What are the distributional and aggregate effects of the rising use of industrial robots across occupations? I construct a novel dataset that tracks the cost of robots from Japan by occupations. The dataset reveals a relative one-standard deviation drop of Japan's robot cost induces a 0.2-0.3% drop in the US occupational wages. I develop a general equilibrium model where robots are internationally traded durable goods that may substitute for labor differently across occupations. The elasticities of substitution between robots and labor within an occupation drive the occupation-specific real-wage effects of robotization. I estimate the model using the robot cost shock from my dataset and the optimal instrumental variable implied by the model. I find that the elasticities of substitution between robots and labor are heterogeneous across occupations, and higher than those between general capital goods and labor in production and material-moving occupations. The estimated model implies that the industrial robots explain a 0.9 percentage point increase in the 90-50th percentile ratio of US occupational wages, and a 0.2 percentage point increase of the US real income from 1990 to 2007.
Media Coverage: VoxEU
We study the impacts of industrial robots on employment in Japan, the country with the longest tradition of robot adoption. We employ a novel data set of robot shipments by destination industry and robot application (specified task) in quantity and unit values. These features allow us to use an identification strategy leveraging the heterogeneous application of robots across industries and heterogeneous price changes across applications. For example, the price drop of welding robots relative to assembling robots induced faster adoption of robots in the automobile industry, which intensively uses welding processes, than in the electric machine industry, which intensively uses assembling process. Our industrial-level and commuting zone-level analyses both indicate that the decline of robot prices increased the number of robots as well as employment, suggesting that robots and labor are grossly complementary in the production process. We compare our estimates with the ones reported by existing studies and propose a mechanism that explains apparent differences between the results.
We investigate the impact of multinational enterprises (MNEs) on the labor share in the source country. We propose an equilibrium model that features a production function with factor inputs in foreign countries. Each firm receives a shock that shifts the productivity of foreign factor inputs. We conduct comparative statics regarding the foreign factor productivity shock and show that the difference in factor demand elasticities with respect to foreign factor prices affects aggregate labor share. To identify these elasticities, we develop a method-of-moments estimator that leverages a foreign factor productivity shock. We then apply the estimator to a unique natural experiment: the 2011 Thailand Floods. The floods had a strong impact on manufacturing clusters in areas north of Bangkok city and affected Japanese MNEs by forcing them to halt operations of plants located in the cluster. We employ a uniquely combined Japanese firm- and plant-level dataset that tracks wages, employment, fixed assets in Japan, and employment in foreign subsidiary plants. The estimated factor demand elasticities indicate that foreign factor augmentation increased capital demand in Japan more than labor demand, suggesting that the foreign factor augmentation contributes to reducing the labor share in Japan.
Media Coverage: VoxEU
Choosing a proper geographic unit is crucial for achieving an accurate analysis of local labor markets. While a small administrative unit such as the municipality is not always ideal because workers commute across units to form a single labor market, a large administrative unit such as the prefecture is often too coarse because prefectures typically include several labor markets. To define appropriate local labor markets for Japan, we first constructed commuting zones (CZs) using the commuting patterns observed in the Population Census from 1980 to 2015 and the hierarchical agglomerative clustering method adopted by Tolbert and Sizer (1996) to delineate CZs in the US. From 1,736 municipalities in 2015, for example, we constructed 265 CZs that are mutually exclusive and exhaustive. We then compared the properties of CZs with those of other potential administrative units including the municipality, prefecture and Urban Employment Area (UEA) proposed by Kanemoto (2002), finding that our proposed CZs capture the actual commuting patterns and heterogeneity of local labor markets reasonably well.