Follow the Money—and the Work
To understand what the 2021–22 yen depreciation did to Japanese multinational firms, we need to compare the foreign affiliates that earned income with the Japanese parents that organized and supported them. The useful question is not simply whether exposed firms became richer in yen, but whether their overseas operations expanded and what changed inside Japan. The linked data make that comparison possible, while also showing where the evidence stops.
How exposure travels through an existing network
Aggregate balance-of-payments data can show that Japan received more direct-investment income, but they cannot show which firms received it, whether their affiliates hired more people, or whether the Japanese parent changed its own workforce. The study therefore links two annual surveys conducted by Japan’s Ministry of Economy, Trade and Industry. One covers foreign affiliates of Japanese firms; the other covers the Japanese parents. A common parent identifier makes it possible to follow an affiliate’s sales, employment, profits, and payments alongside the parent’s wages, employment composition, and fixed assets.
The linked sample is substantial but not universal. The 2020 summary-statistics sample contains 2,210 parent firms, 10,776 affiliates, and 102 host countries; 62 percent of the parents are in manufacturing. The affiliates’ revenue is mostly earned outside Japan: about 67.4 percent comes from the local host market, 17 percent from Japan, and 15.6 percent from third countries. This matters because the firms are not primarily shipping goods back to Japan for resale. They are largely operating as foreign-market businesses in their host economies.
The surveys also reveal two connections that aggregate data conceal. Affiliate-to-parent payments average about 74 percent of affiliate current income. That ratio is highly skewed and includes dividends, royalties, and related transfers, so it is not an exact payout rate or a direct measure of balance-of-payments accrual. It is better understood as evidence that financial links between the affiliate and its Japanese parent are strong. On the parent side, non-production support employment accounts for about 22.5 percent of employment. The category includes commercial branches, sales establishments, offices, and related business units outside the head office. It is large enough to be a plausible margin of adjustment, not a statistical footnote.
The next challenge is to measure which parents were more exposed to the yen movement through their foreign networks. A simple comparison of firms before and after 2021 would be difficult to interpret. The pandemic, supply-chain bottlenecks, changing demand, and local restrictions affected affiliates in different ways. Instead, the paper uses differences that were already present before the depreciation.
For each affiliate, the researchers first calculate how its pre-period revenue and purchases were distributed across the host market, Japan, and third countries. They then combine those shares with the bilateral change in the yen between October 2020 and October 2022. At the parent level, the resulting affiliate exposures are weighted by each affiliate’s 2020 share in the parent’s foreign affiliate network. A parent with important affiliates whose sales were concentrated in currencies that appreciated strongly against the yen receives a larger revenue-side exposure. A separate purchase-side exposure captures the possibility that foreign-currency input costs moved in the opposite direction.
This design is a way of asking a focused counterfactual: among Japanese parents that already had foreign affiliates, did the firms whose inherited networks made the yen depreciation more valuable respond differently from firms with less exposed networks? It is not the same as asking whether Japan as a whole changed after the exchange rate moved. The currency movement is common to firms linked to a given currency; the firm-level variation comes from the network and transaction geography that existed before the episode.
The Taiwan–Korea comparison makes the idea concrete. Japanese affiliates in both countries had similar revenue geography: about 70 percent of revenue was earned in the local market and less than 10 percent came from Japan. Until 2020, the yen moved similarly against the Taiwan dollar and the Korean won. By the end of 2022, however, the cumulative yen depreciation was approximately 22 log points against the Taiwan dollar and 11 log points against the won. A Japanese parent weighted toward Taiwan therefore had greater revenue-side exposure than a parent weighted toward Korea, even though the two networks could look similar if one only counted the presence of an East Asian affiliate.
The exposure measures are not improvised short-run responses. In the 2015–20 pre-period, affiliate fixed effects explain roughly 97 percent of the variation in parent-network weights and between 85 and 92 percent of the variation in transaction shares; year effects explain essentially none. That persistence supports the interpretation that the design is using durable features of the firms’ networks rather than choices made after the yen began to fall.
There is an important measurement qualification. The affiliate survey does not report the currency in which every transaction is invoiced. The baseline assigns host-market transactions to the host currency, transactions with Japan to the yen, and third-country transactions mainly to the U.S. dollar or euro. Those conventions reflect broad invoicing patterns, but they are not direct observations of every contract. The paper reports similar results under alternative mappings, yet the exposure variable remains an informed approximation.
What moved abroad, and what changed at home
The event-study results provide the first check on timing. Before 2020, affiliates with greater revenue-side exposure did not show systematic differential changes in value added or payments to parents. After the depreciation began, both outcomes rose for the more exposed affiliates. This pattern does not by itself prove that exchange rates caused every change, but it is more informative than a simple correlation in levels: the divergence appears when the relevant currency episode begins rather than being an established pre-existing trend.
The long-difference estimates show that the response was broad on the affiliate side. Greater revenue-side exposure is associated with higher value added, current profit, retained earnings, and total payments to the Japanese parent. Moving an affiliate from the 10th to the 90th percentile of the revenue-exposure distribution corresponds to roughly a 19 percent increase in reported current profit and about a 15 percent increase in total payments. Profit and payment responses have similar magnitudes, which is consistent with higher affiliate income being transmitted financially to the parent. Royalties, a small component of total payments, do not show the same statistically precise response.
These are yen-denominated financial outcomes, so they can contain a translation effect. If a foreign affiliate earns the same foreign-currency surplus as before, that surplus is worth more in yen after the yen depreciates. The survey does not reveal every transaction currency or the exact translation conventions used in the accounts, so the study does not claim that the profit estimate alone measures new physical production.
Employment provides a different kind of evidence. Revenue-side exposure is associated with higher total affiliate employment and more workers dispatched from the Japanese parent. The p90–p10 effects are about 4 percent for affiliate employment and 4.8 percent for dispatched employment. Headcount is not mechanically created by converting foreign earnings into yen. It indicates that at least part of the affiliate response involved additional labor use. The dispatched-worker result is especially revealing because it records a narrow form of labor supplied directly by the Japanese parent.
The type of affiliate matters. The dispatched-worker response is about three times larger for sales- and services-oriented affiliates than for manufacturing affiliates. Total affiliate employment does not show the same differential, whether or not dispatched workers are subtracted. That pattern points toward parent-supplied commercial and coordination functions rather than a general claim that every overseas factory expanded in the same way.
The purchase-side results provide an internal comparison. Greater purchase-side exposure generally reduces affiliate value added, employment, and dispatched workers. Revenue exposure and cost exposure therefore do not look like one undifferentiated response to having foreign-currency business. The remaining expenditure outcomes—such as capital investment, research and development, and rental costs—are less precisely estimated. The evidence is strongest for income, payments, and labor use.
The parent-side results are the part most likely to be missed by a headline about foreign profits. At Japanese parents with greater revenue-side affiliate exposure, average wages rise, but the employment response is mainly a change in composition. Non-production support employment increases, other parent employment declines, and total employment remains approximately flat. Head-office employment does not respond. Moving from the 10th to the 90th percentile of revenue exposure raises non-production support employment by about 9 percent and reduces other employment by about 7 percent. Domestic fixed assets also fall by about 5 percent.
That pattern is not evidence that Japanese parents stopped doing anything at home. Support employment includes commercial and business functions that may serve domestic operations, overseas affiliates, or both. The survey does not identify an individual worker’s task or destination. Nor can the average-wage measure tell us whether wages rose within each employment category or whether the workforce’s composition changed. The cautious interpretation is that exposed parents shifted toward support and coordination functions associated with their broader multinational activity, rather than adding a large number of domestic production jobs.
Several details make that interpretation more credible. The support-employment response is concentrated among parents with larger foreign affiliate workforces; parents below the median show no comparable response. The parent-dispatched-worker result independently records some labor moving toward affiliates. Together, these patterns fit a reorganization around foreign operations better than a general domestic hiring boom.
The study also tests two alternatives. If foreign-affiliate income were mainly a financial windfall that relaxed domestic constraints, more leveraged parents should have responded more strongly. The interaction with initial leverage is insignificant across the parent outcomes. If affiliates expanded because a weaker yen made Japanese intermediate inputs cheaper, the response should be stronger among affiliates that sourced more from Japan. The revenue-exposure interaction with Japan-input intensity is not positive across the tested outcomes. These tests weaken particular financial and vertical-input explanations, but they do not prove that every support worker served an overseas affiliate.
The evidence has three further limits. First, the log-change regressions require a positive outcome in both 2020 and 2022, so they describe the intensive margin of continuing observations rather than entry, exit, or newly profitable affiliates. Second, the identification assumes that pre-existing network geography did not proxy for another shock that differentially affected outcomes during the same period. Host-country-by-sector fixed effects and the absence of differential pre-trends help, but no design can make that assumption disappear. Third, the sample is built from reporting and linkage requirements and is concentrated in relatively large firms, many of them manufacturers. The estimates should not be read as a direct average effect for every Japanese business.
What the data do establish is more specific and more useful than “the yen helped” or “the yen failed.” A Japanese parent whose existing affiliate network was more exposed to the depreciation saw stronger foreign-affiliate income and labor use. Some of the resulting value reached the parent through payments, higher wages, and more support employment. Yet total parent employment did not expand broadly, and domestic fixed assets declined. The foreign side of the multinational grew in ways that the home side absorbed partly through reorganization.
The remaining question is not whether these patterns are real, but what constraint could generate them. Why would a firm hire more support while leaving total employment nearly unchanged? Why would the support be directed toward an affiliate rather than a domestic factory? Answering that requires making the parent’s internal allocation problem explicit.