Core Inflation in the Advanced Economies: A Regional Perspective

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1421 September 2025 Core Inflation in the Advanced Economies: A Regional Perspective Daniel O. Beltran and Julio L. Ortiz Please cite this paper as: Beltran, Daniel O., and Julio L. Ortiz (2025). “Core Inflation in the Advanced Economies: ARegionalPerspective,”InternationalFinanceDiscussionPapers1421. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2025.1421. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Core Inflation in the Advanced Economies: A Regional ∗ Perspective † ‡ Daniel O. Beltran Julio L. Ortiz Abstract We explore differences in the dynamics of core inflation between Europe and North America using a Bayesian time series filter that decomposes the level of core inflation in the major advanced economies into regional, global, and country-specific components. We find a prominent role for both regional and global factors. Historically, the two regional components have at times diverged. Using reduced-form regressions, we examinetheeconomicdriversbehindthechangesinourestimatedglobalandregional components of U.S. core inflation, focusing on the post-pandemic inflation surge and subsequent pullback. The global component is associated with global supply frictions and past energy shocks. The North American regional component is associated with labor market tightness in the region. Key Words: Regional inflation. Dynamic Linear Model. Core inflation. JEL Classification: C11, C32, C53, E31, F00 ∗WethankShaghilAhmed,DaniloCascaldi-Garcia,ThiagoFerreira,LucaGuerrieri,MatteoIacoviello, participantsoftheGWforecastingseminar,andothercolleaguesattheFederalReserveBoardforcomments and suggestions. Annika Johnson provided outstanding research assistance. The views expressed in this notearesolelytheresponsibilityoftheauthorsandshouldnotbeinterpretedasreflectingtheviewsofthe Board of Governors of the Federal Reserve System or of anyone else associated with the Federal Reserve System. †DivisionofInternationalFinance,FederalReserveBoard,20thandCSt.NW,Washington,DC20551. Email: daniel.o.beltran@frb.gov. ‡DivisionofInternationalFinance,FederalReserveBoard,20thandCSt.NW,Washington,DC20551. Email: julio.l.ortiz@frb.gov.

1 Introduction Inflation dynamics sometimes exhibit striking regional patterns that cannot be fully explained by either purely domestic or purely global factors. For example, amid the oil shocks ofthe1970s,cross-countryinflationexperiencesvariedinpartduetooil importdependence which tended to be regional in nature. More recently, energy shocks triggered by geopolitical tensions generated disproportionately large inflationary surges in Europe compared to other regions of the world. Intuitively, inflation could be regional in nature for a couple of reasons. First, countries within a region may share a common exposure to global shocks. This exposure could be shaped by a variety of factors such as the degree to which product, labor, and energy markets are regionally integrated, similar economic policies such as wage indexation, and regional similarities in consumers’ preferences. Second, countries within a region may share a common response to global shocks. For example, countries within a region may adopt similar fiscal and monetary policies to mitigate the effects of an adverse shock. Finally, countries within the same region may face an exogenous regional shock. In this paper, we decompose core inflation across 12 countries into regional, global, and country-specific components with a focus on Europe and North America. We find that regional inflation plays an important role in shaping inflation dynamics. In addition, we document differences in regional inflation experiences between Europe and North America. Moreover,wefindthattheglobalcomponentofcoreinflationalsoexplainsasizableshareof the variation in core inflation rates and we estimate that, at present, the global component is elevated relative to recent history. Finally, in an exploratory exercise, we find that during the post-COVID inflation surge, global inflation is largely associated with supply frictions and past oil price fluctuations whereas regional inflation is largely associated with labor market tightness and consumer spending patterns. Accounting for the presence of a regional component to inflation is important for a number of reasons. First, extracting a regional component of inflation matters for our understanding of global and domestic components, whose importance could be overstated if we were to ignore regional inflation. Second, the presence of a regional component matters for our understanding of the drivers of inflation, which could in turn have monetary policy 1

implications. Third, the existence of regional components of inflation raises the possibility that regional-level information could potentially be useful when forecasting inflation. Obtaining a deeper understanding of regional inflation could be particularly important if the world grows increasingly fragmented (Airaudo et al., 2025; Fern´andez-Villaverde et al., 2024). We start by developing a parsimonious multivariate Bayesian time series filter that is tailored to compare the inflation dynamics between Europe and North America. Our model decomposesthelevelofcoreinflationinthemajoradvancedeconomiesintoregional,global, and country-specific components. More specifically, we assume that core inflation consists of a global and regional level factors which follow random walks with time-varying growth rates, while persistent country-specific factors follow a moving average representation. We estimate the model parameters via maximum likelihood estimation (MLE). Our dynamiclinearmodel(DLM)isestimatedusingmonthlycoreinflationratesfor12advanced economies going back to the 1970s. Filtering these rich monthly time series through our Bayesian state-space filter enables us to estimate the posterior means and variances of the time-varying components at each point in time. With the estimated model in hand, we analyze each inflation component, with a focus on the global and regional components. We identify a prominent role for regional factors, consistent with Mumtaz et al. (2011) who find that regional factors account for the bulk of fluctuations in inflation and output for most countries in their sample. The European regional component tended to be higher than the North America component into the 1990s. From the late 1990s until the start of the COVID pandemic, the two regional components behaved similarly. Amid the COVID inflation surge, however, North America regional inflation peaked earlier and at a higher level than the European regional component. Both regional components have largely retraced their post-pandemic increases. Next, we analyze the extent to which national core inflation rates are explained by the global component that is common to all countries in our sample. Our estimated global component explains a sizable fraction of core inflation in the major advanced economies, consistent with Cascaldi-Garcia et al. (2024) who estimate a global component using a dynamicfactormodel.Theinclusionofregionalcomponentsinourmodelaffectsthedynamics of the estimated global component. 2

Finally, we conduct an exploratory exercise in which we identify the main correlates of global and regional core inflation amid the post-COVID inflation surge. We perform a model selection exercise that considers an extensive set of possible explanatory variables. We adopt a general-to-specific model selection strategy to arrive at a parsimonious linear model. Our results suggest that the global component of core inflation is well-described by supply frictions and past fluctuations in oil prices while the regional component of core inflation is well-described by labor market tightness and consumer spending patterns. Overall, our analysis indicates that regional inflation has played an important role in shaping inflation dynamics over the last several decades. Moreover, our exploratory exercise suggeststhatinrecentyears,globalandregionalinflationhavemovedindifferentdirections and appear to have different drivers. Related literature. Our paper contributes to the literature that studies global inflation. While a longstanding literature has analyzed the degree to which inflation is global in nature (Ciccarelli and Mojon, 2010; Mumtaz and Surico, 2012; Monacelli and Sala, 2009), fewer have studied its regional features. One notable study that includes a regional perspective is Mumtaz et al. (2011) which finds that regional factors account for the bulk of the fluctuations in both inflation and output growth in most countries. On the other hand, Fo¨rster and Tillmann (2014) find that inflation is a local phenomenon. Moreover, while Parker (2018) finds that a global factor can explain much of the variationinnationalinflationratesamongadvancedeconomies,thisfindingdoesnotextend to middle- and low-income countries. Furthermore, when examining inflation subcomponents, the authors find that the common factor of inflation only explains the variation in the energy subcomponent. Their answer to “How global is ‘global’ inflation?” is “Not very.” More generally, our paper relates to a broader literature that decomposes inflation into common and idiosyncratic components. Some of this literature leverages rich crosssectional variation across subcomponents of aggregate inflation, to extract trends in U.S. inflation using dynamic factor models (Stock and Watson, 2016; Ahn and Luciani, 2024). Along these lines, in a recent study of global and idiosyncratic inflation, Cascaldi-Garcia et al. (2024) uses a dynamic factor model to estimate common global components of core, noncore, and headline inflation across 26 advanced and emerging market economies. They 3

find that the global component accounted for a large part of the surge in core inflation in the two periods when world core inflation was high and volatile: between 1960 and the mid-1990s, and since the COVID-19 pandemic. We model country, regional, and global components with a DLM rather than a dynamic factor model. Our baseline DLM nests a simpler DLM with just a common global component and country-specific components. The estimated global component from this simpler DLM is nearly identical to that of Cascaldi- Garcia et al. (2024). We find that our baseline DLM with regional factors is favored by the data based on the log-likelihood. We regard a Bayesian DLM as well-suited for our purposes for a few reasons. First, the parsimonious structure of our DLM model offers easily interpretable global, regional, and country-specificcomponentswhoseposteriormeansandvariancescanbeestimateddirectly using the Kalman filter. The Bayesian approach with conjugate priors greatly facilitates recursive estimation of the posterior densities. Another advantage of the DLM approach is that it readily produces the means and variances of the predictive distributions of both the observables and the states, which can be used to forecast future states and observations. Our paper also relates to the time series literature that models inflation dynamics. Stock and Watson (2007) find that U.S. CPI inflation is well described by an integrated moving average process of order 1, which is equivalent to an unobserved components model in which inflation has a stochastic trend that follows a random walk. We adopt a similar time-series specification but in a multivariate setting, and include both a stochastic global trend that is common to all the countries in the sample as well as a stochastic regional trend that is common to just countries in the same region. The results from our approach, which uses higher-frequency (monthly) aggregate inflation measures, corroborate those of the earlier studies that have found a prominent role for regional and global components of inflation. The rest of the paper proceeds as follows. In Section 2 we describe our data and the model used to illustrate our approach. Section 3 presents our model estimates of the global, regional, and country-specific components. Section 4 identifies the economic drivers of the global and regional components. Section 5 concludes with a discussion of implications for monetary policy. 4

2 Data and Model Specification This section describes the inflation data that we feed into our Bayesian filter to derive the unobserved regional, global, and country-specific components for each country. We then specify the model used for the decomposition. We exclude emerging market economies from our data sample for consistency and ease of interpretation of our results. That is, we intentionally restrict our sample to a set of advanced economies that follow similar inflation dynamics that are well described by a moving average representation similar to the one in Stock and Watson (2007), but in a multivariate setting with common stochastic trends. 2.1 Data The inflation data used for our decomposition are the 12-month percent changes in the national core consumer price indexes for 12 advanced economies: Germany, Italy, France, Portugal, Netherlands, Luxembourg, Finland, Austria, United Kingdom, United States, Canada, and Japan. The core consumer price indexes are from the OECD Main Economic Indicators database.1 The monthly data are sourced from Haver Analytics, and the sample period is January 1971 to July 2025. 2.2 Dynamic Linear Model As shown in equation 1, we decompose each country’s core inflation series, Y , into a it commontime-varyinggloballevelfactor,µGlobal,atime-varyingregionallevelfactor(onefor t Europe, µEurope, and one for North America, µNorthAmerica), and a persistent country-specific t t component that follows a moving average representation. The global and regional factors, defined in equations 2 and 4, are random walks with time-varying growth rates, βGlobal t 1We choose to use core consumer prices rather than headline consumer prices in part because the globalcomponentobtainedfromadecompositionofheadlineinflationwouldlikelyreflectenergyandother commodity price fluctuations. By instead focusing on core inflation, which is often regarded as a strong signal for future headline inflation, our results could have more direct monetary policy implications. In particular,totheextentthatenergyandothercommoditypricesareimportantdeterminantsofanyofour estimated components, it is due to the passthrough of these prices to core consumer prices. 5

and βRegional (one for Europe, βEurope, and another one for North America βN.Amer.). These t t t growth rates, in turn, are modeled as first-order auto-regressive processes with respective decayparametersρ ,ρ andρ (equations3and5).The9Europeancountries Global Europe N.Amer. are assumed to share a time-varying European level component. The United States and Canada share a common time-varying North America component. Japan does not share a common regional component with any other country in our sample, so its inflation is decomposed into a global and a country-specific component. Y = µGlobal +µRegional +ϵ +ψ ϵ , ϵ ∼ N(0,σ2 ) (1) i,t t t i,t i i,t−1 i,t ϵ,i µGlobal = µGlobal +βGlobal +wGlobal, wGlobal ∼ N(0,σ2 ) (2) t t−1 t−1 µ,t µ,t µ,Global βGlobal = ρ βGlobal +wGlobal, wGlobal ∼ N(0,σ2 ) (3) t Global t−1 β,t β,t β,Global µRegional = µRegional +βRegional +wRegional, wRegional ∼ N(0,σ2 ) (4) t t−1 t−1 µ,t µ,t µ,Regional βRegional = ρ βRegional +wRegional, wRegional ∼ N(0,σ2 ) (5) t Regional t−1 β,t β,t β,Regional We estimate the model’s 20 deep parameters—the variances and moving average parameters—via maximum likelihood estimation (MLE).2 To facilitate estimation, we reduce the dimension of the parameter space by calibrating some of the model’s parameters. Specifically, we set σ2 = 0.01 and σ2 = σ2 = 0.001 to ensure that the µ,Global β,Global β,Regional global component has a smooth trend (Ahn and Luciani, 2024; Del Negro et al., 2019). In addition, we assume that the variances of the i.i.d. country-specific shocks are the same for countries in the same region. That is, σ2 = σ2 ϵ,US ϵ,CA σ2 = σ2 = σ2 = σ2 = σ2 = σ2 = σ2 = σ2 = σ2 ϵ,GE ϵ,IT ϵ,FR ϵ,PT ϵ,NE ϵ,LU ϵ,FI ϵ,AU ϵ,UK Finally, we set the variance of the observation errors to σ2 = 0.001, a small value which V 2We also explored the role of parameter uncertainty by assuming loose priors for the 20 parameters shown in Panel A of Table 1, and estimating their posterior distributions using an adaptive Markov Chain MonteCarloalgorithmsimilartotheonedescribedinBeltranandDraper(2017).Forthese20parameters, the resulting posterior modes are nearly identical to their MLE estimates, and their respective Bayesian credible sets are extremely narrow, indicating that they are well-informed by the data. Using draws from thethinnedMCMCchainoftheposteriordistributionofthedeepparameters,were-estimatetheposterior means and variances of the forward-filtered-backward-smoothed state variables over time, and find that they are nearly identical to the ones presented in this paper using the MLE-based calibration. 6

facilitates the recursive estimation while ensuring a close fit to the observed core inflation series for each country.3 Table 1 reports the MLE values estimated for the deep parameters. The moving average coefficients ψ are close to 1 for many countries, indicating strong persistence. The i variance of the North America regional level component is larger than that of the European regional level component. After fixing these deep parameters at their MLE values, we write the model in state-space form and estimate the posterior distributions of the regional, global, and country-specific components recursively forward in time using the Kalman Filter,startingwithanuninformativeprioratt = 0.Becauseweareinterestedinretrospective inspection of the state variables using the full sample period, we use a backward recursive (smoothing) algorithm to compute the conditional distributions of θ for any t < T given t the observations y , starting from the filtering distribution π(θ |y ). We ignore the es- 1:T T 1:T timated posterior means during the first 4 years of our sample to allow the Kalman filter to converge. Estimation details are provided in Appendix A. Our objects of interest are the estimated means and variances of the posterior distributions of the state variables (the levels and growth rates of the global, regional, and country-specific components). Because we are interested in explaining past movements in inflation, we use the forward-filtered-backward-smoothed estimates of the state variables, which are smoother and more precise as they incorporate information from both the past and future observations. 3 Regional, global, and country-specific components The estimated regional component for North America has historically behaved quite differently from the one for Europe. Figure 1a shows the estimated posterior means of the two regional components since 1975 and their respective 90% confidence intervals.4 The Europe regional component was higher than the North America component through much of the 3Our model assumes that regional innovations are orthogonal. We examined the sensitivity of our findings to this assumption by estimating a version of our model in which we allow the Europe and North America regional innovations to be correlated. We find that our results are qualitatively unchanged. 4Wetreattheperiodfrom1971to1974asthetrainingsample,andexcludetheestimatesoftheposterior distributions of the state variables during this period from Figures 1a and 2a. 7

Table 1 Model Parameters Panel A: Parameters Estimated Using Maximum Likelihood Parameter Description Estimate σ2 Variance of country shocks: 3.325 i,Europe GE, IT, FR, PT, NE, LU, FI, AU, UK σ2 Variance of country shock: 0.224 i,NorthAmerica US, CA σ2 Variance of country shock: JP 0.153 JP σ2 Variance of Europe regional shock 0.001 Europe σ2 Variance of North America regional shock 0.009 N.America ψ MA coefficient for Germany 0.979 GE ψ MA coefficient for Italy 0.985 IT ψ MA coefficient for France 0.929 FR ψ MA coefficient for Portugal 0.856 PT ψ MA coefficient for Netherlands 0.964 NE ψ MA coefficient for Luxembourg 0.894 LU ψ MA coefficient for Finland 0.843 FI ψ MA coefficient for Austria 0.984 AU ψ MA coefficient for United Kingdom 0.988 UK ψ MA coefficient for Canada 0.648 CA ψ MA coefficient for United States 0.956 US ψ MA coefficient for Japan 0.752 JP ρ AR1 coefficient for Europe growth rate shock 0.970 Europe ρ AR1 coefficient for N.Amer. growth rate shock 0.967 N.America ρ AR1 coefficient for global growth rate shock 0.957 Global Panel B: Calibrated Parameters Parameter Description Value σ2 Variance of observation error (same for all countries) 0.001 V σ2 Variance of global level shock 0.01 µ,Global σ2 ,σ2 Variance growth rate shocks (Global, Europe, N.Amer.) 0.001 β,Global β,Regional Note. Panel A reports the estimated parameters of the dynamic linear model. Panel B reports externally calibrated parameters. 1970s and 1980s. From the late 1990s through 2019, however, both components were at similar levels. More recently, as seen in Figure 1b, the regional components diverged again. The North America component took off in 2021 and fell back the following year. Europe’s regional component of core inflation also increased in 2021, but not as sharply as the North 8

Figure 1 Europe and North America Regional Components (a) Full sample (b) Since 2018 Percent Percent Percent Percent 8 8 8 8 7 7 7 7 Europe level 6 6 5 5 6 6 4 4 5 5 3 3 4 4 2 2 N.America 1 1 3 3 level 0 0 N.America 2 2 −1 −1 level −2 −2 1 1 Europe level −3 −3 0 0 −4 −4 −1 −1 −5 −5 −6 −6 −2 −2 1970 1980 1990 2000 2010 2020 2018 2019 2020 2021 2022 2023 2024 Note. Thick solid line and thick dot-dashed line show the estimated posterior means of the North America and Europe regional components, respectively. The thin solid and thin dot-dashed lines denote the edges of their respective 90% confidence intervals. Panel (A) plots these components from January 1975 to July 2025. Panel (B) plots these components since 2018. America component. And although the North America component was declining in 2022, the European component continued to rise that year. The mean of the posterior distribution of the “global” component of core inflation is shown in Figure 2a. The global component of core inflation peaked in the 1970s and gradually declined through the 2000s. After the COVID pandemic, as shown in Figure 2b the global component rose sharply. The global component then partially retraced to still-elevated levels as of July 2025. Our model includes the growth rates of the global and regional components as state variables that vary over time, allowing us to assess the degree to which the global and regional components of inflation accelerated or decelerated following the COVID pandemic. As shown in Figure 3a, the global component’s growth rate turns negative when the pandemic arrives in 2020, then shoots up as the global economy begins to reopen, before 9

Figure 2 Global Component Over Time (a) Full sample (b) Since 2018 Percent Percent Percent Percent 20 20 4.0 4.0 19 19 18 18 3.5 3.5 17 17 16 16 3.0 3.0 15 15 14 14 2.5 2.5 13 13 12 12 2.0 2.0 11 11 1.5 1.5 10 10 9 9 1.0 1.0 8 8 7 7 0.5 0.5 6 6 5 5 0.0 0.0 4 4 3 3 −0.5 −0.5 2 2 1 1 −1.0 −1.0 0 0 −1 −1 −1.5 −1.5 −2 −2 −3 −3 −2.0 −2.0 1970 1980 1990 2000 2010 2020 2018 2019 2020 2021 2022 2023 2024 Note. Thick line shows the estimated posterior mean of the global component of core inflation. Thin lines denotetheedgesofits90%confidenceintervals.Panel(A)showstheglobalcomponentfromJanuary1975 to July 2025. Panel (B) shows the global component since 2018. decelerating. The current growth rate has returned to zero, consistent with the flattening out of the global component at its still-elevated level. Figure 3b compares the growth rates of the regional components and shows a sharper acceleration and deceleration in core inflation in North America relative to Europe. While the North America growth rate has returned to zero, the growth rate in Europe remains negative, consistent with continued disinflation. Figure 4 decomposes core inflation for the U.S., U.K., Canada, and France into the global, regional, and country-specific components described earlier.5 The country-specific components generally contribute less than the other components (with the exception of the U.K.), and are more volatile as they capture the higher-frequency noise in core inflation for each country. The country components also reflect differences in levels of core inflation between countries in the same region. For example, UK’s core inflation was higher than 5Appendix B contains similar figures for the other 8 countries in our sample not shown here. 10

Figure 3 Growth Rates of Global and Regional Components, βGlobal and βRegional t t (a) Growth rate global component (b) Growth rate of regional components Percent Percent Percent Percent 0.3 0.3 0.4 0.4 0.3 0.3 0.2 0.2 Europe 0.2 0.2 growth 0.1 0.1 0.1 0.1 0.0 0.0 0.0 0.0 −0.1 −0.1 North −0.2 America −0.2 −0.1 −0.1 growth −0.3 −0.3 −0.2 −0.2 −0.4 −0.4 2018 2019 2020 2021 2022 2023 2024 2018 2019 2020 2021 2022 2023 2024 Note. Panel (A) shows the growth rate of the global component, βGlobal, with the thick line denoting the t posterior mean of the growth rate, and the thin lines denoting the edges of the 90% confidence intervals. Panel (B) shows the growth rates of the North America and Europe regional components, βN.Amer., and t βEur.. Thick solid line and thick dot-dashed line show the estimated posterior means of the growth rates t of the North America and Europe regional components, respectively. Thin solid and thin dot-dashed lines denote the edges of their respective 90% confidence intervals. that of the other European countries in our sample, resulting in a large positive countryspecific component estimated for the UK. The opposite is true for France, which has had smallerratesofcoreinflationrelativetotherestofEurope.InNorthAmerica,theestimated country-specificcomponentfortheU.S.risessharplyin2021,reflectingthesharperincrease in U.S. core inflation relative to Canada’s core inflation. 11

Figure 4 Core Inflation Decomposition for Selected Countries (a) US (b) UK Percent Percent 10 10 10 10 Country−specific Global Country−specific Global 9 Regional 9 9 Regional 9 Core inflation Core inflation 8 8 8 8 7 7 7 7 6 6 6 6 5 5 5 5 4 4 4 4 3 3 3 3 2 2 2 2 1 1 1 1 0 0 0 0 −1 −1 −1 −1 −2 −2 −2 −2 −3 −3 −3 −3 2018 2020 2022 2024 2018 2020 2022 2024 (c) Canada (d) France Percent Percent 10 10 10 10 Country−specific Global Country−specific Global 9 Regional 9 9 Regional 9 Core inflation Core inflation 8 8 8 8 7 7 7 7 6 6 6 6 5 5 5 5 4 4 4 4 3 3 3 3 2 2 2 2 1 1 1 1 0 0 0 0 −1 −1 −1 −1 −2 −2 −2 −2 −3 −3 −3 −3 2018 2020 2022 2024 2018 2020 2022 2024 Note. Each panel decomposes core inflation for a given country. The solid black line is 12-month core inflation,alsoequaltothesumofthecomponents.Thelight-shadedbarsrepresenttheregionalcomponent, the dark-shaded bars the global component, and the cross-hatched bars the country-specific component. 12

4 Explaining the post-COVID surge in U.S. core inflation Having extracted the regional, country-specific, and global components of core inflation for each country, we next examine their economic drivers, focusing on the global component and the North America regional component, which are the two largest components from the decomposition of U.S. core inflation shown in Figure 4. We start by examining the drivers of the global component, which is relatively large and shared by the other countries in our sample. Because our estimated global component is assumed to follow a random walk, our dependent variable is the 12-month change in the estimated global component. 4.1 Drivers of the global component of U.S. core inflation To understand the drivers of the global component, we regress the 12-month change in the global component on a broad set of explanatory variables related to global forces, such as global energy prices, supply frictions, and labor market tightness. To arrive at a parsimonious model, we adopt a general-to-specific model selection strategy that begins witha‘generalunrestrictedmodel’(GUM)thatencompassestheessentialcharacteristicsof the underlying data, and then eliminate variables that are statistically insignificant using Autometrics, which is part of the software package PcGive-OxMetrics (Doornik, 2009; Hendry and Doornik, 2009).6 Our fully-specified GUM includes the following variables in their levels and changes, as well as their lags: global supplier delivery times, global manufacturing backlogs, sea freight shipping costs, brent crude spot price, and the average of the unemployment gap for the countries in our sample. It also includes a constant, trend, and 1-year and 2-year lags of the dependent variable.7 We underscore that this is our own researchanalysisbasedonourparticularmethodologyandthatdifferentmethodologiesmay produce different results. In particular, our reduced form approach treats the regressors as exogenous variables and their changes as structural shocks, when in reality they are likely endogenously determined. Thus, our results do not imply causality and should be 6Autometrics uses a tree-search algorithm to detect and eliminate statistically-insignificant variables, avoidingpath-dependence.Atanystage,avariableisonlyremovedifthenewmodelencompassestheGUM. Theterminalmodelis,bydesign,astatisticallywell-specifiedvalidreductionoftheGUM(Doornik,2008). 7See Data Appendix for a description of the variables considered in the GUM. 13

Table 2 Drivers of Changes in Global Component Estimate HACSE Unemployment gap −0.27816∗∗∗ 0.072151 t Change in sea freight costs 0.027515∗∗∗ 0.0071516 t Change in sea freight costs 0.033495∗∗∗ 0.0052750 t−12 Manuf. backlogs 0.040985∗∗∗ 0.017491 t−6 Change Brent crude spot 0.0027519∗∗ 0.0013871 t Change Brent crude spot 0.0028226∗∗∗ 0.0011110 t−24 Brent crude spot 0.018710∗∗∗ 0.0032281 t−12 Trend −0.0023692∗∗∗ 0.00053646 Number of observations: 235 Adjusted R-squared: 0.7713 F-statistic: 113.8 Notes.Ordinaryleastsquaresregressioncoefficientsandtheirheteroskedasticityandautocorrelationconsistentstandarderrors(HACSE).Dependentvariableis12-monthchangeintheestimatedglobalcomponent. Changes computed on a 12-month basis. All variables are demeaned. The regression is estimated from December 2005 through June 2025. *** denotes 1% significance, ** denotes 5% significance, and * denotes 10% significance. Source. OECD Main Economic Indicators, S&P Global Purchasing Manager’s Index (PMI), Current PopulationSurvey,ProducerPriceIndex,OrganizationofthePetroleumExportingCountries,JapanMinistry of Health Labor and Welfare, Statistics Canada, Instituto Nazionale di Statistica, INSEE, Deutsche Bundesbank, UK Office for National Statistics, Statistics Austria, Instituto Nacional de Estatistica, Statistics Finland, Luxembourg Central Service of Statistics and Economic Studies, Statistics Netherlands; all via Haver Analytics. interpreted merely as suggestive evidence of the underlying drivers. The results are presented in Table 2. All coefficients are statistically significant at the 1% level, and the adjusted R-squared of 0.77 indicates a good overall fit. The negative coefficient on the unemployment gap suggests that the global component is associated with overall labor market tightness in the advanced economies in our sample. The positive coefficients on changes in sea freight shipping costs and manufacturing backlogs suggests that global supply frictions also drive the global component. The positive coefficient on lagged changes and level of Brent crude oil prices is consistent with second-round effects of higher energy prices (Alp et al., 2023). There is also a small, negative trend over time. Figure 5 shows the relative contributions of the drivers since 2019. The post-covid surge in the global component is largely explained by global supply frictions as measured by changes in sea freight costs and global manufacturing backlogs, and to lesser degree, past energy shocks and overall tightness in labor markets across the countries in our sample. 14

Figure 5 Drivers of Changes in Global Component Percent 3.0 3.0 Oil prices Global labor market tightness 2.5 Global supply frictions Trend + residual 2.5 Change in Global Component 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0.0 0.0 −0.5 −0.5 −1.0 −1.0 −1.5 −1.5 −2.0 −2.0 2019 2020 2021 2022 2023 2024 2025 Notes. This decomposition uses the estimated coefficients shown in Table 2, and groups the contributions oftherighthandsidevariablesbybroadcategory.‘Supplyfrictions’includechangeinseafreightcostsand manufacturing backlogs. 4.2 Drivers of changes in the North America component We run a similar exercise to explain changes in the North America regional component, using data for the U.S. and Canada to construct regional averages of variables that measure labor market tightness and changes in consumption and savings.8 Specifically, we regress the 12-month change in the North America component on regional measures of tightness in labor and goods markets, while controlling for its own lags, a constant, and a time trend.9 TheestimatedcoefficientsareshowninTable3.TheresultssuggestthattheNorthAmerica component is largely driven by tightness in U.S. and Canadian labor markets (change in job-openings-to-unemployment ratio and average weekly earnings), and lagged changes in households’ savings behavior (as excess savings during the pandemic boosted consumption spending). 8We ran a similar regression to explain the ‘non-global’ part of U.S. core inflation by replacing the dependent variable with the sum of the 12-month change in the North America regional and U.S. countryspecific components (using the same explanatory variables) and found similar results. 9See Data Appendix C for full list of explanatory variables considered in the GUM. 15

Table 3 Drivers of Changes in North America Component Estimate HACSE Change openings-to-unemp. ratioN.Amer. 2.8512∗∗∗ 0.35003 t Change avg.weekly earningsN.Amer. 0.21601∗∗∗ 0.098672 t Change personal savings rateN.Amer. 0.089840∗∗∗ 0.018180 t−12 Number of observations: 108 Adjusted R-squared: 0.7689 F-statistic: 179 Note. Ordinary least squares regression coefficients and their heteroskedasticity and autocorrelation consistent standard errors (HACSE). Dependent variable is 12-month change in the estimated North America component. Changes computed on a 12-month basis. All variables are demeaned. The regression is estimated from April 2016 through March 2025. *** denotes 1% significance, ** denotes 5% significance, and * denotes 10% significance. Source.OECDMainEconomicIndicators,CurrentEmploymentStatistics,CurrentPopulationSurvey,Job Openings and Labor Turnover Survey, National Income and Product Accounts; all via Haver Analytics. Figure 6 shows the relative contributions of the drivers since 2019. The post-Covid surge in the North America component is largely explained by tightness in regional labor markets, and to lesser degree, excess savings during the pandemic which boosted consumption spending of lower-income households.10 5 Conclusion We examine the international co-movement of core inflation from a regional perspective, using a Bayesian dynamic linear model that decomposes core inflation rates in the major advanced economies since the 1970s into global, regional, and country-specific components. We find that the post-pandemic surge in the global component largely reflected global supplyfrictionsandtoalesserdegree,pastenergypriceshocksandoveralltightnessinlabor markets across many countries. Meanwhile, the surge in the North America component was associated with tightness in labor markets and lagged changes in households’ savings behavior. Our model is optimized for decomposing and explaining past movements in inflation, 10Aladangady et al. (2022) find that spending among households in the bottom half of the income distribution rises by about 10 percent above its pre-pandemic trend following the CARES Act in 2020 and remains well above trend through 2022. 16

Figure 6 Drivers of Changes in North America Component Percent 4.0 4.0 3.5 3.5 3.0 3.0 2.5 2.5 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0.0 0.0 −0.5 −0.5 −1.0 −1.0 −1.5 −1.5 −2.0 Regional savings −2.0 −2.5 Regional labor market tightness −2.5 Residual −3.0 −3.0 Change N.America component −3.5 −3.5 −4.0 −4.0 2019 2020 2021 2022 2023 2024 2025 Notes. This decomposition uses the estimated coefficients shown in Table 3 and groups the contributions of change in openings-to-unemployment and average weekly earnings into the labor market tightness category. Explanatory variables are the averages for the U.S. and Canada. not for forecasting future inflation observations. That said, the Bayesian framework easily allows one to compute the predictive densities of both the state variables and the observables k-steps ahead sequentially using recursive algorithms. The strong persistence in both the levels and the growth rates of the global, regional, and country-specific components should, in principle, facilitate out-of-sample forecasting. Further research could explore how to optimize models of this nature to produce reliable out-of-sample forecasts. Finally, although we focus on a dozen advanced economies with similar inflation experiences over the last 50 years, our model could easily be extended to include more countries. Our results could have implications for monetary policy. Monetary policy could be challenging when the global and regional components of core inflation are moving in opposite directions. For example, in the second half of 2022, the regional component of US core inflation had leveled off and was starting to decline, but this decline was masked by continued increases in the global component, leaving overall core inflation largely unchanged, 17

at elevated levels.11 11In a speech at the time, Federal Reserve Chair Jerome Powell acknowledged that core inflation had “mainly moved sideways” and that “despite tighter policy and slower growth over the past year, we have not seen clear progress on slowing inflation.” Jerome H. Powell, 2022, “Inflation and the labor market”, speechdeliveredattheHutchinsCenteronFiscalandMonetaryPolicy,BrookingsInstitution,Washington D.C., November 30, https://www.federalreserve.gov/newsevents/speech/powell20221130a.htm 18

References Ahn, Hie Joo and Matteo Luciani,“Commonandidiosyncraticinflation,”Financeand Economics Discussion Series, Washington: Board of Governors of the Federal Reserve System 2024. Airaudo, Florencia, Francois De Soyres, Keith Richards, and Ana Maria Santacreu, “Measuring Geopolitical Fragmentation: Implications for Trade, Financial Flows, and Economic Policy,” International Finance Discussion Papers 1408, Washington: Board of Governors of the Federal Reserve System 2025. Aladangady, Aditya, David Cho, Laura Feiveson, and Eug´enio Pinto, “Excess Savings during the COVID-19 Pandemic,” FEDS Notes 2022-10-21, Board of Governors of the Federal Reserve System (U.S.) Oct 2022. Alp, Harun, Matthew Klepacz, and Akhil Saxena, “Second-Round Effects of Oil Prices on Inflation in the Advanced Foreign Economies,” FEDS Notes, Washington: Board of Governors of the Federal Reserve System 2023. Beltran, Daniel O. and David Draper, “Estimating Dynamic Macroeconomic Models: How Informative are the Data?,” Journal of the Royal Statistical Society Series C: Applied Statistics, 11 2017, 67 (2), 501–520. Cascaldi-Garcia, Danilo, Luca Guerrieri, Matteo Iacoviello, and Michele Modugno, “Lessons from the co-movement of inflation around the world,” FEDS Notes, Washington: Board of Governors of the Federal Reserve System 2024. Ciccarelli, Matteo and Benoˆıt Mojon, “Global inflation,” Review of Economics and Statistics, 2010, 92 (3), 524–535. Del Negro, Marco, Domenico Giannone, Marc P. Giannoni, and Andrea Tambalotti, “Global trends in interest rates,” Journal of International Economics, 2019, 118, 248–262. Doornik, Jurgen A., “Encompassing and Automatic Model Selection,” Oxford Bulletin of Economics and Statistics, December 2008, 70 (s1), 915–925. 19

, “Autometrics,” in Jennifer Castle and Neil Shephard, eds., The Methodology and Practice of Econometrics: A Festschrift in Honour of David F. Hendry, Oxford University Press, 2009. Fern´andez-Villaverde, Jesu´s, Tomohide Mineyama, and Dongho Song, “Are We Fragmented Yet? Measuring Geopolitical Fragmentation and Its Causal Effect,” Working Paper 32638, National Bureau of Economic Research June 2024. Fo¨rster, Marcel and Peter Tillmann, “Reconsidering the International Comovement of Inflation,” Open Economies Review, 2014, 25, 841–863. Hendry, David F. and Jurgen A. Doornik, Empirical Econometric Modelling using PcGive: Volume I, London: Timberlake Consultants Press, 2009. Monacelli, Tommaso and Luca Sala, “The international dimension of inflation: evidence from disaggregated consumer price data,” Journal of Money, Credit and Banking, 2009, 41 (1), 101–120. Mumtaz, Haroon and Paolo Surico, “Evolving international inflation dynamics: world and country-specific factors,” Journal of the European Economic Association, 2012, 10 (4), 716–734. , Saverio Simonello, and Paolo Surico, “International Comovements, Business Cycle and Inflation: a Historical Perspective,” Review of Economic Dynamics, 2011, 14 (1), 176–198. Parker, Miles, “How global is ”global inflation”?,” Journal of Macroeconomics, 2018, 58, 174–197. Petris, Giovanni, “An R Package for Dynamic Linear Models,” Journal of Statistical Software, 2010, 36 (12), 1–16. , Sonia Petrone, and Patrizia Campagnoli, Dynamic Linear Models with R useR!, Springer-Verlag, New York, 2009. Stock, James H. and Mark W. Watson, “Why Has U.S. Inflation Become Harder to Forecast?,” Journal of Money, Credit and Banking, 2007, 39 (1), 3–33. 20

and , “What Caused the US Pandemic-Era Inflation?,” Review of Economics and Statistics, 2016, (4), 770–784. and , “Slack and Cyclically Sensitive Inflation,” Journal of Money, Credit and Banking, 2021, 52 (S2), 393–428. 21

A Dynamic Linear Model Representation A dynamic linear model is specified by a Normal prior distribution for the p-dimensional state vector θ at time t = 0, θ ∼ N (m ,C ), and the following equations that govern the 0 p 0 0 evolution of the state-space system for each time t ≥ 1: Y = Fθ +v v ∼ N(0,V ) i,t i,t i,t i,t i θ = Gθ +w w ∼ N(0,W ), i,t i,t−1 i,t i,t i where the p×p matrix G−t and the m×p matrix F are known and v and w are t i,t i,t independent Gaussian random vectors with mean zero and known, constant variances V i and W (Petris et al., 2009). i We next define the building blocks of the model and combine them into a single multivariate model. (a) The MA(1) component for each country is defined as follows. Y = ε +ψ ε N(0,σ2 ) i,t i,t i i,t−1 Y,i The MA components are modeled as two separate state variables, θ and θ , with: 1,t 2,t (cid:104) (cid:105) F = 1 0 , V = 0 (A.1)     0 1 1 ψ G =   W i =  i σ w 2 ,i (A.2) 0 0 ψ ψ2 i i (b) Global and regional components The global mean (µGlobal) and two regional time-varying means (µEurope,µNorthAmerica) t t t are unobserved state variables that follow random walks with a time-varying growth rate. These time varying growth rates, in turn, follow an auto-regressive process of order 1. Combining the moving average components with the global and regional means, the full model is specified as follows. 22

Define Y as the vector of observable 12-month core inflation rates (π ) for the 12 t t countries in our sample: Y = (πGE,πIT,πFR,πPT,πNE,πLU,πFI,πAI,πUK,πCA,πUS,πJP)′ t t t t t t t t t t t t t F is a matrix of dimension 12×30. The first 24 columns contain the F matrix defined in Equation A.1, spanning two columns for each row. Column 25 of F corresponds to the global mean shared by every country, so it contains a 1 on every row. Column 27 of F corresponds to the Europe regional mean shared by the 9 European countries in our sample, so it contains a 1 on rows 1−9. Column 29 of F corresponds to the North America regional mean, shared by the US and Canada, so it contains a 1 on rows 10 and 11.   1 0 ... 1 0 1 0 0 0   0 0 ... 1 0 1 0 0 0     . . . . . . ... . . . . . . . . . . . . . . . . . .       0 0 ... 1 0 1 0 0 0   F =  0 0 ... 1 0 1 0 0 0      0 0 ... 1 0 1 0 0 0     0 0 ... 1 0 0 0 1 0     0 0 ... 1 0 0 0 1 0   0 0 ... 1 0 0 0 0 0 G is a matrix of dimension 30×30, with the first 24 columns containing the G matrix defined in Equation A.2 along the diagonals. The last 6 columns of GG specify that the global and regional components follow a random walk and their respective growth 23

rates follow an AR(1) process.   0 1 ... 0 0 0 0 0 0   0 0 ... 0 0 0 0 0 0      0 0 ... 0 0 0 0 0 0      . . . . . . ... . . . . . . . . . . . . . . . . . .       0 0 ... 1 1 0 0 0 0   G =   0 0 ... 0 ρ 0 0 0 0   Gl.    0 0 ... 0 0 1 1 0 0      0 0 ... 0 0 0 ρ 0 0   Eur.    0 0 ... 0 0 0 0 1 1    0 0 ... 0 0 0 0 0 ρ N.Amer. W is a matrix of dimension 30 × 30. The variances of the shocks to each country’s moving average components are along the diagonals of the first 24 rows, given by the 2×2 W matrix defined in Equation A.2. The variances of the global, Europe, and i North America components, as well as the variances of their respective growth rates are in the last 6 diagonal elements of W.   W ... 0 0 0 0 0 0 i   . . . ... . . . . . . . . . . . . . . . . . .        0 ... σ2 0 0 0 0 0   µ,Global   0 ... 0 σ2 0 0 0 0    W =  β,Global   0 ... 0 0 σ2 0 0 0   µ,Eur.     0 ... 0 0 0 σ2 0 0   β,Eur.     0 ... 0 0 0 0 σ2 0   µ,N.Amer  0 ... 0 0 0 0 σ2 β,N.Amer. Our objects of interest are the time-varying means of the posterior distributions of the levels and growth rates of the global, regional, and country-specific components, and their 90% confidence intervals. For a given country, the posterior means of each component adds up to the level of 12-month core inflation; the unexplained residual in the observed series is negligibly small by construction. 24

All relevant marginal and conditional distributions are Gaussian, and completely determined by their means and variances (Petris et al., 2009). The solution to the filtering problem for DLMs is given by the Kalman Filter. As described in Petris et al. (2009), the conditional distribution of the state variables given the full-information sample y 1:T is also Gaussian, and we compute the forward-filtered-backward-smoothed (two-sided, full sample) estimates using the Kalman smoother. Although the posterior variances can be updated sequentially using the Kalman filter and Kalman smoother, computation suffers from numerical instability and possibly non-symmetric or even negative definite calculated variances. We use the ‘dlm’ package in R (Petris, 2010; Petris et al., 2009) because it has a robust algorithm for calculating the posterior variances by sequentially updating the singular-value decomposition. The Bayesian approach allows us to choose the prior mean for the state vector θ at t t = 0. To calibrate our priors, we use the global and regional averages of the 12-month core inflation rates from December 1970 (before the start of our estimation period), with wide confidence bands to make them less informative. The priors for the state vector is specified below. Our dataset is large, with 655 monthly observations for 12 countries. The priors are swamped by the data such that the results are insensitive to the choice of priors. For example, setting all the prior means to zero produces nearly identical results. µGlobal = N(2.48,10), 0 µEurope = N(2.26,10), 0 µN.America = N(2.04,10), 0 µJP = N(4.65,10), 0 µi = N(0,10)∀i ∈ (GE,IT,FR,PT,NE,LU,FI,AI,UK,CA,US). 0 25

B Core Inflation Decomposition for Selected Countries Figure B.1 Core Inflation Decomposition for Selected Countries (a) Austria (b) Finland (c) Luxembourg Percent Percent Percent 10 10 10 10 10 10 Country−specific Global Country−specific Global Country−specific Global 9 Regional 9 9 Regional 9 9 Regional 9 Core inflation Core inflation Core inflation 8 8 8 8 8 8 7 7 7 7 7 7 6 6 6 6 6 6 5 5 5 5 5 5 4 4 4 4 4 4 3 3 3 3 3 3 2 2 2 2 2 2 1 1 1 1 1 1 0 0 0 0 0 0 −1 −1 −1 −1 −1 −1 −2 −2 −2 −2 −2 −2 −3 −3 −3 −3 −3 −3 2018 2020 2022 2024 2018 2020 2022 2024 2018 2020 2022 2024 (d) Netherlands (e) Portugal (f) Italy Percent Percent Percent 10 10 10 10 10 10 Country−specific Global Country−specific Global Country−specific Global 9 Regional 9 9 Regional 9 9 Regional 9 Core inflation Core inflation Core inflation 8 8 8 8 8 8 7 7 7 7 7 7 6 6 6 6 6 6 5 5 5 5 5 5 4 4 4 4 4 4 3 3 3 3 3 3 2 2 2 2 2 2 1 1 1 1 1 1 0 0 0 0 0 0 −1 −1 −1 −1 −1 −1 −2 −2 −2 −2 −2 −2 −3 −3 −3 −3 −3 −3 2018 2020 2022 2024 2018 2020 2022 2024 2018 2020 2022 2024 (g) Germany (h) Japan Percent Percent 10 10 10 10 Country−specific Global Core inflation 9 Regional 9 9 Country−specific Global 9 Core inflation 8 8 8 8 7 7 7 7 6 6 6 6 5 5 5 5 4 4 4 4 3 3 3 3 2 2 2 2 1 1 1 1 0 0 0 0 −1 −1 −1 −1 −2 −2 −2 −2 −3 −3 −3 −3 2018 2020 2022 2024 2018 2020 2022 2024 Note. Each panel decomposes core inflation for a given country. The solid black line is 12-month core inflation,alsoequaltothesumofthecomponents.Thelight-shadedbarsrepresenttheregionalcomponent, the dark-shaded bars the global component, and the cross-hatched bars the country-specific component. 26

C Data used in regressions The explanatory variables used in the model selection process for the regressions in Section 4 are listed in Table C.1 below. Both the levels and changes in these variables are included in the general unrestricted model (GUM). All are sourced from Haver Analytics. Table C.1 Data series used in regressions in Section 4 Description Haver series name UK Brent crude spot, $ per barrel MGBUKB@ENERGY Global manuf. suppliers’ delivery times, index SGBLMD@MKTPMI Global manuf. backlogs of work, index SGBLMB@MKTPMI Global manuf. new orders, index SGBLMO@MKTPMI U.S. PPI deep sea freight transportation, index R483111@PPIR U.S. PPI deep sea freight transportation, index R483111@PPIR Personal Consumption Expenditures: Goods, SAAR, Bil.$ CTGBM@USECON JOLTS: Job Openings: Total, SA, Thousands LJJTLA@USECON Unemployment, 16yr+, SA,Thousands LTU@USECON Civilian unemployment rate, 16yr+, SA % LR@USECON Civilian participation rate, 16yr+, SA % LP@USECON Personal savings rate, SA, % YPSVRM@USECON Univ. of Michigan: Exp. infl. rate next year, % CINF1@USECON Avg. hours at Work: 16+ NSA, Hrs LENCLWHN@USECON Avg. hrly earnings: Pvt Sector SA, $/Hr AWBWPA@USECON Real disposable pers. income, SAAR, Bil.Chn.2017$ YPDHM@USECON Real retail sales and food services, SA, Mil.1982-84$ NRSTH@USECON Government Social Benefits to Persons, SAAR, Bil.$, GTPFM@USECON To construct the unemployment gap used in the regression of the global component, we first collect unemployment rates for the countries in our sample from Haver. Table C.2 lists the unemployment rate series. Next, we average across countries to obtain an averaged unemployment rate, 1 (cid:88) urate = urate . t jt 12 j We define the unemployment gap as the 12-month change of the 12-month average unemployment rate (Stock and Watson, 2021), ugap = urate −urate . t t t−12 27

Table C.2 Unemployment rates Country Haver series name Japan S158ELUR@G10 U.S. S111ELUR@G10 Canada S156ELUR@G10 UK S112ELUR@G10 France S132EURH@G10 Italy S136ELUR@G10 Austria ATNELCR@ALPMED Portugal S182UR@G10 Finland H172ELUR@G10 Luxembourg S137ELUR@G10 Netherlands S138ELUR@G10 Germany DESE315@GERMANY 28