Periodic Reporting for period 3 - AUTOMATION (AUTOMATION AND INCOME DISTRIBUTION: A QUANTITATIVE ASSESSMENT)
Okres sprawozdawczy: 2021-11-01 do 2023-04-30
The project is articulated around three parts. The first part uses patent data to study empirically the causal effect of labor costs on automation innovations. There, we develop a new classification of patents in machinery as automation patents or not. Then, we use international data on labor costs to build firm-level variation in the labor costs of the customers of innovating firms by exploiting variations in firms’ exposure to international markets. We analyze the effect of an exogenous increase in low-skill and high-skill labor costs of firms’ automation and non-automation machinery patents. We also look at the effect of the Hartz reforms, which effectively reduced low-skill labor costs in Germany, on non-German firms, which are more or less exposed to the German market.
The second part of the project aims at studying empirically the causal effect of machinery capital on labor outcomes at the industry-level and in local labor markets in the United States (commuting zone, CZ). Our goal there is to analyze whether capital is complement or substitute to labor. This is a long lasting question in economics that has received renewed interest recently with the development of new automation technologies. We intend to show that capital has a very heterogeneous effect on employment: most capital purchases are complement to labor but automation capital is a substitute.
The third part of the project aims at calibrating an endogenous growth model with firm dynamics and automation using Danish firm-level data. We first analyze the evolution of the labor share at the firm level in the Danish economy using micro data from 1995 onwards. We show that the decline in the aggregate labor share results mostly from a composition effect as a larger share of value-added is produced by low labor share firms today than before while the median firm has seen an increase in its labor share (similar to what has occurred in the US manufacturing sector). We then look at different potential causes for this phenomenon (automation, offshoring, exports, etc.) and plan on building our quantitative model based on this analysis.
We have made significant progress on the second part of the project. We have improved the match between patents and sectors of use already explored in the first project. We use firm-level data on patenting firms to map technology codes with sector of invention. Then we use a capital flow table, which details the sectors from which each sector builds capital goods, in order to map patents with sector of use. This allows us to measure for each sector its automation and its non-automation capital. We find that the two measures are highly correlated but that while the purchase of non-automation capital is associated with an increase in employment, the purchase of automation capital is associated with a decline in employment. We are now in the process of building an instrument based on the network of knowledge spillovers received by patents in a specific technology code. This instrument will allow us to look at an exogenous source of variation for changes in the automation or non-automation content of the capital bought by a sector, and therefore measure the causal effect of a change in the nature of the supplied capital on employment in a sector.
We are also still working on the third part of the project. We have derived the main stylized facts regarding the evolution of the labor share distribution among Danish firms. As mentioned before, we find that the decline in the aggregate labor share is mostly a within industry phenomenon driven by a composition effect where a larger share of the industry value added is produced by low labor share firms, while the median firm experiences an increase in its labor share (in line with previous research). In contrast with most previous research, we also focus on the evolution over time of firms’ labor share: we find that initially low labor share firms have not grown faster. Instead some firms have managed to scale up without increasing employment proportionately and thereby reducing their labor share. We are now in the process of analyzing which factors seem to account for these changes in the labor share distribution. Preliminary findings highlight the role of export markets: indeed, the composition effect is stronger in more export-oriented sectors, and within an exporting sector, low-labor share firms are larger mainly because they sell more internationally.
I was also invited to write an Annual Review of Economics article on “Directed Technical Change in Labor and Environmental Economics’’, which I wrote in 2020 (with Morten Olsen). The part on labor economics is directly related to the topic of the ERC grant since it focuses on models of automation innovations and empirical work on endogenous adoption of automation technologies and endogenous innovation in automation.
The second part of the project aims at showing the causal effect of automation on labor market outcomes but also that automation capital is significantly different in that respect from most capital. The current literature is divided on the aggregate effect of automation on employment and we believe that this reflects measures of automation capital that are either too broad (all machinery) or too restrictive (robots). Instead, we believe that the granularity and flexibility of our measure will allow us to show concomitantly the effects of automation and non-automation capital. Moreover, we also intend to disentangle the effect of automation innovation on the inventing industry / region and the effect from adoption by the using industry / region. This dichotomy has not been studied before.
Once, the first two parts are combined, we hope to be able to trace the effect from an exogenous increase in labor costs to an increase in automation back to its effect on labor outcomes.
The third part of the project belongs to a rapidly growing literature that documents dramatic changes in the firm-level distribution of labor shares (Kehrig and Vincent, 2020, Autor, Dorn, Katz, Patterson and Van Reenen, 2020) and tries to explain this phenomenon through, for instance, an increase in automation, as in Hubmer and Restrepo (2021) or the role of IT as in De Ridder (2020) and Aghion, Bergeaud, Boppart, Klenow and Li (2020). Our approach presents several advantages: first thanks to the richness of the Danish firm data, we can properly compute the labor share outside of manufacturing (other papers typically rely on proxies); second we can empirically investigate different channels before building our own model. Our model will then be able to quantify how much each channel has contributed to the decline of the labour share.