Productivity Growth and Workers Job Transitions: Evidence from Censal Microdata

Using evidence from Chile Chad Syverson finds that when workers change firms they generally move from lower to higher productivity employers raising aggregate productivity, but that mean hides a great deal of churn in both directions. They use earnings as a proxy for skill level.

Three core findings, First, "...while on average individual job transitions move workers from lower-productivity to higher-productivity firms, the fraction of all job changes that flow up the productivity ladder is only marginally higher than those in the opposite direction. Almost half (49% in our preferred specification) of all job changes move workers towards firms that have lower productivity levels than their prior employer. Thus net labor reallocation towards higher-productivity firms hides a very large degree of job churning whose productivity implications mostly cancel...."

Second, "....we show that the fraction of up-the-productivity-ladder transitions varies significantly across job flow types and across firm productivity levels. They are more likely for direct job-to-job transitions than for those passing thorough non-employment spells. This is consistent with standard job-search theory, in the sense that movements resulting from on-the-job search should lead to better quality jobs more likely to be available at higher productivity firms. They are also more likely to occur between firms at the high end of the productivity distribution, with up-the-ladder transitions originating from low productivity firms happening less frequently than implied by a model of completely random movements...."

Third, "... the job transitions of younger workers, workers with high skills, and female workers are more likely to be to higher-productivity firms and correspond to larger productivity gains. In fact, young skilled workers provide the plurality of net labor reallocation to higher productivity firms. The net contributions of other large groupings are modest or even negative. Workers with the highest turnover rates contribute proportionally little to aggregate productivity growth...."

They note that their results imply that labor market churn for lower skilled workers likely isn't productivity enhancing, "... Overall, our findings sound a note of caution, or at least modesty in interpretation, about the connections between labor market fluidity and aggregate productivity growth. While fluidity enables reallocation of resources to more productive firms, many of the observed job transitions appear to happen for reasons that are orthogonal to industry or aggregate productivity growth. The flip side of this coin is that there may be a substantial amount of untapped potential for labor reallocations to enhance productivity growth...."

The data, "...First, we drop all firms that have only one employee for an entire year, as it seems likely that they represent some form of self-employment rather than an actual firm with workers in the spirit of our exercise. We also drop all workers employed in those firms. This is about 1.7% of the initial set of 48 million monthly worker-job matches. Second, we want to avoid spurious job transitions in which workers move to a “new” firm with a different tax ID that is actually directly linked to the previous firm. This includes cases in which the firm tax ID changes, M&As, or separations of a single firm into several business entities for tax reasons. We address this by excluding from the set of transitions all cases in which a significant share of a firm’s workforce jointly reallocates to another firm, whether existing or new. This removes just under 5000 firms.1 Third, we must assign any workers employed by more than one firm in a given month to a single job and employer. We classify the worker’s main job as the one with the longest current tenure and, if there is a tie, the one with the highest average monthly earnings. About 8% of our worker-job observations are dropped as secondary jobs. Finally, as we want to focus on full-time jobs but have no information on hours, we drop all employment relations in which the worker earns an implied full-time wage that is less than the minimum wage for 80% of her tenure. These jobs, which account for 20% of the initial set of observations, are probably part-time. We calculate average labor productivity at the firm level as a measure of annual sales (from form F22) per annual-equivalent worker. Firms’ annual-equivalent workers are simply the sum of workers in firms’ monthly DJ1887 reports, divided by 12. For example, a firm with one worker who is employed the entire year and two employees who each work in six months of the year (whether overlapping months or not) has two annual-equivalent employees.2 Thus, average labor productivity is defined at an annual level...."

The Evidence, "...Panel A presents the results across all 11 million transitions in our sample. Consistent with reallocation being a factor behind productivity growth, the average labor productivity gap between the destination and origin firm in a worker transition is about 8%. However, this aggregate result averages over enormous heterogeneity. This is clear in the large dispersion of the productivity differentials. The 25th percentile of the distribution involves a worker moving to a firm with a labor productivity level about half the size (e −0.625 = 0.535) of the firm she left. The 75th percentile transition is to a firm with a productivity level more than twice as high (e 0.761 = 2.14). The rough symmetry of this dispersion is reflected in the fact that the share of job transitions that move up the productivity ladder is only a shade above half, with 47.6% of worker transitions being to firms less productive than the origin firm. This is a striking result. While net reallocation flows are directed towards more productive firms, the process involves an enormous degree of churning in both directions. Theoretically, even in a frictionless world in which all job transitions were efficient, some transitions down the firm-productivity ladder might be motivated by better firm-worker matches or from other non-pecuniary motives that might drive worker reallocation across different firms…. Nonetheless, the fact that the frequency of upward flows is only marginally higher than that of a completely random reallocation benchmark, with a 50% unconditional change of moving up or down the firm-productivity ladder, is notable...."

"...Panel B addresses a potential concern regarding the exercise shown in Panel A. As discussed earlier, our data defines firm productivity by calendar years. Transitions that occur between calendar years compare productivity measures across different points in time. Secular aggregate productivity growth implies that the average true productivity gaps for across-year transitions would be overestimated. Additionally, the raw productivity gap does not account for systematic differences in average labor productivity across sectors, which are likely related to differences in capital intensity. We therefore adjust productivity across years by subtracting from the measured destination-origin productivity difference both the average annual productivity growth over the period as well as, for transitions across sectors, the difference in mean sector productivity levels. After this adjustment, the average productivity gaps do become smaller, with the average differential for all transitions falling to about 5%, and 51.3% of transitions being upwards. Most of the basic patterns in Panel A remain, however.The biggest difference is that the average and median productivity gaps of job-N-job transitions are now slightly negative.For the rest of the paper, our attention will focus on this measure of adjusted labor productivity differences across transitions, although all results are qualitatively robust to using the unadjusted gaps summarized in Panel A...."

"...To have a sense of the macroeconomic scale of the reallocation process described by the data, we sum sales per worker gaps across all job transitions that take place during a given year and then divide that sum by firms’ total sales that year. This calculation assumes each transition leads to an output change equal to what would happen if the transitioning worker experienced a labor productivity change equal to that between of the two firms she transitions between. The sum of these output changes relative to total output provides a metric of output growth coming from productivity gains through labor reallocation. Over our sample period, the implied average annual sales growth from reallocation computed in this way is 1.35. For comparison, average annual total sales growth in our data is 5.9%. This suggests that labor reallocation from lower to higher marginal product activities is quantitatively important as a potential source of growth..."

"...To explore these elements further and complement our firm-heterogeneity results in the previous section with worker-heterogeneity results, we place every worker into one of 25 age-by-skill worker groups. These use the five skills quintiles derived as above as well as five age groups (less than 25, 25-34, 35-44, 45-54, 55+). We compute the average productivity gap associated with worker transitions within each group. We take further advantage of our data to introduce one additional dimension by dividing the sample into men and women. This is a particularly interesting dimension for an economy like Chile, in which the gender wage gap is large and female labor participation, while growing, is still low. Figure 2 graphs the results. The differences across groups reflect some remarkable patterns. Consistent with the discussion above, productivity differentials are significantly larger for young, skilled workers. They are the largest for workers younger than 25 and in the top skill quintile. Productivity gaps decline monotonically with age, even conditioning on skill, becoming almost negligible for workers over 45. They also tend to increase with worker skill, conditioning on age. The one group where the average productivity difference between the destination and origin firm is clearly negative is for high-skilled workers over 55. This may reflect that a large share of those transitions are non-voluntary and associated with the destruction of such workers’ valuable job ladders. Comparing across genders, productivity gaps for a given age-skill group are consistently larger for female workers. The sole exception is the youngest, least-skilled group, where long time gaps between jobs are frequent. This indicates systematic gender-related difference in the process of labor reallocation. Overall, these results are further evidence of the heterogeneity in the interaction between job transitions and productivity. Productivity-enhancing reallocation is significantly stronger among young, high-skilled, and female workers. We now use these results to decompose productivity gains from reallocation into group-level contributions..."

Elias Albagli, Mario Canales, Chad Syverson, Matias Tapia and Juan Wlasiuk, "Productivity Growth and Workers’ Job Transitions: Evidence from Censal Microdata," National Bureau Of Economic Research, April 2021, https://www.nber.org/papers/w28657