A number of developed countries now have linked employer-employee records, although to date I haven’t seen as many papers doing cool things with such data as I would expect. A new paper  in the AEJ-Applied (ungated here ) by Andrey Stoyanov and Nikolay Zubanov uses Danish data to show what is possible, and help provide some of the most convincing evidence yet that workers carry firm knowledge with them when they move.
First, the data: On the firm side, an annual firm survey of all firms with 50+ workers plus less frequent surveys of smaller firms gives them 173,000 manufacturing firm-year observations over 1995-2007 , with data on the key variables needed for calculating productivity – sales, employment, value-added, materials and energy inputs, profits, capital stock. This is linked to records on all individuals aged 18-65, which has annual information on age, gender, experience in thousands of hours, education, occupation, and firm – between 1995 and 2007 they have 5.8 million worker-year observations, of which 668,034 are job changers.
What happens when workers move?
They then use this data to trace what happens to the productivity of firms when they hire workers from more or less productive firms. They measure the productivity gap as the difference in the log value-added per worker in the receiving firm from that in the firms from which it receives workers one year before hiring takes place. They then look at the impact of this gap on the receiving firms productivity in the year after hiring. They find that hiring workers from a firm that is more productive than yours increases your productivity in subsequent years, whereas hiring workers from a firm that is less productive than yours has no effect on your own productivity.
Of course the immediate concern that comes to mind is endogenous movement – perhaps firms only hire workers from better firms when they experience a positive productivity shock, and workers are only likely to move across from better firms when they sense that the firm is improving. The authors try and alleviate such fears in a couple of ways:
- First, they control for productivity shocks happening before the year after hiring by controlling for up to 5 lags of productivity. This deals with the concerns that firms which are consistently improving are the ones hiring workers from better firms.
- Second, to deal with the concern that contemporaneous productivity shocks are driving both productivity gains and hiring new workers, they use the IO approach of Olley and Pakes et al and control for polynomials in capital and investments – based on the idea that firms adjust their capital as they anticipate these shocks.
The productivity gains are equivalent to 0.35 percent per year for an average firm, and last four years on average. They then look at what types of worker movements are associated with higher productivity gains and find intuitively sensible results: productivity gains are twice as high when you hire a worker from a firm in the same sector as yours as when you hire someone from another sector; and gains are higher when you hire workers who are more educated and more experienced. Nevertheless, significant gains are still present when medium-skilled workers are hired, so this is not just all about hiring the most skilled workers.
The big issue is then what the underlying channel for this productivity gain is if we believe it is a causal effect. The authors try and remove the direct human capital effect by controlling for the observed (agem gender, experience, education, professional status, salary) and unobserved (person-specific fixed effects from a wage equation) components of human capital. They find these productivity gains even after controlling for the human capital of the workers who move. But what such data can’t tell us is whether the productivity gain reflects these workers becoming more productive because they are now in a better job match despite being in a lower productivity firm (e.g. perhaps someone now listens to their ideas, perhaps they move into a position more closely matched to their abilities, etc.) or whether the gains come from them making other workers in the firm also more productive. Another nice thing about the paper is that they don’t overreach in discussing their results or concluding – words like “association”, “caution the reader” and “limitation to interpreting our results” are all suitably present.
So what are some implications for developing countries? The first is the importance of worker turnover in reducing productivity dispersions (with the implication that barriers to smooth labor market reallocation such as tenure systems which don’t reward productivity, excessive labor regulations, etc. do have aggregate impacts on productivity). This figure at the start of the paper is interesting – we see that productivity dispersion is higher in industries in which labor turnover is lower – nothing causal here of course, but an interesting association:
Second, the paper shows the value of having good data. I can imagine the results might look quite different in countries with much more dysfunctional labor markets – but don’t know of any developing countries with data like this. Of course one huge issue in developing countries is the large informal sector for which such data would not be available – but it would still be interesting to see whether such patterns occur in the formal sector. Moreover, linking such administrative data to experiments which work with firms and workers, and to more detailed specialized surveys would yield even more insights into how exactly knowledge can and cannot spillover from one firm to another.