Across-Country Wage Compression in Multinationals

Layer 1 — Summary

Many multinationals do not fully adjust wages to the local context of their foreign establishments; instead, they partially link the wages of foreign workers in a given position to the wages paid in the same position at headquarters — a practice the authors call “wage anchoring.” Using yearly establishment-level compensation data on roughly 1,200 multinationals operating across 174 cities worldwide (2000–2015) and matched employer-employee administrative data (RAIS) from Brazil, Hjort, Li, and Sarsons document that a 10 percent higher headquarters wage is associated with 1.63–2.8 percent higher wages for workers in the same occupation at foreign establishments, with the within-firm across-country correlation substantially exceeding the correlation between a given establishment’s wages and the local average paid by other multinationals for the same position. To establish a causal link between externally imposed headquarters wage changes and subsequent foreign establishment wage responses, the paper exploits two identification strategies: minimum wage shocks in the headquarters country or U.S. state and exchange rate fluctuations, both of which generate plausibly exogenous variation in headquarters wages that is then partially transmitted to foreign workers in the same position. Wage change transmission appears to be direct and to operate through firm-wide wage-setting procedures rather than through associated changes in technology or employment at foreign establishments, a conclusion the Brazil RAIS data support because total employment at multinationals’ Brazilian establishments shows little change following positive external shocks to headquarters wages. Wage anchoring is strongest for low-skill occupations (cleaners, drivers, security guards), where a 10 percent higher headquarters wage is associated with a 2.8 percent higher foreign establishment wage, versus roughly 1.2 percent for middle- and high-skill occupations; the resulting spatial compression of wages is in line with how many multinationals themselves report setting pay across locations.

Layer 2 — Q&A

Q1: What is the central phenomenon documented in this paper, and what are the two broad empirical components of the analysis?

The central phenomenon is “wage anchoring”: multinationals link wages at their foreign establishments to the wage level at headquarters for the same narrowly-defined occupation, so that the within-firm across-country wage distribution is more compressed than what local labor-market conditions alone would imply. The first empirical component is descriptive — documenting the high cross-sectional correlation between headquarters and foreign establishment wages within a firm×occupation cell, controlling for city×year effects and local wage benchmarks. The second component is causal — using minimum wage shocks in the headquarters country or U.S. state and exchange rate shocks to generate externally imposed changes in headquarters wages, and tracing whether and how quickly those changes are partially transmitted to foreign establishments.

Q2: What is the primary dataset, what does it cover, and what are its key limitations?

The primary dataset was compiled by an unidentified consulting company that gathers compensation information from client employers and harmonizes positions globally into 309 occupations across 16 skill levels and 26 occupational categories. It covers roughly 1,200 multinationals (private-sector firms and multinational public-sector employers such as NGOs and multilateral organizations), operating in more than 170 cities, with yearly observations spanning 2000–2015. The data report average nominal gross total monthly wages for domestic (non-expat) workers in each establishment-occupation-year cell. Key limitations: the panel is unbalanced because multinationals choose which establishments report each year and often rotate establishments in and out; matching between the headquarters and any given foreign establishment requires observing the same occupation in the same year at both, which reduces the headquarters-matched sample to 80 employers and 611 foreign establishments (Sample 3, the most comparable subsample). The publicly listed U.S. firms in the data account for about one-third of total revenue of all publicly listed U.S. firms, so the sample is skewed toward unusually large employers.

Q3: How do the authors define and measure “wage anchoring” in the descriptive section?

The authors regress log average wages of workers in occupation j at a firm f’s foreign establishment in city c in year t (wjfct) on log average wages for the same occupation at the firm’s headquarters (HQwjft), controlling for firm×occupation fixed effects, city×year fixed effects, and a local market wage benchmark measured either as the average paid by other multinationals in the same city-occupation-year cell or as a city×occupation×year fixed effect. The estimated coefficient on the headquarters wage — around 0.163 using the benchmark-wage control and about 0.09 using the more restrictive city×occupation×year fixed effect — measures how much of a headquarters wage difference is “passed through” to foreign establishment wages within the same firm and occupation. They further document that the within-firm wage slope (the difference between wages in consecutive skill levels within an occupational category) at foreign establishments is similarly anchored to the corresponding slope at headquarters, with a 10 percent greater consecutive-skill wage gap at headquarters associated with about a 1.4 percent greater gap at the foreign establishment.

Q4: What exactly do the minimum wage and exchange rate identification strategies exploit, and what do they identify?

The minimum wage strategy compares multinationals whose headquarters are located in a country or U.S. state that experiences a minimum wage increase (“treated”) against multinationals whose headquarters are not exposed (“control”), conditioning on establishments being in the same foreign city. Within the treated group, it also exploits cross-occupation variation: within a given foreign establishment, workers in positions whose headquarters counterparts are more exposed to the minimum wage increase (because their wages are closer to the new minimum) experience larger foreign wage gains. The exchange rate strategy exploits appreciation of a non-U.S. headquarter country’s currency against the dollar: when the USD-measured headquarters wage of such a multinational increases following an appreciation, this tests whether foreign establishment wages in USD also rise. Because exchange rates increase and decrease, are less stable than minimum wages, and have different underlying drivers, the exchange rate design provides an independent corroboration of the minimum wage findings. Both strategies identify the effect of externally imposed headquarters wage changes on wages at the same firm’s foreign establishments in the same narrowly defined occupation.

Q5: What evidence is marshaled against indirect pathways (technology changes, employment changes, offshoring) as the driver of foreign wage transmission?

The paper presents three types of evidence against indirect pathways. First, including headquarters country×year fixed effects in the descriptive wage regressions — which absorbs any technology shocks originating in the headquarters country that affect all occupations uniformly — leaves the estimated wage anchoring coefficient essentially unchanged. Second, event study and panel regressions using the Brazil RAIS data show little change in total employment at multinationals’ Brazilian establishments following positive external shocks to headquarters wages, which is hard to reconcile with employment-driven or offshoring-driven wage adjustment. Third, a causal forest analysis of the conditional average treatment effect of minimum wage shocks on foreign wages — estimated allowing responses to vary with a wide range of job, employer, sector, and location characteristics — finds that occupation characteristics and sector have little explanatory power for which establishments transmit more, while differences in transmission are more closely related to characteristics of the headquarter-establishment country pair (proximity, similarity, shared language), which are more naturally associated with administrative coordination than with technology or production-style linkages.

Q6: How does occupation skill level moderate wage anchoring, and what does this heterogeneity imply?

Wage anchoring is strongest for low-skill occupations. In the descriptive correlations, a 10 percent higher headquarters wage is associated with 2.8 percent higher foreign wages in low-skill jobs (cleaners, drivers, data entry clerks, security guards) but only about 1.2 percent higher foreign wages in both middle-skill and high-skill jobs. The occupation heterogeneity is visible graphically (Figure 1 Panel C) and holds in regressions interacting the headquarters wage with skill-level indicators. A natural interpretation, consistent with the firm-wide wage-setting procedure explanation, is that firms are most likely to apply standardized pay rules to lower-level positions where local market customization may be seen as less important; higher-skill workers may be more likely to have individually negotiated contracts responsive to local conditions. The heterogeneity also implies that the spatial compression effect — wages in foreign establishments being pulled toward headquarters levels — is particularly pronounced at the lower end of the within-firm wage distribution, affecting positions like cleaners and guards in ways that can result in wages that are, relative to GDP per capita, an order of magnitude higher than what headquarters workers in the same position receive.

Q7: What is the “spatial compression” implication and how does it relate to within-firm wage inequality?

Wage anchoring implies that workers in the same occupation at foreign establishments located in lower-income countries receive wages that are compressed toward headquarters levels rather than fully adjusted to local wages. The paper shows that nominal wages at foreign establishments average about 89 percent of headquarters wages in the same occupation and year — and about 78 percent for establishments in countries poorer than the headquarter country — a ratio that is roughly stable across the within-firm headquarters wage distribution. This partial equalization is what the authors call “across-country wage compression”: it reduces the within-multinational cross-country wage dispersion relative to what would arise from purely market-based, locally responsive wage-setting. The spatial compression is consistent with how many firms self-report setting wages: a survey of primarily North American employers (Culpepper & Associates, 2011) found 29 percent report paying the same nominal wages across locations, and several large employers (Amazon, IKEA, Walmart) have self-imposed country-wide wage floors.

Q8: What role do headquarter-establishment country-pair characteristics play in predicting which establishments exhibit stronger wage transmission?

Using a causal forest algorithm to estimate the conditional average treatment effect of a minimum wage shock at headquarters and then constructing above- versus below-median predicted treatment groups, the paper finds that differences in transmission are “generally not large” but that higher transmission is somewhat associated with characteristics of the headquarter-establishment country pair: pairs that are more closely connected and share more similarities (e.g., common language, closer geographic distance) transmit more. Some foreign-establishment-country characteristics such as inequality and urbanization also appear related. In contrast, occupation characteristics (such as offshorability), the sector the multinational operates in, and characteristics of the headquarter country alone have little explanatory power. The paper notes these findings do not conclusively rule out alternative explanations but are more consistent with administrative coordination channels than with technology- or employment-based ones.

Q9: What role do potential fairness preferences and firm-wide wage norms play in the paper’s interpretation?

The authors suggest several possible mechanisms through which firm-wide wage-setting procedures could operate. Firms may adopt uniform wage-setting to reduce the menu and information costs of localized wage-setting (Lemieux et al., 2012); to increase foreign worker morale, particularly if workers are averse to pay inequality relative to headquarters peers (Card et al., 2012; Dube et al., 2019); or to respond to fairness preferences from headquarters workers or consumers (Harrison & Scorse, 2010). Survey evidence from Alfaro-Urena et al. (2019) explicitly records that multinationals pay high wages abroad in part to “ensure cross-country pay fairness within the MNC.” Alternatively, the authors note that firm-wide wage-setting may represent a form of firm inertia or mistakes — an inability or unwillingness to fully adapt pricing and compensation to local contexts — consistent with DellaVigna & Gentzkow (2019). The paper presents this as an open question for future research rather than definitively adjudicating among the explanations.

Q10: How does the Brazil RAIS data corroborate and extend the global multinationals findings?

The RAIS matched employer-employee administrative data cover all employees at each Brazilian establishment of the 44 multinationals in the global dataset that operate in Brazil, with individual-level information on wages, education, race, gender, age, and tenure. Because RAIS is an administrative census of formal-sector employment rather than a consulting firm’s client dataset, it provides independent corroboration of the main findings. The paper confirms using RAIS that wages of individual workers at multinationals’ Brazilian establishments rise abruptly when their foreign headquarters experience positive external shocks. The RAIS data then enable the additional step of examining employment responses, where event study and panel regressions find little change in total employment at multinationals’ Brazilian establishments following such shocks — evidence against employment- or technology-driven indirect pathways as the primary explanation for wage transmission.

Key Concepts

Wage anchoring: The practice by which a multinational ties wages at its foreign establishments, for workers in a given occupation, to the wage level at its headquarters for the same occupation. In this paper’s usage, anchoring does not mean wages are set identically across locations but that they are partially linked — externally imposed changes in headquarters wages are partially transmitted to foreign establishment wages — rather than being independently set based on local labor-market conditions.

Across-country wage compression: The reduction in the cross-country dispersion of wages within a multinational that results from wage anchoring. Because foreign establishment wages are partially pulled toward headquarters levels rather than fully adjusting to local wages, the multinational’s within-firm wage distribution is more compressed across countries than it would be under purely localized wage-setting. In the paper’s data, this compression is particularly pronounced for low-skill occupations in lower-income host countries.

Firm-wide wage-setting procedures: Administrative practices, such as applying a single pay scale or a fixed wage ratio across all of a firm’s establishments regardless of location, that mechanically link foreign establishment wages to headquarters wages. The paper argues these procedures — rather than correlated technology shocks or employment adjustments — are the proximate driver of wage anchoring, on the basis of the employment non-response in Brazil, the persistence of anchoring after controlling for headquarters-country technology shocks, and the pattern of heterogeneity across country pairs.

Partial transmission: A load-bearing qualifier in this paper describing the magnitude of wage anchoring: headquarters wage changes arising from external shocks are not fully extended to foreign workers, but a fraction of the change is passed through. The estimated pass-through in descriptive regressions ranges from about 0.09 to 0.31 depending on specification and sample, and is highest (around 0.28) for low-skill occupations. The partial nature of transmission means that the spatial compression is real but incomplete.

Wage slope: The difference between log average wages paid by an employer to workers in jobs of consecutive skill levels within an occupational category, at a given establishment. The paper documents that the wage slope at foreign establishments is correlated with the wage slope at headquarters — a 10 percent greater consecutive-skill wage gap at headquarters is associated with a roughly 1.4 percent greater gap at the foreign establishment — suggesting that the anchoring extends beyond the level of wages to the internal wage structure.

External shocks to headquarter wages: Minimum wage increases in the headquarters country or U.S. state, and exchange rate fluctuations that change the USD value of wages set in local currency. These shocks serve as instruments or quasi-experimental sources of variation in headquarters wages that are plausibly exogenous to conditions at foreign establishments, enabling causal identification of the effect of headquarter wage changes on foreign establishment wages.

Causal forest (heterogeneous treatment effect estimation): A machine learning algorithm used in the paper to estimate the conditional average treatment effect of a minimum wage shock at headquarters, allowing the size of the foreign wage response to vary flexibly with a large set of characteristics (job, employer, sector, headquarter country, establishment country, headquarter-establishment country pair). The resulting predicted treatment effect scores are used to construct above- and below-median transmission groups, which are then compared across observable characteristics to identify what predicts stronger wage anchoring.

Summary based on NBER Working Paper 26788 (February 2020, Revised April 2025). Source text was truncated after the beginning of Section 4.1 (minimum wage event study analysis); all causal evidence descriptions draw on the introduction and Section 3–4 framing rather than the full Section 4 tables and Section 5 heterogeneity analysis. AI-assisted, human review pending.