The US labor share of income: what shocks matter?

Abstract

We propose a novel methodological approach to disentangle the main structural shocks affecting the US labor share of income. We motivate an SVAR model to derive four structural shocks: aggregate demand, labor supply, shocks to wages, and productivity; and quantify the dynamic responses of the labor share to each structural shock. We find substantial differences between the immediate post-war era and the neoliberal period. In order of magnitude, the labor share responded mainly to productivity, aggregate demand, and shocks to wages during the immediate post-war era; whereas shocks to wages, productivity, and to aggregate demand mattered most during the neoliberal era. These effects are statistically significantly different across the two periods only for wage and productivity shocks. Increased (decreased) sensitivity to wage (productivity) shocks during the neoliberal period suggests that the decline in the labor share is mainly driven by the factors that govern wage setting.

Keywords:

• US labor share of income
• structural shocks
• neoliberalism

JEL Classifications:

• B50
• E12
• E24
• E25
• E32

Acknowledgements

We are deeply grateful to two anonymous reviewers and Deepankar Basu for their valuable comments and suggestions. We also thank participants at conferences and seminars in Boston and Salt Lake City, where the current research was presented. The usual disclaimer applies.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 To label the period post-Volcker shock the ‘Great Moderation’ or ‘neoliberal era’ is standard; discussion and references are supplied further below. The period from 1948 to 1973 is often labeled a ‘golden age’, but since we include the years of crises throughout the 1970s and early 1980s, we label it simply as the immediate post-war era.

2 The labor share is, of course, definitionally related to the level of output, employment, real wages and labor productivity. Our approach, however, does not offer a decomposition of the labor share, as the latter cannot be simply recovered using the four variables employed.

3 We should also emphasize that financialization and globalization (among other factors) are likely to have played a role in the pronounced changes in the distribution of income in advanced countries (as discussed by other papers reviewed in Section 2); however, it is difficult to identify the structural shocks derived from these conceptualizations. In this sense, our approach only provides a direct test on the relative role of ‘wage setting’ vs. ‘technology’ in the decline of the labor share.

4 This pattern is robust across various studies even when top wages and salaries are included in the calculation of respective averages. Since executive compensation packages through the inclusion of boni and stock options incorporate what should more properly be accounted for as profits, the decline in the ‘true’ labor share would be more pronounced, and the measured gap between real wage and productivity growth wider.

5 As the analysis put forth pertains to the short-run, movements in the output series are typically considered to emanate from the demand side (as is standard in neoclassical synthesis and also old, new and post-Keynesian thinking); while we follow the discussion provided by Foroni et al. (2018), who showed that the incorporation of both the unemployment rate and real wages in a VAR model aids in the identification of labor supply shocks and wage (bargaining) shocks. Corresponding to this standard line of thinking, the structural shocks identified in each of the equations are labeled as suggested here (see also footnote 10).

6 Our methodology is similar to the one proposed by Kilian (2009), who also followed a two-step econometric procedure based on a SVAR model and the use of IRFs to identify the underlying demand and supply shocks in the global crude oil market and to estimate their respective macroeconomic effects. Of course, our research question is entirely different.

7 The literature is voluminous, and we will not review it here in detail. For important influences, see Flaschel (1993) and Barbosa and Taylor (2006). For a recent book treatment, see Blecker and Setterfield (2019, Chapter 5). Here and in subsequent sections, the reader might recall that our target in this paper is not to directly speak to questions on the nature of the growth regime, but instead speak to questions regarding critical factors in the evolution of the labor share.

8 The model could potentially be extended to include an interest rate rule since monetary policy played an important role in the onset of recessions, especially during the immediate post-war period (see, for example, Proaño et al., 2006). The starting point here, however, is neo-Goodwinian theory, in which the interest rate is not a variable of central importance; while the interaction between production and demand on the one hand and social conflict over the distribution of income on the other take precedence. Also, the inclusion of an interest rate creates additional layers of complexity in order to identify the system as the evolution of the inflation rate should also be incorporated. We leave this topic for future research.

9 Namely: (a) static and dynamic returns to scale associated with increases in the volume of output and the technical progress embodied in capital accumulation; (b) macro-increasing returns in Young (1928)'s sense: an initial demand expansion leads to a series of changes that propagate themselves in a cumulative way; and (c) the learning-by-doing process, which means that labor productivity is a function of cumulative output.

10 It is worth pointing out that our choice of variables contained in ${\mathbf{z}}_{\mathbf{t}}$ is similar to that of Foroni et al. (2018). However, our research question and empirical identification strategy differ substantially from theirs, since our main interest is to determine the structural drivers of the labor share of income, rather than to separate labor supply and wage (bargaining) shocks. Furthermore, they incorporate prices (or inflation) in a VAR model to identify demand shocks, while we consider an actual measure of real GDP preferable. Similarly, Basu and Gautham (2019) put forth a VAR to investigate the dynamics between the labor share and other key macroeconomic aggregates. As with Foroni et al. (2018), their research question critically differs, since their focus lies on the effect of a shock of the labor share itself on economic activity variables. Still, careful consideration of the IRFs presented in Basu and Gautham (2019) suggests that their results are – broadly speaking – consistent with our first-step SVAR results.

11 Note that restrictions on $\mathbf{A}$ assume restrictions regarding the structure of contemporaneous feedback between the variables; while restrictions on $\mathbf{B}$ take on the form of assumptions with respect to the correlation structure of the errors in the SVAR.

12 Note that, although our target here is not to add to the debate on demand regimes, the first-step SVAR estimation implies, in line with the post-Keynesian literature, a contemporaneous impact from real wage rate and labor productivity – the main constituent variables of the labor share – on the aggregate demand variable.

13 It is important to mention that our identification strategy does not correspond to the well-known Cholesky scheme, which creates a recursive contemporaneous ordering among the variables. Because of this, the relevant IRFs obtained from the first-step SVAR estimations (reported in Figures 1 and 4) do not necessarily start from zero.

14 We introduced twelve lags in the estimation of (9(9) ${\mathrm{\Psi }}_t={\text{β}}_j+\sum _{i=0}^{12}{\delta }_{j,i}{\hat{\epsilon }}_{j,t-i}+v_{j,t},j=1,2,3,4,$(9) ) in order to incorporate approximately three years of data, as in Kilian (2009). The great majority of the regression models estimated according to (9(9) ${\mathrm{\Psi }}_t={\text{β}}_j+\sum _{i=0}^{12}{\delta }_{j,i}{\hat{\epsilon }}_{j,t-i}+v_{j,t},j=1,2,3,4,$(9) ) – with variables as stationary series (Section 4), or in levels (Section 5) – do not show serial correlation problems at the 5% level of significance when twelve lags were included; and none of these models presented autocorrelation problems at the 10% level of significance.

15 This allows us to focus on the short-run dynamics of the labor share without making assumptions concerning any long-run equilibrium, which would require the estimation of an error correction model and the use of cointegration analyzes. Section 5 presents results with all variables in levels, showing that the main results are robust to this change in specification.

16 An estimation in which real wages and productivity are used separately could potentially lead to a composition bias (see also Cantore et al., 2020). In brief, the composition bias in the labor force arises because less productive (low wage) workers tend to exit the labor force during recessions and enter during expansions, which makes the average wage less pro-cyclical since there is a larger share of the wage bill going to low wage workers during recessions. This bias would not affect directly the labor share as it is a ratio of two potentially biased responses (productivity and wages); but it can affect the real wage and productivity when evaluated separately. The latter can become a problem if the results show counter-cyclical responses of real wages and productivity: a recession (expansion) would lead to a rise (fall or less pro-cyclical response) of real wages or productivity. However, the results for the first period show that a positive shock to aggregate demand (that is, an expansion) leads to a rise in real wages and has no effect on productivity, so the bias can only reinforce our results since you would remove the impact of low wage/low productivity workers on the averages. In the second period, the results show that a positive shock to aggregate demand leads to no effects on either real wages or productivity, which again means that the composition bias does not seem to be important in our results. Cantore et al. (2020) also found that the composition bias did not affect their estimations. We thank an anonymous referee for drawing our attention to this point.

17 The four structural shocks employed as regressors, ${\hat{\epsilon }}_{j,t}$, j = 1, 2, 3, 4, are shown in Figure A1 in Appendix 1. Based on the Jarque-Bera normality test, these shocks are normally distributed, except for productivity shocks during the neoliberal era.

18 We also tested whether the structural shocks can be treated as predetermined with respect to $\mathrm{\Delta }{\psi }_t$, following Kilian (2009): first, an AR model for $\mathrm{\Delta }{\psi }_t$ with three lags – as in the SVAR – is estimated; second, the contemporaneous correlation between the residuals obtained therefrom and the structural shocks is calculated. The great majority of the correlations are low (i.e. below 50%), with the exception of productivity shocks during the post-war era (59%), and wage shocks during the neoliberal era (64%).

19 An alternative comparison between moving average representations and local projections to recover impulse responses in the presence of persistence in the shock is provided by Alloza et al. (2019), who find that both methods treat persistence differently (namely, standard local projections identify responses that include an effect due to the persistence of the shock, while moving average representations implicitly account for it). They propose the inclusion of leads of the shock in local projections in order to control for its persistence, which renders the resulting responses equivalent to those associated to counterfactual non-serially correlated shocks.

20 Figure A2 in Appendix 2 plots a direct comparison of the responses of $\mathrm{\Delta }{\psi }_t$ obtained from both LPs and VAR IRFs.

21 Figure A3 in Appendix 2 also plots the IRFs of ${\psi }_t$ to the shocks obtained using LPs and VAR.

22 During the neoliberal era, the response of the labor share is statistically significant for only one quarter, except for aggregate demand shocks, which have statistically significant effects also in quarters six through twelve.

Additional information

Notes on contributors

Ivan Mendieta-Muñoz

Ivan Mendieta-Muñoz is Assistant Professor in the Department of Economics at the University of Utah. He received his Ph.D. in Economics in 2016 and his M.Sc. in Economics and Econometrics in 2012 from the University of Kent. His primary research fields are: Macroeconomics and Financial Economics; Time Series Econometrics and Statistical Methods; and Economic Development in Latin America. His work has appeared published in academic outlets such as Competition and Change, Economics Bulletin, Economics Letters, International Review of Applied Economics, Journal of Post Keynesian Economics, Macroeconomic Dynamics, Metroeconomica, and Review of Income and Wealth, among others. He has received various awards, recognitions and grants for his academic work.

Codrina Rada

Codrina Rada is Associate Professor in the Department of Economics at the University of Utah. She received her Ph.D. in Economics from the New School for Social Research in 2007. Her current research gravitates around issues of economic growth and income distribution, with a specific interest in the effects of structural change and global economic integration on macroeconomic dynamics. Over the years she has also published theoretical and empirical work that explores the economics of pensions and aging, and structural transformation in emerging economies. Her work has appeared in Review of Income and Wealth, Cambridge Journal of Economics, Columbia University Press, Journal of Policy Modeling, Agricultural Economics, Metroeconomica, Structural Change and Economic Dynamics, and Development and Change, among others.

Márcio Santetti

Márcio Santetti is Ph.D. candidate in the Department of Economics at the University of Utah. His current research interests are technological change and its effects on economic growth, income distribution, and the environment. He received a Master's degree in Development Economics from the Pontifícia Universidade Católica do Rio Grande do Sul (Brazil), and his work has appeared in outlets such as CEPAL Review and Economia e Desenvolvimento.

Rudiger von Arnim

Rudiger von Arnim is Associate Professor in the Department of Economics at the University of Utah. He received his Ph.D. in Economics from the New School for Social Research in 2008. His research focuses on macroeconomics, specifically the linkages between growth, the business cycle and the distribution of income and the ex-ante assessment of international trade agreements on developing economies with structural heterogeneity. On these themes, he has extensively worked with ÖFSE (Austrian Foundation for Development Research) and also G24, UNDP, UNCTAD, and ILO's Institute for International Studies (IILS).