A New Interpretation Of Productivity Growth Dynamics In The Pre-Pandemic And Pandemic Era U.S. Economy, 1950-2022

“….Has the cyclical behavior of TFP changed through time to echo that of labor productivity? Table 3 repeats the productivity gap change regressions of Table 2 with the change in the TFP gap substituted as dependent variable for the change in the labor productivity gap. The main results of Table 2 are replicated here for the same three sample periods of 1950-85, 1986-2006, and 2007-2019, where the 2007-19 results repeat the recovery treatment of the right side of Table 2. The current-quarter TFP gap response to a change in the output gap is strongly positive — 0.71, 0.81, and 0.88 across the three intervals. The long-run response of TFP is significantly positive in all three periods, although substantially higher in the first (0.50) and third (0.66) intervals than in the middle interval (0.39). The long-run recession responses are insignificant in the first two periods but a highly significant -0.52 in the third period. The constant terms for all three periods are insignificant and close to zero….”

A Post-Sample Simulation to Detect a Productivity Growth Revival in the Pre- Pandemic Economy, 2017-19

"...For perspective on the magnitude and trajectory of 2017-19 productivity growth behavior, we first perform the same procedure on the productivity growth revival that we know occurred in the late 1990s. To match the three-year duration of the 2017-19 experience, we examine only the first three years of the late 1990s revival during 1996-98. Figure 5 illustrates how our procedure reveals a revival of productivity trend growth during this three-year interval. We re-estimate the change gap version of our labor productivity equation for the business sector from 1986 to 1995 (instead of 1986-2006 as in Table 2) and calculate the predicted growth rate of productivity during 1996 to 1998, holding constant throughout the 1996-98 interval the value of the Kalman productivity growth trend at its 1995:Q4 value of 2.13 percent per year. The difference between the actual and predicted growth rates of productivity measures the estimated increase in trend relative to the 1995:Q4 starting point. igure 5 extends from 1990:Q1 to 1998:Q4 and plots in red a four-quarter moving average of the actual growth rate of output and in green displays a similar average of the actual growth rate of productivity. The black line through 1995:Q4 is the four-quarter moving average of the predicted productivity gap change from the 1986-95 equation plus the Kalman trend that was used to calculate that gap, hence the predicted value for productivity change (gap plus trend). The continuation of the black line from 1996:Q1 to 1998:Q4 is the calculated prediction of the equation when the Kalman trend is fixed at its value of 2.13 percent for 1995:Q4. The fitted values do a relatively good job of predicting the change in productivity growth during the 1990-95 interval shown in Figure 5 to the left of the vertical bar; in particular the 1992-93 early recovery “hump” following the 1990-91 recession is tracked very closely. This was the episode christened at the time as “the jobless recovery” (Gordon, 1993).

"...We now turn to an interpretation of productivity behavior in the nine quarters between 2020:Q1 and 2022:Q1, the “pandemic economy.” This includes the period of the sharp lockdown of the economy in 2020:Q2, the rapid but partial opening up in 2020:Q3, and the subsequent recovery of the following six quarters. We are immediately confronted by a marked departure from the procyclical relationship between the productivity and output growth gaps that characterized the 2010-19 decade. As output collapsed in 2020:Q2, business- sector productivity growth moved in the opposite direction, rising at an annual rate of 6.6 percent in that quarter.The apparent countercyclical behavior of productivity in 2020:Q2 reflected not a sudden outburst of creative innovation, but rather in large part a shift in the mix of output and employment toward higher productivity sectors of the economy. While output and employment dropped everywhere, they declined at a much faster rate in the sectors where labor productivity and wages are relatively low. This shift in the industry mix reflects a relatively large decline in the employment of workers in low-paid industries where work involves close contact among employees, customers, or both, such as bricks-and-mortar retail trade and leisure/hospitality, and a relative increase in the employment of workers who could continue to work at home in relatively high-paid industries such as finance and information technology. The importance of this mix shift is highlighted by the enormous 9-to-1 difference in the level of productivity between the highest and lowest industries, as we see in Table 6 below….”

Robert Gordon and Hassan Sayed, "A New Interpretation Of Productivity Growth Dynamics In The Pre-Pandemic And Pandemic Era U.S. Economy, 1950-2022,"National Bureau Of Economic Research, July 2022, https://www.nber.org/papers/w30267

“…The first two columns of Table 1 provide the annual growth rates of labor productivity for both the total economy and the business sector, where total output is measured by the geometric average of GDP and gross domestic income (GDI), following the analysis of Nalewaik (2011), who showed that the average of GDP and GDI is more accurate than either one examined separately. This distinction turns out to be important for pandemic-era growth in 2020-22, when GDI grew considerably faster than GDP, implying that productivity growth rates based on the average of GDP and GDI, as in Table 1, outpaced by a substantial margin the BLS-published productivity growth rates calculated with GDP as the sole measure of output. Thus, the Table 1 business growth rate for 2020-22 of 2.11 percent contrasts with the BLS- published growth rate that is a much slower 1.15 percent. There is no published BLS equivalent of our data for the total economy in the first column, as our numbers are based on the average of aggregate GDP and GDI divided by an unpublished BLS series on hours in the total economy….As expected, the growth rate for the business sector in the second column is always more rapid than in the total economy, due to the relatively slow measured productivity growth for those portions of the economy excluded from the business sector, notably the government and nonprofit sectors (shown separately in the third column). The shortfall of productivity growth in the total economy compared to the business sector was similar in 2010-16 (0.24 percent) and in 2017-19 (0.41 percent). This difference was a much larger 0.70 percent during the dot.com era of 1996-2004. Thus, the slowdown from 1996-2004 to 2010-19 was substantially greater for the business sector than for the total economy. Recall that it is productivity growth in the total economy, not just the business sector, that matters for the growth of potential output and real income per capita, i.e., the standard of living….The contrast between the total economy and business sector was reversed in 2020-22. The revival from 2017-19 to 2020-22 was an impressive 0.81 percentage points in the total economy compared to 0.64 in the business sector. As a result, productivity growth in the non- business sector jumped from -0.18 to 1.16 percent, and this is the fastest productivity growth registered by the non-business sector since 1950-72. Why? We will return to this puzzle in a subsequent section which examines the widely differing pandemic-era productivity experiences of particular industries within the business sector. Shown in the fourth column of Table 1 are the growth rates for the manufacturing sector, which is notable because unlike the total economy or the business sector, it had virtually no slowdown from 1950-72 to 1973-95. During that interval of more than two decades, healthy productivity growth in the manufacturing sector contrasted with a disappointing record for the rest of the economy. The manufacturing sector also enjoyed a robust revival in excess of two percentage points during the dot.com interval. But then the tables were turned as the manufacturing sector suffered a stunning 4.93 percentage point growth slowdown from 1996- 2004 to 2010-19, much larger than the slowdown in either the total economy or the business sector. The unusually poor relative performance of the manufacturing sector after 2009 has not yet received much attention in the productivity growth literature. The 2020-22 bounce-back of manufacturing productivity growth is much greater than for the total economy or the business sector. Further insight into the performance of the business sector is provided in the fifth column, which lists growth in total factor productivity (TFP). The growth in TFP is usually smaller than that of labor productivity, differing by the normally positive contribution of capital deepening, with a minor additional contribution from improving labor quality. TFP growth was less than labor productivity growth in the business sector by 1.02 points in 1950-72 and a similar 1.00 points in 1972-95 but by a larger 1.50 points in 1996-2004, reflecting the marked rise in the growth rate of investment (i.e., capital deepening) in the dot.com era. Somewhat surprisingly, the TFP shortfall rose to an even larger 1.95 percent in 2005-2009 before dropping to 0.34 percent in 2010-16, 0.71 in 2017-19, and a similar 0.84 in 2020-22. The growth revival from 2010-16 to 2017-19 for TFP of 0.22 percent was substantially less than the revival of 0.59 percent for business-sector productivity, indicating that capital deepening contributed to the 2017-19 revival. The further revival of TFP growth from 2017-19 to 2020-22 of 0.51 percent was somewhat less than the 0.81 percent rebound of business-sector productivity growth. We return in Table 5 below to a detailed decomposition of the difference between business-sector labor productivity and TFP growth in 2010-16, 2017-19, and 2020-22….”
Pre-Pandemic Trends of Labor Productivity and TFP

"...The first step in our analysis is to estimate the underlying growth trends of labor productivity and TFP in the pre-pandemic economy. Our preferred method is the Kalman filter which removes correlations between any time series and the unemployment gap, defined as the difference between the unemployment rate and the “long-run NAIRU” estimated by the Congressional Budget Office. Thus, if a decline in productivity growth occurs at the same time as a rise in the unemployment gap, as during the 1981-82 recession, the Kalman procedure uses that correlation to eliminate the decline in productivity growth and instead show a smooth evolution of the productivity growth trend. In contrast, the commonly used Hodrick-Prescott (H-P) filter is univariate and smooths a series using only information on deviations from average growth of the series itself, without any “outside information” on the behavior of the unemployment gap. In practice the H-P filtered series for productivity growth still exhibits substantial fluctuations at business cycle frequencies....Our examination of trends in this section terminates the data in 2019 and leaves the interpretation of 2020-22 for later in the paper, where we deal with unique aspects of the evolution of productivity during the pandemic era. Figure 1 contrasts our Kalman filter for business sector productivity growth, shown in blue, with two alternative series. One is a simple 20-quarter moving average of one-quarter annualized changes, shown in yellow, and the other, plotted in green, is the H-P filtered series based on the same data (using the standard H-P quarterly smoothing parameter of 1600). Note that the H-P filter retains considerable cyclical sensitivity and appears to be a slightly smoothed replica of the 20-quarter moving average. For reference, shown on the right side of Figure 1 as a horizontal line to the right of the black vertical bar is the average annual growth rate of business-sector productivity for the nine quarters between 2020:Q1 an 2022:Q1...."

"... Table 1 introduces the behavior of productivity (Y/H) growth from 1950 to 2022. The rows divide this long span of seven decades into separate intervals that mark off different eras of productivity growth. First comes the rapid growth era of 1950:Q1-1972:Q4, followed by the slowdown interval of 1973:Q1-1995:Q4 that received so much attention during the 1970s and 1980s. Next comes the return of rapid growth during the “dot.com” era between 1996:Q1 and 2004:Q4, a revival usually attributed to the effects of high investment in the information technology revolution and the impact on TFP of personal computers, the internet, and search engines...The subsequent post-2004 interval is divided into four sub-periods. The first of these, 2005:Q1-2009:Q4, includes the last stages of the 2001-07 economic expansion, the 2007-09 recession, and the first two recovery quarters after the end of the recession. We treat 2010:Q1- 2016:Q4 as a distinct sub-period marked by unusually slow productivity growth and then break out the 12 quarters of 2017-19 that show a modest but noticeable revival. The last line of Table 1 includes the nine pandemic quarters 2020:Q1-2022:Q1, with a further acceleration beyond 2017-19 in every column of the table….”

Here is the key table

The Growth Rates of Labor Productivity and TFP by Major Sector

The paper begins by contrasting the dismal decade of 2010-19, when business sector productivity growth proceeded at only 1.1 percent per year, with the ebullient nine years 1996- 2004 when productivity growth reached 3.3 percent or the entire 1950-2009 postwar era with its productivity growth achievement of 2.5 percent. Could the slow growth of the 2010-19 decade imply another disappointing decade in the 2020’s or might the 2020’s become a decade of robust growth propelled by technological breakthroughs, robots, and artificial intelligence? The decade opened on an optimistic note with business-sector productivity reaching 4.1 percent during the four quarters of 2020. This paper provides a new interpretation of variations of productivity growth for the entire postwar era with particular emphasis on the 15 years since the start of the financial crisis recession of 2008-09. Our first insight is that short-term productivity changes during recessions and business expansions, and from one quarter to the next, do not reflect autonomous gyrations of productivity growth itself, the so-called “productivity shocks” that have played such a large role in the macroeconomics literature. Instead the behavior to be explained is the response of the change in hours of work to autonomous movements in output, which themselves are driven by demand-induced fluctuations in consumption, fixed investment, government spending, net exports, and particularly short-run changes in inventory accumulation. Output growth proceeds with sharp irregular quarter-to-quarter ups and downs during both expansions and recessions, with additional sustained downward movements during recessions. Hours respond partially and gradually to these output changes, and productivity changes are a residual, the definitional difference between output changes and the induced hours changes. Because of the gradual adjustment of hours, the residual productivity changes typically involve a large positive response in the current quarter followed by negative reactions in the next few subsequent quarters as hours complete their adjustment. If the sum of the positive initial productivity response and the negative lagged responses is significantly positive, then productivity growth is said to be procyclical.We concur with the previous literature that productivity growth was strongly procyclical during 1950-85; our coefficient of the response of productivity growth to output changes over the current quarter and four lagged quarters is a highly significant 0.28, similar to the one-third response embedded in the original formulation of Okun’s Law. With our focus on the gradual adjustment of hours, we make a distinction between the normal response of hours and an extra reaction to output changes that occurs in recessions and was particularly marked in2008-09. During that recession hours declined with a much higher elasticity to output changes than in other periods, which we interpret as “excess layoffs” due to the panic of business firms as output collapsed during and after the financial crisis of 2008. Since productivity change is a residual, equal to output change minus hours change, it behaved counter-cyclically and rose in the four quarters of 2009 by a massive 6.4 percent. This counter- cyclical episode when combined with non-recession data has led much of the previous literature to conclude that productivity growth was no longer procyclical after 1985. The excess layoffs of 2008-09 did not occur in isolation; eventually the lost jobs returned as workers were rehired in the post-recession recovery. We model the total recession and recovery effect as netting out to zero, so the cumulative extra rehiring addition of jobs in the recovery exactly offsets the estimated excess reduction of jobs during the recession. Because the excess adjustment of hours in 2008-09 was a special phenomenon, we divide up our regression analysis into three eras (1950-85, 1986-2006, 2007-19) rather than two divided at 1986. The set of current and lagged coefficients on our combined recession-recovery variable is highly significant for the 2007-19 interval but not for the earlier periods. As for the “normal” non- recession long-term response of productivity to output changes, when the current quarter response is combined with four lagged quarters, we conclude that procyclicality did not disappear after 1985. Our long-run sum of coefficients is 0.28 for 1950-85, 0.20 for 1986-2006, and 0.66 for 2007-19 (although the 0.20 for the middle interval is insignificant). Our novel recession/recovery treatment has six important implications that resolve several of the puzzles about productivity behavior that have emerged both in the pre-pandemic interval of 2007- 19 and in the pandemic era of 2020-22. First, the estimated coefficients for the 2008-09 recession/recovery effect imply that average annual productivity growth in the two years 2008-09 would have been -0.8 percent instead of the actual 3.2 percent if the excess layoff phenomenon had not occurred. Second, our estimated recession/recovery coefficients imply that a major explanation for slow productivity growth in 2010-16 was the extra post-recession rehiring that offset the excess hours reduction that had previously occurred in the recession. We estimate that without that post-recession rehiring spread out over the seven years 2010-16, the annual rate of productivity growth on average would have been almost twice as high — 1.7 percent per year rather than 0.9 percent per year. Third, the appearance in the productivity data of a growth revival from 2010-16 to 2017- 19 is more than explained by the winding down of post-recession recovery rehiring that had characterized the 2010-16 period. Fourth, when allowance is made for the recession/recovery effect, the regression coefficients indicate that productivity growth was strongly procyclical in 2007-19. Regarding the pandemic era, our fifth conclusion emerges when our estimated 2007-19 regression coefficients are applied in a post-sample simulation to actual data for 2020-22. We find that our coefficients are able to track almost exactly the marked acceleration of productivity growth during 2020, once adjustments are made for the shorter duration of the 2020 recession in comparison with the 2008-09 recession. Sixth, our post-recession rehiring treatment in the same simulations explains why productivity growth slowed so markedly from 4.1 percent at an annual rate in the four quarters of 2020 to only 0.6 percent in the following five quarters of 2020- 22.Further insight into productivity behavior in 2020-22 is provided by our new quarterly data base of productivity levels and changes for 17 separate industries in the private business sector extending from 2006:Q2 to 2022:Q1. The numerator of our industry indexes is real- value-added, that is, the business component of GDP. The aggregate growth rate of the industry indexes in 2020-22 is 1.3 percent, which is slower than the 2.1 percent growth rate of the business-sector data used elsewhere in the paper for which the numerator is the average of GDP and GDI (GDI grew considerably faster than GDP in 2020-22). Our first insight from the quarterly industry data is that published productivity indexes greatly exaggerate aggregate productivity growth in 2020:Q2 and understate it in 2020:Q3 as a result of the lockdown’s effect in causing sharp shift in the industry mix away from industries like restaurants and hotels with low levels of productivity toward work-from-home industries like information and financial services that have high levels of productivity. We illustrate this distortion by creating an alternative aggregate of the 17-industry data that holds constant industry output weights at the 2019 level. Because 2020-22 productivity growth across the 17 industries is so heterogeneous, we combine these industries into three groups: goods, work-from-home (WFH) services, and contact services. The three groups behave very differently. For the nine quarters of 2020-22 taken together, productivity growth in the goods group is near zero, in WFH services is a strongly positive 3.3 percent, while in contact services is a mediocre -2.6 percent. The average across the three groups is 1.3 percent. Comparing the four quarters of 2020 with the five quarters of 2021-22, productivity growth in the goods group declines from strongly positive to strongly negative, which is consistent with the hypothesis of excess layoffs followed by a gradual rehiring recovery. In WFH services productivity growth actually improved from 2020 to 2021-22, suggesting a more permanent phenomenon. In contact services productivity growth remained negative by about the same amount in both 2020 and 2021-22. Thus it appears that the apparent revival of productivity growth in 2020-22 was entirely caused by the outstanding performance of the WFH services. We cite A survey study showing that the WFH respondents assess their own productivity as substantially higher than their expectations, which may provide a comparison between productivity of WFH activity compared to the productivity of the same individuals in their previous office environments. We cite another study suggesting that pandemic-era GDP growth may be understated by neglecting the shift of residential capital from non-work to work activities and the large personal investment in technology hardware and communications software needed to make WFH effective. Where does this leave our assessment of the long-run trend in productivity growth? The paper has suggested that productivity growth was distorted by excess layoffs in 2008-09, shifting some productivity growth into that two-year period and moving it away from 2010-16 when rehiring cancelled the effects of the excess layoffs. The business-sector data displayed in Table 1 and used in our regressions and simulations records growth rates of 1.3 percent in 2005- 07, 3.2 percent in 2008-09, and 0.9 percent in 2010-16. The underlying trend of productivity growth is better represented by averaging across these three intervals, resulting in a 2005-16 growth rate of 1.4 percent. This is remarkably close to the 1.5 percent in the apparent revival interval of 2017-19 and also close to the 1.5 percent achieved in the long “slowdown” period of 1973-95. A cautionary note is that a suggestion of 1.5 percent for the long-term growth of productivity in the business sector translates into only 1.1 percent for the total economy, since productivity in the total economy grew an average of 0.4 percent slower than in the business sector during 1950-2019. This 1.1 percent suggestion is relevant for predictions of future potential output growth and is slower than the current CBO ten-year forecast of 1.4 percent total-economy productivity growth. Our conclusion is thus consistent with the long-term forecast of 1.2 percent for the total economy included in our previous long-term evaluationsEconomic commentary in mid-2022 is dominated by speculation about the imminence of a new recession. There is much puzzlement about the reality of negative GDP growth in the first quarter of 2022 and the likelihood of negative GDP growth again in the second quarter, juxtaposed with continued robust growth in payroll employment. Our paper resolves this puzzle in its post-recession recovery simulation which indicates that hours growth in 2021-22 is 2.5 percent faster at an annual rate in each quarter than otherwise, due to rehiring of workers to replace those who lost their jobs in the excess layoffs of 2020. The paper predicts also that continued rehiring will result in significant negative productivity growth in 2022:Q2 and likely in subsequent quarters, having the effect of further reducing the 2020-22 average productivity growth rate down from 2.1 percent closer to the 1.4 percent average of 2005-19. This assessment leaves no room for a pandemic-era revival in productivity growth as has been widely suggested. Instead, there appears to be a consistent growth rate of 1.4 percent (2005-19) roughly equal to 1.5 percent (1973-95), leaving the dot.com achievement of 3.3 percent as a historic outlier as it recedes further into the past.

Their Takeaway

The Mechanics Of 2020 Uptick

"...How do productivity changes in the three industry groups compare when the nine quarters of 2020-22 are compared with the previous ten years, 2010-19? This comparison is based on fixed output-weight (FYW) indexes for each group and is shown in Figure 10. During 2010-19 average productivity growth was relatively similar across the three groups, with respective growth rates of 0.6, 1.1, and 0.6 percent, implying an average of 0.8 percent annual growth for the FYW index of all three groups taken together. The experience in 2020-22 could not have been more different. Productivity growth in the goods group slowed slightly from 0.6 percent in 2010-19 to 0.4 percent in 2020-22. But productivity growth for WFH services soared from 1.1 percent in 2010-19 to 3.3 percent in 2020- 22. In contrast the performance of contact services slumped from 0.6 to -2.6 percent. Changes between the two intervals are plotted separately in the bottom section of Figure 10. Growth for the total of all three groups increased from 0.8 to 1.3 percent (recall again that the industry data are based on the GDP measure of productivity and exhibit slower growth in 2020-22 than the business-sector data used in Table 1 and our previous regression/simulation analysis based on the average of GDP and GDI in the numerator of productivity)...."

Key takeaway, "...The dismal decade of 2010-19 recorded the slowest productivity growth of any decade in U.S. history, only 1.1 percent per year in the business sector. Yet the pandemic appears to have created a resurgence in productivity growth with a 4.1 percent rate achieved in the four quarters of 2020. This paper provides a unified framework that explains productivity growth in both the pre-pandemic and pandemic-era U.S. economy. The key insight is that in their panicked reaction to the collapse of output in the 2008-09 recession, business firms overreacted with “excess layoffs,” adjusting hours to the output decline with a far higher elasticity than normal. Our regression analysis, which allows post-recession rehiring that gradually unwinds the excess layoffs, explains why productivity growth was countercyclical in 2009 and why it was so slow in 2010-16 as rehiring boosted hours growth. Post-sample simulations explain why productivity growth was so high in 2020 and why it fell to only 0.6 percent in the five quarters of 2021-22. The paper includes implications for the future long-term evolution of productivity growth in the business sector and total economy. A new data file on quarterly productivity levels and changes for 17 industries provides new perspectives for 2006-22 and particularly for the nine pandemic quarters of 2020-22. Positive pandemic-era productivity growth can be entirely explained by a surge in the performance of work-from-home service industries, while goods industries soared and then slumped, while contact services recorded strongly negative productivity growth throughout 2020-22...."