Job postings and aggregate stock returns

https://doi.org/10.1016/j.finmar.2023.100804Get rights and content

Highlights

  • The job openings-to-employment ratio (JOE) is a strong predictor of the equity premium.

  • JOE outperforms other popular predictor variables in forecasting tests.

  • Return forecasts based on JOE are useful to real-time investors.

Abstract

The job openings-to-employment ratio (JOE), defined as the number of job postings divided by the employment level, is among the strongest known predictors of the equity premium. We find that JOE outperforms a broad set of over two dozen popular predictor variables in both in-sample and out-of-sample forecasting tests. Forecasts based on JOE also produce gains of 2.91% in annualized certainty equivalent return and 0.20 in annualized Sharpe ratio relative to forecasts based on the historical mean equity premium. The empirical results are consistent with a standard production-based asset pricing model with labor inputs and search frictions.

Introduction

Fama and French (1989) find that expected excess returns on common stocks are countercyclical such that investors demand a higher risk premium to hold stocks at the end of a recession. Similarly, Cochrane (2005) emphasizes that the most economically appealing forecasting variables for equity returns are those that trend at business cycle and longer horizons. Because employment levels and wages are integral components of business cycle variations, predictor variables that summarize labor market conditions seem likely to be associated with future stock market excess returns. Previous studies on the time-series predictability of aggregate returns have identified a long list of forecasting variables with empirical success, including valuation ratios (e.g., Campbell and Shiller, 1988a, Campbell and Shiller, 1988b, Fama and French, 1988, Hodrick, 1992, Lewellen, 2004), interest rates (e.g., Fama and Schwert, 1977, Fama, 1981), default and term premiums (e.g., Keim and Stambaugh, 1986, Fama and French, 1989), and inflation (e.g., Fama and Schwert, 1977, Fama, 1981, Campbell and Vuolteenaho, 2004).1 Relatively few studies, however, examine the role of labor market conditions in aggregate return predictability.

In this paper, we show that the job openings-to-employment ratio (JOE), defined as the number of job postings in the United States divided by the employment level, is among the strongest known predictors of the equity risk premium. The intuition for this relation mirrors the intuition for the negative relation between aggregate investment and expected stock returns predicted by the standard production-based asset pricing model with capital inputs (e.g., Cochrane, 1991). In the standard setting, a high cost of capital implies a low net present value of new investment and a low investment rate (holding constant expected productivity). Similarly, a production-based model with labor inputs and labor adjustment costs predicts that a high discount rate is associated with a low net present value of new hires and a low job postings rate.2 From an economic perspective, Kuehn et al. (2017) note that more than 10% of U.S. workers separate from their employers each quarter. In the presence of search-matching frictions in the labor market, replacing these workers and adjusting employment levels in response to productivity and discount rate shocks is costly to firms. Our aggregate job postings variable thus summarizes firms’ forward-looking intentions to hire new workers and contains important information about expected aggregate stock market returns.

We motivate our empirical work using a production-based asset pricing framework with quadratic adjustment costs (e.g., Cochrane, 1991) and labor inputs (e.g., Chen and Zhang, 2011, Belo et al., 2014, Belo et al., 2022). The model economy consists of identical firms with no leverage and identical workers. To either maintain the current size of the workforce (by replacing departing workers) or change the size of the workforce, firms advertise job openings and recruit new workers. The model also allows for adjustment costs, which represent the resources spent on hiring activities, such as advertising open positions, screening candidates, training new hires, and implementing a new organizational structure. The quadratic specification for the adjustment costs implies that they are increasing and convex in the number of job postings, reflecting an increasing marginal cost of additional hires. The model’s first-order condition for optimal hiring equates the marginal cost of an additional job posting with its expected discounted marginal product. Because adjustment costs increase with the ratio of job postings to employed workers, the model implies a negative relation between this ratio and the expected stock market return.

The job openings-to-employment ratio is the primary variable of interest for our empirical tests. We follow Barnichon (2010), Cajner and Ratner (2016), and Petrosky-Nadeau and Zhang (2021) to measure the total number of vacancy postings and scale this variable by the U.S. civilian employment level. In Fig. 1, we plot the monthly time series of JOE from February 1951 to December 2021. The shaded regions in the plot indicate recessions as defined by the National Bureau of Economic Research. The job openings-to-employment ratio displays strong cyclicality, peaking just prior to the start of each recession in the sample. As such, JOE appears to be a leading indicator of the business cycle.

We evaluate the predictive power of JOE for future excess returns on the Center for Research in Security Prices (CRSP) value-weighted index using standard predictive regressions (e.g., Fama and French, 1989). We consider forecast horizons ranging from one month to three years. At each horizon, JOE is a negative and significant predictor of returns at the 1% level. The estimated slope coefficients also suggest that the predictive effects are economically large. At the monthly forecast horizon, for example, a one standard deviation increase in JOE is associated with a 0.49% per month decline in the expected excess market return.

To provide additional perspective on JOE’s predictive content for excess market returns, we compare its performance with that of 27 other predictor variables with data availability at a monthly frequency from February 1951 to December 2021 (i.e., the sample period for JOE). We term this set of 27 predictors as the “primary” benchmark set, and we further classify these predictors into three groups. The first group includes 14 forecasting variables examined by Goyal and Welch (2008). This set of predictors is widely used in the literature on stock market return predictability (e.g., Ferreira and Santa-Clara, 2011, Rapach et al., 2016). The second group consists of nine variables introduced in articles published in top journals subsequent to the Goyal and Welch (2008) study, and the third group is composed of four labor market predictors.

The job openings-to-employment ratio compares favorably with the primary benchmark predictor variables in the in-sample tests. At the monthly forecast horizon, JOE is the strongest individual predictor of excess stock market returns. The R2 value for the regression with JOE is 1.35%, whereas the R2 values for the 27 benchmark predictors range from 0.00% to 1.06%. At the annual forecast horizon, just one variable (Eiling et al.’s (2021) cross-sectional volatility of lagged industry returns) produces an R2 value that exceeds JOE’s R2 of 8.18%. We also consider multiple regression specifications for each forecast horizon and find that JOE’s explanatory power for returns remains strongly significant after controlling for any of the primary benchmark predictors.3

Goyal and Welch (2008) demonstrate that predictor variables often exhibit poor performance in out-of-sample tests. Using their testing methods, we find that the predictive content of JOE extends to the out-of-sample setting and generates out-of-sample R2 statistics of 0.82%, 2.10%, 3.40%, 4.40%, and 4.76% at the monthly, quarterly, semiannual, annual, and biennial horizons, respectively. The positive values for these statistics suggest that out-of-sample forecasts based on JOE produce lower mean square prediction errors than do forecasts based on the historical mean equity premium. These out-of-sample R2 values are statistically significant at the 5% level in each case. By contrast, the out-of-sample R2 for three-year returns is negative. This estimate does turn positive, however, once we incorporate Campbell and Thompson’s (2008) suggestion to restrict the parameters in the forecasting model to be consistent with theory.

The 27 primary benchmark predictors often perform poorly in out-of-sample tests. At the monthly horizon, the R2 values for the benchmark predictors range from −1.10% to 0.78%, and 17 of the 27 R2 estimates are negative. Just two of the R2 values are positive and statistically significant at the 5% level. We also consider out-of-sample encompassing tests following Harvey et al. (1998). These tests confirm that forecasts based on JOE contain superior information relative to those based on the primary benchmark predictors.

We characterize the economic significance of the predictive ability of JOE and the 27 primary benchmark predictor variables using asset allocation analyses. We consider a mean–variance investor who allocates between the CRSP value-weighted index and the risk-free asset based on the return forecasts from a given predictor variable. The baseline results correspond to an investor who rebalances on a monthly basis and also feature reasonable constraints on investment positions. The portfolio strategy that uses JOE to form return forecasts generates an annualized out-of-sample Sharpe ratio of 0.47 compared with just 0.27 for the strategy based on the historical mean equity premium. A mean–variance investor also gains 2.91% per year in certainty equivalent return by moving from the forecast based on the historical mean to the forecast based on JOE. The estimated Sharpe ratio and utility gains are both statistically significant at the 5% level. The utility improvements from JOE exceed those from any of the 27 benchmark predictor variables. The gains in certainty equivalent returns associated with these predictors, for example, range from −1.16% to 2.56% per year, and only three of the estimates are statistically significant at the 5% level.

Our findings contribute to the growing literature on production-based asset pricing and the relation between labor inputs and assets returns. A large portion of the literature focuses on the role of labor investment decisions in explaining the cross-section of expected stock returns. A partial list of labor-based characteristics linked to asset prices includes industry unionization rate (Chen et al., 2011), organizational capital (Eisfeldt and Papanikolaou, 2013), industry labor mobility (Donangelo, 2014), firm-level hiring rate (Belo et al., 2014, Belo et al., 2017), exposure to labor market tightness (Kuehn et al., 2017), growth rate in labor hours per worker (Gu and Huang, 2017), and labor leverage (Donangelo et al., 2019). In contrast to these studies, we emphasize the predictive content of aggregate job postings for the time series of stock market returns.

The three papers closest to ours are Chen and Zhang (2011), Eiling et al. (2021), and Belo et al. (2022). These studies show that employment growth, sectoral labor reallocation shocks, and the hiring rate of public firms, respectively, forecast aggregate returns at various horizons. We confirm the empirical success of each of these predictors, but also demonstrate that they exhibit just moderate correlation with JOE. Moreover, JOE compares favorably with the existing labor-based predictors in both in-sample and out-of-sample tests, and JOE’s forecasting ability for aggregate returns remains strongly significant after controlling for these variables.

This paper also contributes to the macroeconomic literature that studies theoretical links between labor and financial markets. For example, Phelps (1994) examines the impact of interest rates on unemployment. Merz and Yashiv (2007) propose a model to link firm value to aggregate investment and hiring, and Yashiv (2016) extends this model to incorporate the interaction between the adjustment costs for investment and hiring. These studies feature a neoclassical production-based economy with labor market frictions, but do not consider implications for return predictability. Danthine and Donaldson (2002), Uhlig (2007), and Hall (2017) demonstrate theoretical relations between aggregate expected returns and aggregate labor share, wage rigidity, and unemployment, respectively. None of these studies tests whether the corresponding relation holds empirically.

Prior studies on return predictability highlight two desirable properties of a proposed forecasting variable. First, Goyal and Welch (2008), Harvey et al. (2016), and Elliott and Timmermann (2016) emphasize that a given variable that describes expected returns should be linked to economic theory to alleviate concerns that its empirical success is spurious. Second, Fama and French (1989) and Cochrane (2005) note that the most plausible time-series predictors of stock market returns are variables linked to the business cycle and long-term business conditions. The job openings-to-employment ratio satisfies both of these requirements. It is motivated through standard investment theory and is more economically appealing relative to predictors that lack a theoretical foundation. The job openings-to-employment ratio also captures conditions in the labor market that are integral components of the business cycle.

The rest of the paper is organized as follows. In Section 2, we present a production-based asset pricing model that links job postings to subsequent period expected returns. In Section 3, we describe the data sources and introduce the predictor variables. Section 4 contains our empirical tests on the in-sample and out-of-sample performance of JOE and the benchmark predictors. We conclude in Section 5. We present additional details on variable definitions and data construction in the Appendix.

Section snippets

Model

In this section, we establish a theoretical link between the job openings-to-employment ratio and expected returns. We build on the standard production-based asset pricing model with labor inputs and stochastic adjustment costs associated with labor market frictions.

Data

In Sections 3.1 Aggregate stock market excess returns, 3.2 Job openings-to-employment ratio, 3.3 Primary benchmark predictor variables, 3.4 Secondary benchmark predictor variables, we describe the data sources and variable construction for the stock market excess return, JOE, the 27 primary benchmark predictors, and the nine secondary benchmark predictors. In Section 3.5, we present summary statistics for the forecasting variables.

Results

In this section, we discuss our in-sample and out-of-sample tests to assess the predictive power of JOE and the benchmark variables for aggregate excess stock returns. Section 4.1 presents results from in-sample predictive regressions. In Section 4.2, we compare out-of-sample R2 values for the various predictors, and in Section 4.3 we discuss out-of-sample forecast encompassing tests. In Section 4.4, we examine the economic value of the predictors for real-time investors.

Conclusion

We show that JOE, measured as the ratio of total U.S. job postings to the employment level, is among the best known predictors of aggregate stock market returns. Over the period from March 1951 to December 2021, JOE is negatively related to future excess market returns and generates impressive predictive regression R2 estimates at forecast horizons ranging from one month to three years. These empirical results compare favorably with those generated by forecasting variables proposed in previous

References (85)

  • FamaE.F. et al.

    Asset returns and inflation

    J. Financ. Econ.

    (1977)
  • FerreiraM.A. et al.

    Forecasting stock market returns: The sum of the parts is more than the whole

    J. Financ. Econ.

    (2011)
  • HuangD. et al.

    Gold, platinum, and expected stock returns

    J. Financ. Econ.

    (2019)
  • JiangF. et al.

    Manager sentiment and stock returns

    J. Financ. Econ.

    (2019)
  • JondeauE. et al.

    Average skewness matters

    J. Financ. Econ.

    (2019)
  • KeimD.B. et al.

    Predicting returns in the stock and bond markets

    J. Financ. Econ.

    (1986)
  • LewellenJ.

    Predicting returns with financial ratios

    J. Financ. Econ.

    (2004)
  • LiJ. et al.

    Investor attention, psychological anchors, and stock return predictability

    J. Financ. Econ.

    (2012)
  • LudvigsonS.C. et al.

    The empirical risk-return relation: A factor analysis approach

    J. Financ. Econ.

    (2007)
  • MaioP.

    Cross-sectional return dispersion and the equity premium

    J. Financial Mark.

    (2016)
  • Petrosky-NadeauN. et al.

    Unemployment crises

    J. Monetary Econ.

    (2021)
  • RapachD.E. et al.

    Short interest and aggregate stock returns

    J. Financ. Econ.

    (2016)
  • RapachD. et al.

    Forecasting stock returns

  • StambaughR.F. et al.

    The long of it: Odds that investor sentiment spuriously predicts anomaly returns

    J. Financ. Econ.

    (2014)
  • YashivE.

    Capital values and job values

    Rev. Econ. Dyn.

    (2016)
  • YuJ.

    Disagreement and return predictability of stock portfolios

    J. Financ. Econ.

    (2011)
  • BeloF. et al.

    What drives firms’ hiring decisions? An asset pricing perspective

    Rev. Financial Stud.

    (2022)
  • BeloF. et al.

    Labor force heterogeneity and asset prices: The importance of skilled labor

    Rev. Financ. Stud.

    (2017)
  • BeloF. et al.

    Labor hiring, investment, and stock return predictability in the cross section

    J. Polit. Econ.

    (2014)
  • BollerslevT. et al.

    Expected stock returns and variance risk premia

    Rev. Financ. Stud.

    (2009)
  • CajnerT. et al.

    A Cautionary Note on the Help Wanted Online DataFEDS Notes

    (2016)
  • CampbellJ.Y. et al.

    The dividend-price ratio and expectations of future dividends and discount factors

    Rev. Financ. Stud.

    (1988)
  • CampbellJ.Y. et al.

    Stock prices, earnings, and expected dividends

    J. Finance

    (1988)
  • CampbellJ.Y. et al.

    Predicting the equity premium out of sample: Can anything beat the historical average?

    Rev. Financ. Stud.

    (2008)
  • CampbellJ.Y. et al.

    Inflation illusion and stock prices

    Amer. Econ. Rev.

    (2004)
  • CavanaghC.L. et al.

    Inferences in models with nearly integrated regressors

    Econom. Theory

    (1995)
  • CederburgS. et al.

    On the economic significance of stock return predictability

    Rev. Finance

    (2022)
  • ChenH.J. et al.

    Labor unions, operating flexibility, and the cost of equity

    J. Financ. Quant. Anal.

    (2011)
  • ChordiaT. et al.

    Index option trading activity and market returns

    Manage. Sci.

    (2021)
  • CochraneJ.H.

    Production-based asset pricing and the link between stock returns and economic fluctuations

    J. Finance

    (1991)
  • CochraneJ.H.

    Asset Pricing

    (2005)
  • CooperI. et al.

    Time-varying risk premiums and the output gap

    Rev. Financ. Stud.

    (2009)
  • Cited by (0)

    Acknowledgments: We thank Ronald Balvers, Scott Cederburg, Don Chance, Stephen Ferris, Ryan Flugum, Tyler Jensen, Raynolde Pereira, Ashish Tiwari, Sterling Yan, an anonymous referee, and seminar participants at the 2018 Financial Management Association Doctoral Student Consortium, Cornerstone Research, McMaster University, Oakland University, Richmond University, the University of Minnesota Duluth, the University of Missouri, the University of South Dakota, and the University of Wisconsin-Eau Claire for helpful comments and suggestions.

    View full text