Productivity and Pay: Is the link broken? Anna Stansbury and Lawrence Summers November 2017 Preliminary 1
50 100 150 200 Index 1973=100 Productivity and median compensation have diverged since 1973; the typical worker s compensation has stagnated Productivity and typical compensation 1940 1960 1980 2000 2020 Labor productivity Median compensation Production/nonsupervisory average compensation Data from BLS, BEA and Bivens and Mishel (2015) 2
Does this mean that raising productivity growth no longer raises average Americans incomes? although boosting productivity growth is an important long-run goal, this will not lead to broad-based wage gains unless we pursue policies that reconnect productivity growth and the pay of the vast majority Bivens and Mishel (2015) productivity gains haven't translated into broadly shared gains for the entire workforce Bunker (2015) there s no law that everybody s going to benefit from technology... Ever since the Industrial Revolution, we ve experienced a rising tide that has helped most people but... those trends have diverged Brynjolfsson (2015) 3
Two views of the productivity-compensation divergence: linkage and delinkage Delinkage Increases in productivity growth do not currently translate into additional growth in workers' compensation Linkage Productivity growth does translate into pay, holding all other factors constant - but a variety of other orthogonal factors have been putting downward pressure on median workers' compensation even as productivity growth has been acting to lift it. 4
-.02 0.02.04 Change in log, 3-year moving average Visually, productivity and compensation seem to move together though compensation growth has been slower Change in log productivity and compensation: 3 year moving avg. 1940 1960 1980 2000 2020 Labor productivity Median compensation Production/nonsupervisory average compensation Data from BLS, BEA and Mishel and Bivens (2015) 5
An empirical test of the linkage and delinkage views log compensation = α + β log productivity + γ unemployment + ε Strong delinkage : β = 0 Strong linkage : β = 1 Test with: Moving average specification (bandwidth of 1-5 years) Distributed lag specification (up to 4 lags) 6
p Results of various specifications: avg. and P/NS compensation Average compensation: 1948-2015 0.8 1.1 1.1 1.1 1.2 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) Decade dummies (3yma) No unemployment control (3yma) Nonfarm business sector (3yma) These charts show the coefficient estimates from various specifications Lines represent 95% confidence intervals. Solid line: moving average specification Dashed line: distributed lag specification p 0.0 0.5 1.5 Estimated relationship Production/nonsupervisory compensation: 1948-2015 0.3 0.8 0.8 0.8 0.0 0.5 1.5 2.0 Estimated relationship 1.1 1.1 1.2 1.2 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) Decade dummies (3yma) No unemployment control (3yma) Nonfarm business sector (3yma) 7
p Results of various specifications: median compensation Median compensation: 1973-2015 0.3 0.3 0.5 0.5 0.8 0.8 0.8 1.1 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) Decade dummies (3yma) No unemployment control (3yma) Nonfarm business sector (3yma) These charts show the coefficient estimates from various specifications Lines represent 95% confidence intervals. Solid line: moving average specification Dashed line: distributed lag specification 0.0 0.5 1.5 2.0 8
p Results of various specifications: sample split, P/NS comp. Production/nonsupervisory compensation: 1948-1973 0.2 0.3 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) Decade dummies (3yma) No unemployment control (3yma) These charts show the coefficient estimates from various specifications for regressions of average production and nonsupervisory compensation on productivity. Lines represent 95% confidence intervals. Nonfarm business sector (3yma) -0.5 0.0 0.5 1.5 Estimated relationship Production/nonsupervisory compensation: 1973-2015 Single year only p Two-year moving average Three-year moving average Four-year moving average Five-year moving average -0.5 0.0 0.5 1.5 Estimated relationship 0.5 Time trend (3yma) Decade dummies (3yma) No unemployment control (3yma) Nonfarm business sector (3yma) 9
Results: percentiles of the wage distribution 0.5 10 th percentile 20 th percentile 30 th percentile 40 th percentile These charts show the coefficient estimates from the baseline 3-year moving average regressions of percentiles of the wage distribution on productivity. Lines represent 95% confidence intervals. 50 th percentile 60 th percentile 70 th percentile 0.5 80 th percentile 90 th percentile 95 th percentile -0.5 0.0 0.5 1.5 Estimated relationship Note: this is wage data not total compensation data. Comparison with the median compensation regressions suggest that these may be underestimates. 10
Results: average compensation for advanced economies 1.1 1.1 Canada 1972-2015 France 1972-2015 West Germany 1972-1990 These charts show the coefficient estimates from the baseline 3-year moving average regressions of average compensation on productivity in advanced economies. Lines represent 95% confidence intervals. 0.0 Germany 1993-2015 0.3 Italy 1985-2014 Japan 1995-2015 United States 1950-2015 0.0 2.0 Estimated relationship 11
Index 1973=100 100 120 140 160 180 The overall productivity-compensation gap is the result of three separate divergences Divergence decomposition (Bivens and Mishel 2015) Labor share Deflator wedge Mean-median inequality 1970 1980 1990 2000 2010 2020 Year Net labor productivity Mean comp., CPI deflation Mean comp., NDP deflation Median comp., CPI deflation Data from BLS, BEA and Mishel and Bivens (2015) 12
Technology-based theories have a natural implication: divergence is greater when productivity growth is faster Assuming that greater technological progress implies faster productivity growth: Labor share: log labor share = α + β log productivity + γ unemployment + ε H 1 :β <0 Mean/median inequality: log mean compensation = α + β log productivity + γ unemployment + ε median H 1 :β >0 13
Divergences increased less during productivity booms than productivity slowdowns 14
p Results: labor share and productivity 1973-2015 -0.3-0.2-0.0 0.0 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) These charts show the coefficient estimates from various specifications Lines represent 95% confidence intervals. Solid line: moving average specification Dashed line: distributed lag specification Decade dummies (3yma) No unemployment control (3yma) -0.2 Nonfarm business sector (3yma) - -0.5 0.0 0.5 15
p Results: mean/median compensation and productivity 1973-2015 0.2 Single year only Two-year moving average Three-year moving average Four-year moving average Five-year moving average Time trend (3yma) These charts show the coefficient estimates from various specifications Lines represent 95% confidence intervals. Solid line: moving average specification Dashed line: distributed lag specification Decade dummies (3yma) No unemployment control (3yma) Nonfarm business sector (3yma) - -0.5 0.0 0.5 16
A quick counterfactual: productivity growth and widening inequality If the mean/median compensation ratio had been the same in 2015 as it was in 1973, median compensation would have been around 32% higher If the productivity/mean compensation ratio (labor share) had been the same in 2015 as it was in 1973, mean and median compensation would have been around 5% higher If the rate of productivity growth had been the same over 1973-2015 as it was over 1948-1973, our estimates of the relationship between productivity and compensation suggest that mean compensation would have been around 59-76% and median compensation 65-68% higher in 2015. 17
Conclusions: The substantial variations in productivity growth that have taken place during recent decades have translated into significant positive changes in the compensation of middle income workers. This suggests that if productivity accelerates holding other factors constant, the likely impact will be increased pay growth for middle income workers. At the same time, there is little co-movement between productivity growth and widening inequality either for the labor share, or the mean/median ratio. This tends to imply that technology is not the key driver of changes in the labor share, or in the mean/median compensation gap. It instead suggests the importance of factors not associated with the rate of productivity growth in explaining the pay-productivity divergence. 18