Will robots be good or bad for our jobs? Here are lessons from Japan

How will automation affect the future of employment? Policymakers, journalists, and academics alike are all attracted by this question due to the fear of the replacement of human jobs by machines. Taking up the industrial robots as an example, Acemoglu and Restrepo (2020) indeed draw a dismal picture that the penetration of industrial robots in the US depressed both employment and wages of workers of all educational backgrounds. To what degree can this experience be extrapolated to other regions and other times? In Adachi et al. (2020b), we address this question by looking at Japan’s long-term experience spanning three decades from 1978 to 2017.

Figure 1 Robot stock by country:

Note: Data are from the International Federation of Robots (IFR) and the Japan Robot Association (JARA).

Japan’s experience of industrial robot adoption is peculiar in terms of the very early timing of robot adoption and the fact that the robots are produced domestically. Japan’s industrial robot producers in the late 1970s and the early 1980s expanded production through cut-throat competition and continuous innovation to meet the need of industries to remove the human labour force from the harsher portions of the work environment. As a result, the timing of robot penetration in Japan was much earlier than in other developed countries, as Figure 1 shows.

This peculiar feature of robot penetration in Japan makes it an interesting case to study the effect of new technology penetration on employment among the early adopters of technologies that emerge domestically. This unique feature of Japan, however, poses a substantial challenge for researchers who attempt to estimate the causal effect of robot penetration on employment, because technology adoption is endogenously determined. Increase in technology adoption and increase in labour demand might occur simultaneously when product demand expands in an industry.

Figure 2 Average unit value of robots by application type

Note: Unit value is calculated by dividing the total shipment amount by the number of robots shipped.

In our study, we overcame this challenge by exploiting the technological progress in robot production that is captured by the fall in the price of robots. As shown in Figure 2, the decrease in the price of robots differs substantially across robot application types. The price of welding robots fell substantially while the price of assembling robots was relatively constant. This heterogeneous change of the prices of robots across application types generates different robot price trends for robot adopting industries, depending on the different composites of application types per industry. The automobile industry, which uses welding robots more intensively than other industries, saw a significant fall in prices, whereas the electronics industry, that intensively use assembling robots, faced relatively constant robot prices. We exploit these inter-industry differential trends of effective prices to identify the causal impact of adoption on employment.

We find that robots are complementary to employment. We examine this finding both at the industry level and the region level. At the industry level, we show that a 1% decrease in robot price increased adoption by 1.54%. Perhaps more surprisingly, we also find that a 1% decrease in robot price increased employment by 0.44%. This finding implies that robots and labour are gross complements. Therefore, taken together, our two-stage least-squares estimates suggest that a 1% increase in robot adoption caused by price reduction increased employment by 0.28%.

At the region level, we apply commuting-zone (CZ) level analysis to better compare our results with existing estimates in the literature, using the Japanese CZ defined in Adachi et al. (2020a). By constructing a shift-share analysis of robot exposure, we conduct a two-stage least-squares estimation resembling the one by Acemoglu and Restrepo (2020), but with our cost-based instrumental variable. Our results indicate that an increase of one robot unit per 1,000 workers increases employment by 2.2%, corroborating the finding that the robots and labour are gross complements. This contrasts with the finding by Acemoglu and Restrepo (2020), whose corresponding estimate was -1.6%. The difference of the results is not surprising, given the difference of the country and the time period covered. Particularly, considering the export-oriented nature of Japan’s automobile and electric machine sectors, robot adoption and its cost-reducing effect have contributed to the expansion of exports and increased labour demand. This scale effect may well have offset the substitution effect of robots for labour.

The CZ-level analysis enables us to conduct further analyses regarding spillovers from the manufacturing sector to the non-manufacturing sector. First, we find that employment in non-manufacturing sectors neither increased nor decreased upon robot adoption. This fact shows that within-region-across-industry reallocation from services to manufacturing did not occur, suggesting that an across-region reallocation of workers occurred. In other words, in the context of the declining population in Japan, robots might work like magnets that keep workers from leaving for other regions. Second, we find that although the total employment increased when robots were adopted, the hours worked per worker decreased. This finding suggests that robots may have worked as work-sharing and time-saving technological changes. In turn, this implies that the hourly-wage effect might be even more positive, and we confirmed this fact in our data.

In Adachi et al. (2020b) we demonstrate that the adoption of automation technology could in fact expand employment and increase wages. While the detailed mechanism behind this relationship is not yet completely clear, the expansion of production scales through the reduction in production costs seems to be the key. This study at least shows that the adoption of automation technology is not always bad news for human labour.