still be to reduce labor market activity.

Though neither Frenkel and Ros nor the analysis in Figure 1 pertains specifically to female employment and FLFP, it stands to reason that in an economy where women are disproportionately employed in low-skilled and export-oriented jobs, the employment consequences they highlight will be felt most acutely by women workers.

Data and Methods

These ideas yield the following.

I test this hypothesis using time-series cross-sectional data for 1990 through 2015, covering about 150 low- and middle-income countries.Footnote ²⁰ The dependent variable is the rate of FLFP. The main independent variable is reer overvaluation, measured according to the procedures outlined by Rodrik.Footnote ²¹ Start by calculating the logarithm of the real exchange rate,

(1)$${\rm logRE}{\rm R}_{i, t} = {\rm log}\left({\displaystyle{{{\rm XRA}{\rm T}_{i, t}} \over {{\rm PP}{\rm P}_{i, t}}}} \right), \;$$

where XRAT is the nominal exchange rate between country i and the US dollar and PPP is the purchasing power parity conversion factor. Then, adjust logRER to account for the Balassa–Samuelson effect by regressing it on per capita GDP and year fixed effects:

(2)$${\rm logRE}{\rm R}_{i, t} = \alpha + \beta _1{\rm GD}{\rm P}_{i, t} + f + \mu _{i, t}, \;$$

and finally, calculate

(3)$${\rm logOvervalue}{\rm d}_{i, t} = {-}1 \, \ast \, ( {{\rm logRE}{\rm R}_{i, t}-\widehat{{{\rm logRE}{\rm R}_{i, t}}}} ) , \;$$

where $\widehat{{{\rm logRE}{\rm R}_{i, t}}}$ in (3) are the predicted values from (2). Rodrik shows that Overvalued = exp (logOvervalued) is a cross-nationally comparable measure of overvaluation. However, as the distribution for Overvalued is highly skewed, the logarithmic transformation directly calculated in (3) is more appropriate for empirical analysis and is used in the regressions that follow. Figure 2 shows that variable's distribution for low and middle-income countries between 1990 and 2015.Footnote ²²

FIGURE 2. Distribution of logOvervalued among low- and middle-income countries, 1990–2015

The regression also includes several control variables. Together, per capita gdp (logged) and squared per capita gdp (logged) account for the U-shaped relationship between economic development and FLFP.Footnote ²³ I also control for oil wealth. This is for several reasons. First, oil wealth has long been linked to overvalued exchange rates.Footnote ²⁴ Second, to this Ross adds that oil might be especially bad for FLFP for two reasons.Footnote ²⁵ On the demand side, oil causes overvaluation, which then lowers FLFP through the processes developed here. On the supply side, oil wealth increases women's nonlabor income and thereby reduces the number of hours women allocate to market labor. The supply-side effect constitutes a “backdoor path” for which my model must adjust, lest the effects of logOvervalued be biased.Footnote ²⁶ I make the adjustment by controlling for the logarithm of per capita oil and gas rents. Finally, I include country fixed effects to adjust for all time-invariant and slowly moving confounds. For instance, democracy influences both exchange rate orientations and FLFP.Footnote ²⁷ Likewise, a country's culture affects both FLFP and exchange rate choices.Footnote ²⁸ History matters, too. For instance, countries that implemented import substitution often chose overvalued exchange rates to make imported capital equipment cheaper,Footnote ²⁹ but such countries were often also associated with lower FLFP.Footnote ³⁰

Tests cannot rule out the possibility of nonstationarity for several of these variables, though all are stationary when differenced.Footnote ³¹ Accordingly, I estimate the single-equation error-correction model recommended by DeBoef and Keele, wherein the change in FLFP is regressed on its lagged level, lags and differences of the independent variables, and a sufficient number of lagged differences of the dependent variable to rule out serially correlated errors.Footnote ³² In the present context, this yields

(4)$$\eqalignb{& {\rm \Delta }Y_{i, t} = \beta _1Y_{i, t-1} + \beta _2{\rm \Delta }Y_{i, t-1} + \beta _3{\rm logOvervalue}{\rm d}_{i, t-1} + \beta _4{\rm \Delta logOvervalue}{\rm d}_{i, t} \cr & \qquad \quad + \beta _5{\rm GD}{\rm P}_{i, t-1} + \beta _6{\rm \Delta GD}{\rm P}_{i, t} + \beta _7{\rm GDP}_{i, t-1}^2 + \beta _8{\rm \Delta GDP}_{i, t-1}^2 \cr & \qquad \quad + \beta _9{\rm resourceRent}{\rm s}_{i, t-1} + \beta _{10}{\rm \Delta resourceRents}_{i,t} + \gamma + \epsilon _{i, t}, \;} $$

where γ represents the vector of county dummies, i indexes countries, and t indexes time. Hypothesis 1 predicts negative coefficients on either β ₃ or β ₄, or both. Table 1 shows the summary statistics for the variables in the equation.Footnote ³³

TABLE 1. Summary Statistics

Note: L.X and ΔX refer to the lagged and differenced values of X.

Demonstrating the Mechanisms

The Table 2 results support Hypothesis 1. Here, I use the case of Mexico to assess whether the argument identifies a plausible causal pathway linking overvaluation to FLFP. Mexico is a useful case in this regard because in the mid-1990s it experienced two pertinent macroeconomic changes. First, NAFTA went into effect in 1994 and brought with it an explosion of export-oriented factories along the country's northern border. The maquiladoras had long been an important source of female employment in the country—for many of the reasons documented earlier—and that trend continued with NAFTA's arrival.Footnote ³⁹ Second, following the balance-of-payments crisis in late 1994, the government floated the peso in 1995. If the theory developed here is correct, we should expect to observe correspondence between movements in the now-floating peso and movements in women's labor market participation.

The argument's crucial assumption is that low-skilled export-oriented manufacturing is an especially important source of female employment. To assess whether this claim holds for Mexico, Table 3 regresses female employment in Mexico in industry i in year t (in thousands, logged) on a categorical variable that identifies the twenty-one economic sections specified by Revision 4 of the International Standard Industrial Classification (ISIC). What we would like to know is whether the manufacturing industry is an important source of female employment in the country. To that end, manufacturing is the excluded category in the regression, such that all coefficient estimates indicate whether industry i employs more or fewer women than manufacturing. Notice that nearly every industry has a negative and statistically significant coefficient estimate, implying that it employs fewer women than manufacturing. The exception is wholesale and retail trade, which employs more women.Footnote ⁴⁰

Though useful, the Table 3 results do not speak to whether women are disproportionately employed in low-skilled occupations by export-oriented firms in the manufacturing sector. However, an extensive literature has found that they are. For instance, Anderson and Dimon conduct labor surveys in two cities: Tijuana, a major hub for export-oriented manufacturing; and Torreón, where most production is domestically oriented agriculture. They draw two important conclusions. First, demand for female labor is higher in Tijuana than in Torreón. Second, while labor force participation is higher in Tijuana for both men and women, the effect for women is proportionally larger, so the participation gap between men and women is smaller.Footnote ⁴¹ Villarreal and Yu concur. Using data from a nationally representative labor survey, they find that export-oriented firms employ more women than firms producing for the domestic market do.Footnote ⁴² Likewise, Aguayo-Tellez and colleagues find that trade liberalization in Mexico increased the demand for female labor by expanding industries that already employed female labor intensively.Footnote ⁴³

There is similarly compelling evidence that, in these maquiladoras, women are disproportionately employed in low-skilled jobs. Kopinkak presents a particularly striking piece of evidence. In a study of Nogales, Mexico, she notes that, at the time of her study, it was common for maquiladoras to place job advertisements in the local paper that specified the gender of the successful candidate. Some 40 percent of job advertisements did so, and of that set of advertisements, nearly 90 percent were for an unskilled job, while none of the advertisements for skilled jobs specified a female employee.Footnote ⁴⁴ Fussell concurs.Footnote ⁴⁵

Finally, there is also evidence that overvaluation causes negative growth in Mexico's manufacturing sector and that this effect is particularly pronounced in labor-intensive manufacturing.Footnote ⁴⁶ Given the combination of all these results, Hypothesis 1 suggests that movements in the peso will correspond with movements in FLFP. Regression analysis supports this prediction. The results in model 1 of Table 4 derive from an error-correction model predicting the change in FLFP with lagged and differenced values of logOvervalued for Mexico. Because there are so few observations (only twenty-three, after adding the appropriate lagged and differenced variables), only a subset of the control variables included in the cross-national results are included here—namely just lagged and differenced GDP. The results are consistent with expectations. Though we do not observe an immediate effect of overvaluation (ΔlogOvervalued is negative but statistically insignificant), the lagged level of the variable in question is both negative and statistically significant. Model 2 in Table 4 uses the labor force ratio as the dependent variable and reaches a similar conclusion, though in this case both logOvervalued entries are statistically significant. All of the results presented thus far are consistent with Hypothesis 1: An overvalued REER appears to be particularly bad for women's labor market participation.

TABLE 4. REER Overvaluation and Female Labor Force Participation in Mexico

Note: *p < 0.10; **p < 0.05; ***p < 0.01.

The Choice of Exchange Rate Valuation

To this point, we have left a country's exchange rate valuation exogenous. This is helpful for expository purposes, but as a practical matter, governments are able to keep currencies undervalued or overvalued for considerable lengths of time,Footnote ⁴⁷ and they are presumed to do so in response to competing societal interests. For instance, an undervalued currency promotes exports and helps import-competing firms, but also raises inflation and reduces purchasing power. Naturally, an overvalued currency has the opposite effects.Footnote ⁴⁸ Accordingly, consumers and producers are expected to oppose each other with regard to the exchange rate and to compete to have their preferences reflected in national policy.

I do not dispute these arguments. But notice that, according to the results here, women might also constitute a social group with preferences relevant to the choice of exchange rate. Indeed, in low- and middle-income countries, women might face a trade-off that has been underappreciated in current scholarship. Overvaluation has the aforementioned purchasing power effect, but by reducing women's labor force participation, it also exerts a negative income effect. And consider also that because working outside of the home gives women more bargaining power within it,Footnote ⁴⁹ overvaluation also has a second-order (but no less important) household bargaining effect. How might women navigate this trade-off? It seems plausible that the combination of the latter two effects trumps the first, such that women in developing economies will tend to prefer a neutral or undervalued currency to an overvalued one.

That said, our interest here is not these preferences per se but national policy—or, more accurately, the consequences thereof as reflected in exchange rates. However, we can presume that whether women's opposition to an overvalued currency is reflected in national policy will depend on two factors: the extent to which women are engaged in political life, and whether the country is democratic and therefore responsive to citizen demands. These conditions allow us to formulate an extension of this paper's main argument:

A complete investigation of Hypothesis 2 is beyond the scope of this brief note, but a preliminary investigation can proceed by regressing logOvervalued on a measure of women's political engagement, on regime type, and on the multiplicative interaction of the two (Table 5). The expectation in Hypothesis 2 is that the marginal effect of political engagement will be statistically or substantively insignificant in nondemocracies and negative and statistically significant in democracies. Political engagement is measured with an item in Wave 6 of the World Values Survey, “How interested would you say you are in politics?” (possible responses: very, somewhat, not very, not at all).Footnote ⁵⁰ For each female respondent, I code a dummy variable as 1 for “very” or “somewhat” and 0 otherwise. Then, for each country, I calculate the proportion of female respondents with a score of 1, producing a country-level measure of women's political engagement. Finally, note that because the engagement variable uses survey data, the regression is necessarily cross-national rather than relying on panel data. Specifically, Wave 6 of the World Values Survey was conducted between 2010 and 2014, depending on the country, and all the economic and political variables in the model are measured during the same year the survey was fielded in country i.

TABLE 5. Women's Political Engagement, Democracy, and Exchange Rate

Note: Robust standard errors in parentheses. *p < 0.10; **p < 0.05; ***p < 0.01.

The coefficient estimates are presented in Table 5, but a more useful way to understand the results is with Figure 3. The solid line traces the marginal effect of women's political engagement on logOvervalued as levels of democracy change in the sample; the shaded area is the 95 percent confidence interval. The rug plot displays the distribution of regime type in this sample of about forty low- and middle-income countries. The results are consistent with H2 Women's engagement has no effect on logOvervalued in the most authoritarian regimes. For values of the democracy measure—which comes from the Varieties of Democracy data setFootnote ⁵¹ and spans the interval [0–1]—between 0.01 (Qatar) and 0.4 (Algeria), there is no statistically reliable effect of women's political engagement on logOvervalued. In more democratic settings, though, women's political engagement has the predicted statistically significant negative effect.

FIGURE 3. Marginal Effect of Women's Political Engagement on logOvervalued

Though these results comport with Hypothesis 2, caution should be taken with them. They are correlational and thus may suffer from reverse causality or some other endogenous process, and they derive from a small sample, since they are based on coverage of Wave 6 of the World Values Survey. Nevertheless, the results are provocative and consistent with the hypothesis. In that regard, they invite further research into the topic.