The Role of Fertility in Business Cycle Volatility

Size: px

Start display at page:

Download "The Role of Fertility in Business Cycle Volatility"

Marion Rice
5 years ago
Views:

1 The Role of Fertility in Business Cycle Volatility Sarada Duke University Oana Tocoian Claremont McKenna College Oct Preliminary, do not cite Abstract We investigate the two-directional relationship between fertility decisions and economic conditions, and extract implications for future business cycle volatility. First, we develop a theoretical framework and formulate hypotheses about how fertility is influenced by unemployment rates and GDP growth, and also how labor force participation and unemployment are then driven by childbearing decisions. We then test these hypotheses using US national, state, and individual panel data. Keywords: fertility; business cycle volatility; unemployment JEL Classification: J11, J13, J21 1 Introduction A substantial literature spanning the past century has sought to explain demographic patterns, notably the rates of marriage, births and divorce. While there is consensus on the general trend towards reduced fertility as countries develop, it is not clear how finer variations in GDP impact fertility. As early as 1922, Ogburn and Thomas found that birth rates are pro-cyclical, increasing slightly in periods of prosperity and decreasing during recessions. Galbraith and Thomas (1941) find an even stronger pro-cyclical pattern for the interwar period, and show that it is due mainly to a direct effect, rather than an indirect (i.e. through varying marriage rates) one. Kirk and Thomas (1960) find a much reduced (but still positive) correlation between economic growth and fertility for the immediate post-war period. sarada.x@duke.edu ( sx19/) otocoian@cmc.edu ( 1

2 These findings fit with the theoretical framework of the family developed by Becker, Mincer, and Willis, 1 where children serve as consumption goods, and so the number of children a household opts for increases with household income. More recently, however, Butz and Ward (1979) explain the baby bust of the 1960s with the observation of rising female wages. They propose a model in which child services are provided mainly by the wife, and so a rise in women s opportunity cost of time leads to fewer children. They then conclude that we can expect fertility to be counter-cyclical henceforth, as women continue to enter the labor force in greater numbers. What has not yet been studied, to the best of our knowledge, is the reverse direction of causality - namely, how fertility rates can affect business cycle volatility when women are present in the labor force. The link between past fertility rates and current economic fluctuations has been examined through the demographic age distribution channel by Jaimovich and Siu (2008). What we investigate, however, is the (almost) immediate feedback loop between childbearing decisions and macroeconomic outcomes, through changes in female labor force participation. We develop a theoretical framework in which fertility decisions interact with macro-economic conditions We formulate hypotheses and test them using a combination of macro-level (US national and state-level) and individual level data. 2 Theoretical framework and predictions The discussion below sketches a model of how fertility relates to macroeconomic outcomes as female labor market presence increases. When women do decide to have children (which should occur more frequently at times when the opportunity cost of participating in the labor market is lower), they create a demand for their time at home, thus decreasing their effective wage for at least the following 6 years. They leave the labor force, and will continue to stay home even as their counterfactual wages rise. As they exit the labor force, unemployment drops even before there is any real recovery in terms of output. These women s labor supply is elastic downwards, but then (temporarily) inelastic to wage increases. Compound this with the fact that fertility decisions are lagged by at least one year, and that business cycles are difficult to predict even months in advance, and we end up with a situation where women may be locking themselves out of the labor force even as more jobs are created. In the first years 1 See for instance Becker et al. (1960) or Willis (1973) 2

3 of recovery from a recession, unemployment (measured as a fraction of the active labor force) will appear to drop faster, wages will rise faster, but industrial production and GDP will grow at a slower pace than otherwise. The first observation that should fall out of this logic is that over the past years, as female presence in the labor force has gone up, business cycles have become less volatile in terms of unemployment and wages, but more volatile in terms of GDP growth. We should also see that female unemployment becomes less volatile than male unemployment. In addition, the logic outlined above also explains why, in the wake of a jump in births in 1977, there was a drop in the share of working women, even as female unemployment dropped dramatically. We propose further testing this by looking separately at women under and over 45. If in fact our logic is accurate, we should find that women over 45 follow the female (and male) unemployment rate closely, whereas women under 45 deviate (since their absence from the labor force won t be recorded as unemployment In terms of future predictions, the model outlined above would suggest that, as more childcare is outsourced, and the cost of outside-the-home child services is less volatile than female wages (an assumption), we should observe a return to pro-cyclical fertility, where an increase in either spouse s income leads to an increase in the demand for children. Business cycles should return to where volatility in unemployment closer approximates volatility in GDP. The model described above puts forth four testable hypotheses: Hypothesis 1 Fertility co-varies with the first and second lag of female unemployment, and varies inversely with the first two lags of male unemployment. Hypothesis 2 Fertility has become more counter-cyclical/ less pro-cyclical as female labor force participation has increased. This follows from hypothesis 1. 2 Hypothesis 3 Female labor force participation is depressed for up to 6 years following childbirth. Hypothesis 4 Female unemployment is less volatile than male unemployment, beyond what can be explained by the relative stability of female-intensive industries. This follows from hypotheses 1 and 3. 2 Once we refine our model to allow for increased rates of childcare outsourcing, we will obtain a richer set of predictions. Given static female labor force participation, we expect fertility to become more pro-cyclical as childcare services become more common. 3

4 3 Empirical tests We use both macro (national and state) and micro-level data. Our macro data are drawn from the following sources: from the Bureau of Economic Analysis we obtained data on industry-specific output and employment, dating back to 1947 at the national level, and 1963 at the state level. Data on gender-specific labor force participation and unemployment rates come from the Bureau of Labor Statistics, based on the Current Population Survey. At the national level, this disaggregation is available by gender and age group back to 1948; at the state level, unemployment is disaggregated by gender starting in National birth rates are supplied by the Census Bureau starting in 1970 (as well as one-time observations in 1960 and 1965). State-level birth statistics are drawn from the Center for Disease Control s National Center for Health Statistics, and currently span the years (with further disaggregation about the age of the mother from 1995 on). Individual panel data come from the Panel Survey of Income Dynamics (PSID), hosted by the University of Michigan, and spans the years National-level analysis We first test hypotheses 1 and 2 at the US national level, using the regression specifications shown in table 1. Columns 1 through 3 test hypothesis 1: we regress the growth rate in the number of births by the first two lags of female and male unemployment. We find that, indeed, higher female unemployment predicts higher fertility: when women are out of work, the opportunity cost of having children is lower. Also consistent with our theoretical framework, higher male unemployment predicts lower fertility: when the husband is out of work, the family can t afford to have (more) children. Column 3 shows that GDP growth has no further measurable effect on fertility, beyond what is transmitted through the unemployment channels. Naturally, male and female unemployment are positively correlated, as they are both indicators of macroeconomic conditions. Since they have opposite predictive powers on the number of births, the interesting question of which effect dominates arises- i.e. is fertility pro- or counter-cyclical? Specifications 4 and 5 in table 1 address this question by regressing birth rates on the first two lags of real GDP growth. The coefficient on the first lag is positive and significant, showing that fertility is pro cyclical and that the effect of GDP growth on fertility is immediate. 4

5 Table 1: National-level examination of the business cycle determinants of fertility, s1 s2 s3 s4 s5 s6 (1) (2) (3) (4) (5) (6) Female unemployment Lag (1.08) (1.71) (1.71) Lag (1.39) (1.38) Male unemployment Lag (.75) (1.09) (1.18) Lag (.92) (1.00) Growth in real GDP Lag (.29) (.16) (.18) (.31) Lag 2.07 (.23) time trend (.02) (.02) (.02) (.02) (.02) (.02) time trend squared (.003) (.003) (.003) (.004) (.004) (.004) Lag 1 of real GDP growth time trend (.12) Obs R Notes: dep. variable = growth in the national birth rate (birth rate is the number of births per 1,000 people). Time trend = decades elapsed since the beginning of the sample. Robust standard errors are shown in parentheses. Significance indicated is at 10%(*), 5%(**), and 1%(***). 5

6 The next question is whether the pro-cyclicality of fertility has declined over time (hypothesis 2). Specification 6 tests this by introducing an interaction term between lagged GDP growth and the time trend (expressed as the number of decades since 1970). The coefficient estimated is negative, but not statistically different from zero. We therefore have no indication that the response of birth rates to GDP growth has changed over the 4 decades of the regression. However, the national sample size is clearly very limited - only 40 observations, covering 6 recessions - so insignificant estimates are to be expected. To find more variation, we have to move to the state level. Before we do so, however, it is worth assessing hypothesis 3: is national-level female unemployment less volatile than male unemployment? Figure 1 suggests that the answer is yes. Before 1980 women s unemployment was higher than men s, whereas following 1980 the two track each other closely. However, across all time periods, the variance in female unemployment is lower: it increases by a lesser fraction during recessions, and also drops by less during times of strong economic growth. Since some of this selective variation can be due to composition effects, we will revisit the question at the state and (in later drafts) at the industry level. Figure 1: Male vs. female unemployment,

7 3.2 State-level analysis We test hypotheses 1 and 2 in a pooled sample of all 50 states. 3. State-level analysis is currently limited by the availability of the data on births (as supplied by the CDC) which only goes back to We expect to be able to enlarge the sample using historical population estimates from the Census Bureau, disaggregated by age. For the moment, however, our state regressions only cover the two decades from 1991 (we lose one year since we measure growth) to We find similar results in terms of the role of gender-specific unemployment, as we did at the national level: higher female unemployment leads to higher birth rates, but higher male unemployment reduces fertility. When testing whether GDP growth has a time-varying impact on fertility, we noticed an uneven trend and therefore introduced separate controls for 5-year intervals. We find a positive and significant effect from 1996 to 2010, with the strongest being in the early 2000s, but no discernible relationship during the early 1990s. In other words, over these two decades, fertility becomes first more, then less pro-cyclical. We hesitate to assign a wider interpretation to this apparent time variation, since female labor force participation did not change significantly over these last 2 decades. However, once we have expanded the dataset, 4 it will be interesting to run this analysis over a longer time horizon. 4 Microdata analysis We expand our empirical analysis by using data from the Panel Survey of Income Dynamics (PSID). Since this dataset allows us to track individuals and families over time, it is the best suited tool for analyzing personal choices in the wake of major life events like childbirth. When we ask whether women choose to have children in response to poor labor market prospects, aggregate data is a useful tool. High state-level female unemployment is a good indicator that the opportunity cost of bearing children is low, on average, for the women in that state. Your neighbor s labor market opportunities mirror your own. However, post-natal withdrawal from the labor market comes in response to a householdspecific, localized circumstance. There is no sense in which your neighbor s newborn baby impacts your work decisions. Furthermore, the presence of an infant in the home is easily observable in survey data, whereas measuring labor market prospects is less straightforward. 3 We exclude the District of Columbia, since many DC workers are commuters from neighboring states. 4 The most promising way of obtaining earlier state-level fertility information appears to be by using historical population age distribution estimates from the Census 7

8 Table 2: State-level examination of the business cycle determinants of fertility, s1 s2 s3 s4 s5 (1) (2) (3) (4) (5) Female unemployment Lag (.10) (.10) Lag (.10) (.11) Male unemployment Lag (.08) (.08) Lag (.09) (.08) Growth in real GDP Lag (.03) (.03) (.04) Lag (.02) (.02) Interactions of real GDP growth, lag 1 time trend.03 (.04) indicator for (.03) indicator for (.05) indicator for (.06) indicator for (.05) State FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Obs. 1,000 1,000 1,000 1,000 1,000 e(n-clust) R Notes: dep. variable = growth in the state-level birth rate (birth rate is the number of births per 1,000 people). Time trend = decades elapsed since the beginning of the sample. All specifications include state and year fixed effects, and standard errors are clustered at the state level. Significance indicated is at 10%(*), 5%(**), and 1%(***). 8

9 Figure 2: Male vs. female unemployment at the state level, top 6 states,

10 For these reasons, we use the PSID to test hypothesis 4: that women withdraw from the labor force for several years after childbirth. We feel that this is a well established and widely known fact, but one that is still worth pointing out because of its scale. Figure 3 shows the working status for married women and men, following the birth of their first child (at -1 on the horizontal axis). It paints a striking picture of the differential impact of offspring on parents presence in the labor market. The labor force participation rate of married women drops from 85% before having children to around 60% for the first 6-7 years, starts to rise slowly afterwards and levels off a full 16 years later. Even then, it never reach the level of men s participation rate. Figure 3: Married women s working status, relative to the birth of their first child. 5 Next steps We first plan to expand the dataset to encompass earlier time periods in the state level data, so that we can better test hypothesis 4 - the change in the pro-cyclicality of fertility decisions. We also plan to identify and exploit variation in women s presence across industries and occupations. We expect that business cycle shocks will be transmitted differently (be attenuated) through industries where women form a significant part of the labor force. 10

11 6 Conclusion We are constructing a framework that encompasses the two-directional relationship between fertility decisions and macroeconomic conditions, in particular labor force participation and unemployment. Preliminary analysis shows empirical support for both channels of causation, and suggests that household economics and female labor force participation play an important role in determining business cycle volatility. References Becker, Gary S, James S Duesenberry, and Bernard Okun, An economic analysis of fertility, in Demographic and economic change in developed countries, Columbia University Press, 1960, pp Butz, William P and Michael P Ward, The emergence of countercyclical US fertility, The American Economic Review, 1979, 69 (3), Galbraith, Virginia L and Dorothy S Thomas, Birth rates and the interwar business cycles, Journal of the American Statistical Association, 1941, 36 (216), Jaimovich, Nir and Henry E Siu, The young, the old, and the restless: Demographics and business cycle volatility, Technical Report, National Bureau of Economic Research Kirk, Dudley and Dorothy Swaine Thomas, The influence of business cycles on marriage and birth rates, in Demographic and Economic Change in Developed Countries, Columbia University Press, 1960, pp Ogburn, William F and Dorothy S Thomas, The influence of the business cycle on certain social conditions, Journal of the American Statistical Association, 1922, 18 (139), Willis, Robert J, A new approach to the economic theory of fertility behavior, The Journal of Political Economy, 1973, 81 (2), S14 S64. 11

Economic Uncertainty and Fertility: Insights from Japan. James M. Raymo 1. Akihisa Shibata 2

Economic Uncertainty and Fertility: Insights from Japan James M. Raymo 1 Akihisa Shibata 2 1: Department of Sociology and Center for Demography and Ecology University of Wisconsin-Madison 2: Kyoto Institute