The Global Pattern: Fast, Uneven, and Changing
The global evidence gives exchange-rate pass-through a measurable baseline, but it does not turn that baseline into a world constant. This chapter asks what the BIS estimates actually show across economies and decades: how quickly consumer prices respond, how the reported advanced-economy and emerging-market patterns differ, and why those differences are an empirical puzzle rather than a ready-made explanation. The value of reading the results at this level is to know what a group average can establish—and what must still be learned from the conditions surrounding each country-period estimate.
A broad sample reveals a fast average response
The BIS study builds its comparison from quarterly consumer-price inflation, nominal effective exchange rates, and output gaps. It retains economies with at least 20 quarters of data within a given decade, producing a sample of 98 economies across the 1980s, 1990s, 2000s, and 2010s. The country count is not identical in every decade: 72 economies appear in the 1980s and 98 in the 2010s as data availability improves, especially for emerging market and developing economies. That expansion increases the range of experience in the sample, while also warning the reader not to interpret every decade contrast as if it came from a perfectly balanced panel.
The minimum-data rule has a useful logic. A country with only a few observations could generate a dramatic-looking coefficient that reflects one unusual episode rather than a persistent relationship. Requiring observations within a decade makes it possible to estimate a local dynamic relationship, though it also excludes countries and periods for which the data are thinner. The resulting averages describe the economies that meet the rule; they are not automatically representative of every country that was absent. This is a general lesson for cross-country macroeconomic comparisons: the denominator is part of the result.
The estimates are dynamic rather than purely contemporaneous. The BIS model allows current and lagged movements in the nominal effective exchange rate to affect quarterly CPI inflation, while also accounting for inflation persistence and the output gap. The study reports that most of the exchange-rate effect appears in the same quarter or the following quarter. Across the full sample, the mean response accumulated over one year is 13.4 percent, with a t-statistic of 9.6. In the paper's specification, that is a precise estimate of a nonzero average relationship; it is not evidence that every economy transmits exactly 13.4 percent, nor that the exchange rate is the only reason consumer prices moved.
A coefficient written as 13.4 percent is easy to misread. In a simplified interpretation, if the relevant exchange-rate movement is normalized to 1 percent and the economy behaved like the estimated full-sample average, the associated CPI response over the specified one-year horizon would be about 0.134 percent in magnitude, subject to the study's sign convention. It is not 13.4 percentage points added to the inflation rate, and it is not 13.4 percent of every individual price. The number summarizes a fitted relationship between changes in an exchange-rate index and changes in an aggregate price index. Its meaning depends on the normalization, the horizon, and the endpoint that produced it.
The distinction between speed and magnitude matters here. A response that appears largely in the current or next quarter tells us that the exchange-rate link is not necessarily a distant, purely long-run phenomenon. But the one-year figure is still a sum of several quarterly responses. It can combine a rapid first movement with smaller later adjustments, or an early response with later offsetting behavior. A reader who sees only the cumulative number cannot tell how much pressure arrived immediately, how much was delayed, or whether different price components moved in opposite directions.
The sample also has a deliberate scale. It is designed to compare the macroeconomic environment of a country in a decade, not to describe the repricing decision of a single firm or product. Consumer-price inflation is an aggregate outcome. It reflects imported goods, domestic services, taxes, wages, margins, demand, and other prices whose exposure to the exchange rate differs. The global estimate is therefore useful for asking why an economy's overall CPI response tends to be larger or smaller than another's. It is not a substitute for a sectoral or firm-level decomposition.
The construction can be pictured as a sequence from observations to a question rather than as a single statistic:
This sequence is important because aggregation happens late. The underlying model first preserves country-specific slopes; only then can the study summarize them for broad income groups and decades. If the order were reversed—if every country were assumed to share one slope at the start—the later differences would be statistically invisible. The global pattern is consequently not just a comparison of averages. It is an average built on an attempt to retain the variation that needs explaining.
The one-year average also sits between two possible mistakes. One mistake is to dismiss it because it is not a firm-level causal estimate. That would throw away useful information about the timing and scale of a macroeconomic relationship. The other is to treat it as a forecast for the next currency movement in every country. That would ignore the state dependence the study is designed to uncover. The proper reading is narrower: under the sample, specification, and horizon used by the BIS, the typical estimated CPI response is positive or negative according to the exchange-rate convention and has a cumulative magnitude summarized by 13.4 percent.
The averages differ, but they do not explain themselves
The BIS results show a marked contrast between the two broad country groups. Over the full sample, the estimated one-year pass-through is 7.6 percent for advanced economies and 17.1 percent for emerging market and developing economies. In the 2010s, the advanced-economy estimate falls further to 4.4 percent, while the EMDE estimate is 20.0 percent. The paper describes the decline over time as especially clear for advanced economies; the broad EMDE estimate changes less in a statistically decisive way and remains within the range associated with the full sample.
The contrast is large enough to matter for interpretation, but its exact size should not be detached from the design. These are means of estimated country-period relationships, not direct observations of a fixed group property. The reported group averages use the study's main specification with controls for international disturbances, and the group labels collect countries with different histories and structures. A reader should therefore take the figures as evidence of a broad difference in the sample, not as a claim that an economy changes its pass-through simply by crossing an income classification.
These figures make a tempting story available: richer economies pass through less, poorer economies pass through more. But the estimates do not establish that income level itself is the mechanism. The group labels collect several characteristics that can change together. Economies differ in their average inflation, exchange-rate volatility, product composition, size, pricing-to-market, policy credibility, and exposure to shocks. The advanced-economy and EMDE averages show that the environments are different in ways that matter for CPI transmission; they do not identify which difference is doing the work.
The time pattern reinforces that caution. The advanced-economy reduction from 7.6 percent over the full sample to 4.4 percent in the 2010s coincides with a period in which inflation and exchange-rate volatility were generally lower in advanced economies, but the BIS analysis does not treat coincidence as proof that one caused the other. Policy frameworks, trade patterns, product markets, and the composition of the sample also changed. Among EMDEs, the reported 2010s estimate of 20.0 percent is higher than the full-sample estimate of 17.1 percent, but that does not mean emerging economies followed a single path in the opposite direction. A group average can move because the underlying countries changed, because the distribution of conditions changed, or because different episodes received different weight.
There is also a difference between a statistical group and an economic type. “Advanced economy” and “EMDE” are useful organizing categories for the reported comparison, but neither is a complete theory of price setting. A small, open economy can share some characteristics with both groups. A country can have sophisticated firms and a volatile currency, or low inflation and a concentrated import basket. The classification helps the reader see broad heterogeneity; it should not be used to predict an individual country's pass-through without examining its actual state.
The changing country coverage makes the same point from another angle. If the 2010s include economies that were not present in the 1980s, then part of an apparent time change may reflect improved representation rather than only a change within the countries observed in both periods. The BIS estimates still provide a valuable long-run comparison, but a strong claim about a global trend would require the reader to ask whether the sample is balanced, how observations are weighted, and whether the same dynamic specification behaves similarly in different groups. The paper's result is best treated as a carefully defined comparison, not as a frictionless time series of one world economy.
The BIS paper's second stage makes the hidden distribution visible. It links 332 country-decade pass-through estimates to a small set of macroeconomic and structural variables rather than asking only why two group means differ. This is a change in question. The first stage asks, in effect, “how much did this country in this period transmit exchange-rate movements to CPI?” The second asks, “which conditions line up with the estimated answer?” The distinction prevents the group average from being mistaken for a causal explanation.
It also changes how to think about the advanced-economy decline. A lower average could reflect a broad shift in the state of transmission—for example, lower inflation or volatility—or a reweighting toward country-periods with conditions that produce lower responses. It could reflect changes in firms' pricing behavior, product composition, or policy credibility. The reported average alone cannot distinguish these possibilities. That is not a weakness unique to this study; it is a general limit of interpreting aggregated coefficients without the variation from which they were formed.
The evidence is still informative even when it does not settle the mechanism. First, the timing suggests that exchange-rate movements can reach aggregate consumer prices quickly enough to matter for near-term inflation assessment. Second, the group contrast says that the magnitude is not portable from one macroeconomic environment to another. Third, the advanced-economy time pattern says that historical averages can become less relevant when the inflation and policy landscape changes. These are substantive conclusions, not merely warnings about methodology.
A useful way to state the result is that the BIS study has identified a conditional empirical surface, not a single height. The 13.4 percent full-sample value locates the center of the broad comparison. The 7.6, 4.4, 17.1, and 20.0 percent figures show that the center shifts across broad groups and periods. But none of those numbers tells us where a particular economy sits today, because the position depends on the variables that differ within and across those categories.
That is why the next analytical step cannot be another ranking of countries. It must examine the drivers that can move an economy from a low-response state to a high-response state. The BIS study uses a random-forest method for this purpose, allowing the relationship between the estimated pass-through and its potential drivers to bend rather than forcing a straight line. Its predictor ranking and partial effects are not causal estimates, but they can reveal patterns that a single average would suppress.
The method's evidence boundary should remain visible while the puzzle is developed. The pass-through estimates are fitted from historical data and depend on the chosen controls, sample, and dynamic specification. The changing country coverage complicates simple decade comparisons. The income groups contain diverse economies. And a statistically precise average can still be a poor forecast if the current state lies far from the observations that generated it. These limits do not make the global evidence unusable; they tell us how to use it.
For a reader, the most defensible interpretation is therefore layered. The global panel supports the claim that exchange-rate pass-through to CPI is often relatively fast. It supports meaningful differences between advanced economies and EMDEs in the reported samples. It supports a further claim that time and country group are not enough to explain those differences. To understand why the surface bends, one must look at inflation, volatility, product markets, economy size, and policy conditions. The global averages have done their job when they turn a supposedly universal number into a clearly defined empirical question.