Policy Research Working Paper 10582 Capital Controls in Emerging and Developing Economies and the Transmission of U.S. Monetary Policy Jongrim Ha Haiqin Liu John Rogers Development Economics Prospects Group October 2023 Policy Research Working Paper 10582 Abstract Emerging markets and developing economies (EMDEs) policy shocks from the literature, this paper tests the propo- exhibit significantly greater volatility in asset returns than sition that countries with less open capital accounts exhibit advanced economies. The commonalities in these returns systematically smaller responses to U.S. monetary policy (and flows) across countries are particularly strong for shocks than low capital control countries. This paper also EMDEs. If these occur independently of the exchange considers the role of other institutional features such as rate regime and if these global financial cycle effects are exchange rate regimes and foreign exchange interventions furthermore independent of countries’ financial openness, in explaining cross-country differences in the responses to the result is Obstfeld (2022)’s “Lemma”: countries can do the shocks. The empirical results suggest that more stringent nothing to decouple from the global financial cycle. Under capital controls exhibit smaller responses of interest rates the prevalent view that U.S. monetary policy is the key and exchange rates to U.S. monetary policy shocks and that driver of the global financial cycle, countries then inherit this result holds more firmly for EMDEs than advanced U.S. monetary policy no matter what they do on exchange economies. In contrast, the analysis finds only weak evi- rates or capital control policies. Using structural vector dence that the degree of exchange rate flexibility affects autoregression models for 78 countries over 1995–2019, U.S. spillovers to foreign interest rates and exchange rates. as well as different methods of identifying U.S. monetary This paper is a product of the Prospects Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at jongrimha@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Capital Controls in Emerging and Developing Economies and the Transmission of U.S. Monetary Policy Jongrim Ha Haiqin Liu John Rogers World Bank Fudan University Fudan University jongrimha@worldbank.org liuhq21@m.fudan.edu.cn johnrogers@fudan.edu.cn∗ Keywords: Federal Reserve; Spillovers; Capital Flow Management JEL codes: E44, E52, F38 ∗ We thank Carlos Arteta, Steve Kamin, Michael Klein, M. Ayhan Kose, Franziska Ohnsorge, Hayley Pallan, Franz Ulrich Ruch, and many other participants in the World Bank seminars for useful comments. The findings, interpretations and conclusions expressed in this paper are those of the authors and should not be attributed to the World Bank, Fudan University, or the institutions the authors are affiliated with. 1 Introduction The current period of unprecedented monetary policy tightening in the U.S. and other advanced economies has given rise to renewed urgency to understanding the extent of international monetary policy spillovers as well as ascertaining the policy options that may be on the table for countries to mitigate potentially harmful effects from these spillovers. According to the classic Trilemma, countries can choose only two of: free international capital mobility, a fixed exchange rate, and sovereign monetary policy. Capital controls allow monetary policy to be set independently of foreign monetary policies while also being used to limit volatility in the exchange rate. Floating exchange rates allow a country to have open international capital markets while its central bank sets monetary policy independently of foreign monetary policies if it so chooses.1 The literature on spillovers and the Trilemma was given significant impetus by work on the Global Financial Cycle (GFC). Rey (2015) and Miranda-Agrippino & Rey (2020) document persuasive evidence that significant commonalities exist between the movements of credit aggregates, credit flows, and asset prices across countries, correlations that are particularly high during crises. Emerging-market economies are particularly subject to the GFC (Kalemli-Özcan 2019). It has been argued that U.S. monetary policy is the key driver of the global financial cycle, a proposition that has key implications for understanding what policy options may or may not be “on the table”.2 In his discussion of Miranda-Agrippino & Rey (2020), Obstfeld (2022) cleverly enunciated the implications of this literature: “If being relatively more closed financially does not insulate one from the Global Financial Cycle, then perhaps we have: – not Trilemma – not even Dilemma – but instead: Lemma. No matter what policies countries follow, they cannot decouple from the Global Financial Cycle.” In this paper, we estimate the spillover effects of identified U.S. monetary policy shocks to 1 Note that the Trilemma does not posit that with floating exchange rates, a country’s domestic financial markets and asset prices are fully insulated from foreign monetary policies and financial developments. Even with a floating exchange rate and independently-set policy interest rates, a country’s stock prices, longer-term bond yields, and other asset values will be influenced by global developments, and the Trilemma does not contradict that. 2 When the Fed eases, the reasoning goes, global intermediaries and other investors are reassured about the economic outlook. Consequently, volatility falls and risk appetite increases, leading in turn to higher leverage and rapid expansion of credit. When the Fed tightens, the process is reversed. See Miranda-Agrippino & Rey (forthcoming) for an exhaustive review of the literature on the GFC. 1 foreign interest rates and exchange rates in a panel of up to 78 countries. Aligning those spillover effects with measures of capital control stringency, we examine if countries with less open capital accounts exhibit systematically smaller responses. We also consider the role of other country factors such as the exchange rate regime, economic region and foreign exchange intervention in explaining cross-country differences in responses to U.S. monetary policy shocks. We find some evidence that countries with more stringent capital controls exhibit smaller responses of interest rates and exchange rates to U.S. monetary policy shocks. This result is more evident for samples of emerging markets and developing economies (EMDEs) and advanced economies (AEs) viewed separately. In contrast, we find only weak evidence that the degree of exchange rate regime flexibility affects U.S. spillovers to foreign interest rates and exchange rates. Our findings indicate that (i) the global financial system lies between the classic Trilemma and Obstfeld’s “Lemma”, and (ii) policy tools like capital controls are still powerful under this backdrop, for EMDEs in particular. Related Literature A vast academic literature since the late 1990s examined a wide range of experiences with capital controls without consensus on their “effectiveness”. Rodrik (1998) found no evidence of a positive correlation between capital account openness and growth or investment/GDP ratios, and argued against capital account convertibility. Edwards (2001) reached the opposite conclusion, using a different measure of capital controls, as did in Grilli & Milesi-Ferretti (1995). Edison et al. (2004) considered a host of international financial openness measures and were unable to find robust evidence that openness accelerates economic growth. Still, Chinn & Ito (2002) found that financial development, measured by private credit creation and stock market capitalization, was negatively correlated with the extent of capital controls, a correlation that was stronger in developed countries with solid institutional frameworks. Dooley (1996), Eichengreen (2002), and Edison et al. (2004) surveyed this earlier literature. Building on this to shed light on the effectiveness of capital controls, Miniane & Rogers (2007) tested the proposition that countries with more stringent capital controls experienced smaller financial market spillovers from U.S. monetary policy shocks. They found no evidence to support this notion of capital control effectiveness. More recently, Georgiadis (2016) assesses spillovers from identified U.S. monetary policy shocks in a global VAR (GVAR) model. He finds that U.S. monetary policy generates sizable output spillovers to the rest of the world, the magnitude of which depends on the receiving country’s trade and financial integration, financial openness, exchange rate regime, financial market development, labor market rigidities, industry structure, and participation in global value chains. Ahmed et al. (2021) find that 2 vulnerabilities in EMDEs matter for the transmission of U.S. monetary policy changes to EMDEs; Camara (2021) addresses the role of information effects; Landi & Schiavone (2021) examine capital controls; and Brandao-Marques et al. (2020) find that having a modern monetary policy framework matters more for EMDEs in monetary transmission than financial development. This work speaks to a broader literature on the role of institutional frameworks. Our analysis using a medium-sized panel of countries complements case studies on the effectiveness of capital controls. For example, Chamon & Garcia (2016) analyze the impact of restrictions on capital inflows that Brazil adopted after 2009. These measures had some success in segmenting the Brazilian and global financial markets but did not translate into significant changes in the exchange rate. They argue that capital controls may have contributed to the sizable depreciation of the real in 2012. In work contemporaneous to our own, Arteta, Kamin & Ruch (2022) examine the importance of spillovers by classifying them according to their underlying cause. They distinguish between the effects of three different types of shocks that can boost U.S. interest rates: (1) inflation shocks, which are prompted by rising expectations of U.S. inflation; (2) reaction shocks, which are prompted by investors’ assessments that the Federal Reserve has shifted toward a more hawkish stance; and (3) real shocks, which are prompted by anticipation of strengthening economic activity. The authors note that although sources (1) and (2) should lead to deleterious spillovers, the third one should not. Finally, although the empirical characterization of the global financial cycle described above (co-movements of risky asset prices) is widely accepted, there are critiques of the literature. For example, Cerutti et al. (2019) report OLS regression R2 values indicating that U.S. fundamentals explain at most 25% of the variation in capital flows. Applying a robust variance decomposition method, Habib & Venditti (2019) use a combination of zero and sign restrictions to identify four shocks: U.S. monetary policy; U.S. aggregate demand; a global financial shock; and a geopolitical risk shock. They find that changes in global risk caused by pure financial shocks have the largest impact on capital flows, larger than U.S. monetary policy shocks. Boehm & Kroner (2020) emphasize the importance of U.S. macroeconomic news on global risky asset prices. They show that U.S. macro news can explain as much as 15% of the quarterly variation in foreign stock markets and take issue with the conclusion that U.S. monetary policy drives global financial conditions. Rogers, Sun & Wu (2022) document new evidence that U.S. corporate bond spreads play the predominant role in driving the global financial cycle. According to their findings, the hegemony of the U.S. in the global financial system may be rooted in its powerful financial intermediaries rather than the Fed itself. 3 2 Data and Methodology 2.1 Data Description and Variable Construction The variables we use for the U.S. economy include first a measure of U.S. monetary policy shocks (MP), as explained in the next section. We also control for domestic credit conditions and financial market liquidity by including the excess bond premium (EBP) from Gilchrist et al. (2021). Other U.S. variables include commodity prices (CP), consumer price index (P), industrial production index (IP), and the 10-year government bond yield (R10).3 As our main focus is to evaluate the efficacy of capital controls, we begin considering which non-US (“foreign” hereafter) countries to include in the study. We do this based on the availability of data on capital controls. We use the overall capital restriction index (KC) by Fernández, Klein, Rebucci, Schindler & Uribe (2016) (FKRSU hereafter) in our main analysis. This measure covers 99 foreign countries over 1995-2019, ranging from 0-1, with 1 indicating the greatest restriction.4 For each of these 99 countries, we collect exchange rate per USD (in logs and denoted as S) from the IMF’s International Financial Statistics (IFS) database. We obtain foreign interest rate (R∗ ) and industrial production (IP∗ ) from three data resources: IFS; Federal Reserve Economic Data (FRED); and Haver Analytics. We supplement interest rate data with the BIS policy rates, and also statistics published by countries’ central banks if possible. When multiple interest rate series are available, we choose the one closest to the U.S. 3-month Treasury bill rate.5 6 We select data series—and hence countries—based on the criterion that the series has no more than five missing values.7 Among the 99 countries, we find 44 with sufficient data for all foreign variables, hence are allowed to be included in a “Full” VAR analysis described later. There are 34 3 The specific series we use are: “Producer Price Index by Commodity: All Commodities”(PPIACO), “Con- sumer Price Index: Total All Items for the United States”(CPALTT01USM661S), “Industrial Production: Total Index”(INDPRO) and “Long-Term Government Bond Yields: 10-year: Main (Including Benchmark) for the United States”(IRLTLT01USM156N). They are all downloaded from FRED St. Louis and adjusted for seasonality by Fed X-11 routine whenever needed. In computing CP, P and IP, we take the log of each raw series and multiply by 100. 4 The measure integrates 0/1 indicators of controls in 32 transaction categories. Each of these indicators belongs to one of the 10 asset categories, and represents restrictions either on inflows or outflows. The subset of indicators under each asset category are then averaged to get two indicators - one for outflow, another for inflow. FKRSU then define the overall inflow (outflow) restrictions index - KCI (KCO) - as the average of the 10 inflow (outflow) indices. Our main KC measure is the average of KCI and KCO. We check robustness to use KCI or KCO, or on selected asset categories. 5 We refer to the IFS World Notes to determine the exact maturities of each interest rate variable for each country. For instance, when there are no series explicitly defined as “3M Tbills”, we search for descriptions like “91-day” or “3-month”, if these do not exist either, we further try “6M”, “9M” etc. We examine robustness to a sub-sample of countries for which only short-term rates with maturities under 1-year are available. 6 To enlarge sample size, we use interest rate series with the most sufficient data. In subsection 6.3, we also look at specific “policy rate” following the definition of De Leo et al. (2022). 7 Relaxing this criterion up to 12 enlarges the sample only by retaining small islands or countries with annual GDP less than 20 bil.$; while further restricting the cutoff to below 5 would drop some Latin American countries like Argentina and Chile with interesting capital control histories. 4 additional countries for which S and R* are available (but not IP*). For the other 21 countries, they are insufficient in either S or R*. Since we wish to analyze the responses of exchange rates and interest rates, we keep the former 34 countries in a “Partial” analysis which investigates S and R* responses in models without IP*. We drop the other 21 countries. This selection produces a broader sample of 78 countries, 34 of which we are unable to control for output. See Table A2 for details. 2.2 Monetary Policy Shocks and Econometric Methodology As our baseline measure of U.S. monetary policy shocks, we use the policy news shock (NS shock) of Nakamura & Steinsson (2018).8 As highlighted by Nakamura & Steinsson (2018), the NS shock is similar to the “path factor” in Gurkayanak et al. (2005). It contains an information component that alters private-sector beliefs in non-monetary fundamentals, in contrast with the implications of standard New Keynesian model featured by price rigidity. To check robustness, we consider two alternative measures: the “unified measure” of Bu, Rogers & Wu (2021) (BRW shock) and the forward guidance shock (SS-FG shock) of Swanson (2021). The BRW shock is derived from a two-step, partial-least squares estimation using daily interest rate data across a wide spectrum of maturities. It has three appealing features, which together distinguish it from other shock series in the literature. First, by using the full maturity spectrum of interest rates, this series is able to stably bridge periods of conventional and unconventional monetary policy. Second, the shock is largely devoid of the central bank information effect, the notion that monetary policy announcements also reveal information regarding the central bank’s future macroeconomic outlook (Nakamura & Steinsson 2018). Finally, the Bu, Rogers & Wu (2021) monetary policy shock series is largely unpredictable from available information, including Blue Chip forecasts, “big data” measures of economic activity, news releases and consumer sentiment.9 These considerations make it more plausible to assume that the shock series is exogenous and thus appropriately to be placed first in the ordering of a Sims (1980) style VAR. Finally, we use Swanson (2021)’s forward guidance shock (SS-FG), the extended “path factor” of Gurkayanak et al. (2005) that covers the post-2005 period. We plot all three shock series in Figure 1. In subsection B.3, we also apply the sign restriction identification in Jarociński & Karadi (2020) to disentangle the pure monetary policy shock (JKMP) 8 It is the first principal component of unexpected changes over 30-minute windows around FOMC announcements in five underlying variables: the Fed funds rate following the FOMC meeting, the expected Fed funds rate following the next meeting, and three-month eurodollar future rates with two, three, and four tenors. The original NS series spans 1995-2014, and was extended to 2019 by Acosta & Saia (2020). We drop unscheduled FOMC meetings to avoid potential noise and set shocks in no-meeting months to zero as the standard practice. 9 See Ramey (2016), Miranda-Agrippino (2016), and Bauer & Swanson (2020) for critiques of earlier monetary policy shock series that exhibited predictability. 5 and a central bank information component (CBI). Figure 1: U.S. Monetary Policy Shocks Notes: (1) The figure shows monetary policy shocks at each scheduled FOMC meeting (200 dates in total); (2) NS is the “policy news shock” downloaded from Miguel Acosta’s website; (2) BRW is the pure MP shock constructed by Bu, Rogers & Wu (2021) (Updated: March 4, 2021); (3) SS-FG is the forward guidance shock downloaded from Eric Swanson’s website. We estimate the international spillover effects of identified shocks to U.S. monetary policy using Cholesky VARs (Sims 1980), a recursive model imposing the ordering (MP, CP, IP, P, IP∗ , R∗ , EBP, R10, S) for the 44 countries for which all series are available.10 For the broader sample that adds 34 countries, we estimate the VAR without IP∗ . We include three lags in all cases, as suggested by the Akaike Information Criterion. As in Bu, Rogers & Wu (2021) and noted above, we place the monetary policy shock first because it is exogenous by construction and so is consistent with the identifying assumption that it does not respond contemporaneously to other variables in the system. We also add a dummy for the Zero Lower Bound (equals to 1 during 2009M1-2015M12) to allow for structural change in the deterministic component.11 We concentrate on the impulse responses of the exchange rates and foreign interest rates. To assess the role of capital controls (and also other country factors as introduced in the next section), we test whether the impulse responses of the foreign interest rates and exchange rates are 10 Prices and outputs (CP, IP, P, IP∗ ) and the exchange rate S are in logs, while interest rate variables (R∗ , R10 and EBP) are in percentage. 11 The results are robust with or without the dummy, or to using a dummy of the global financial crisis instead. 6 significantly different for, e.g., “strict” versus “loose” capital controls regimes or based on different degrees of exchange regime flexibility. This follows Miniane & Rogers (2007). Of the different ways to report results from the panel of countries, we choose two. In the first, we estimate the first principal component (PC1) from the vector of time series in the group of countries within relevant percentiles across the capital controls distribution.12 We then estimate the 9-variable VAR with the U.S. variables plus the within-group PC1 of IP*, R*, and S. In the second, we estimate the VARs country-by-country and report median impulse responses within country groups. Since the results are largely robust, we report “PC1 IRFs” in the main text and “Median IRFs” in the Appendix.13 2.3 Country Factors We display our main capital control measures for the broad sample of 78 countries in Figure 2. The two panels depict our KC measures across two time ranges: 1995 vs. 2007 and 2007 vs. 2019. From the top panel, we see that many countries liberalized their capital markets during the early period of our sample. However, these “Early Liberalizers” kept their KC stances largely unchanged after 2007, as implied by the narrow domain of countries around the 45-degree line in the bottom panel. In subsection 6.2, we provide evidence that capital account restrictiveness typically persists, in that countries switch their KC stances very infrequently. That said, capital controls episodically imposed on selected sets of assets can differ from controls that are longstanding and broad. As highlighted by Klein (2012) and Klein & Shambaugh (2015), the duration and scope of capital controls matter in evaluating their efficacy. Thus, we follow Klein (2012) to classify countries as {“Open”, “Gate” or “Wall”} and compare their responses to U.S. monetary policy shocks.14 We show in Figure A1 how our KC index maps into the episodic vs. longstanding distinction over time. The much more frequent flips in KC stances are evident among the “Gates”, while in sharp contrast the “Walls” (“Open”) stayed in high (low) level of capital controls over the whole sample. We display the Open-Gate-Wall classification directly in Table A1 for each country together with average KC index in ascending order. We see that (i) countries with high (low) capital controls are typically emerging (advanced) economies, (ii) there are many (50%) “Gate” countries crowded in the middle of the overall KC distribution, and (iii) countries with high average 12 We standardize the PC1 to have zero mean and unit standard deviations. 13 An alternative approach is to include three PC1 (or median) in one VAR and estimate a 13-variable model. This to some extent controls for interactive impacts among countries. We find essentially the same as results as with the 9-variable model, so we stick to the smaller system to maintain parsimony. 14 “Open” (“Wall”) countries are those with capital controls on less than 10% (more than 90%) of transaction subcategories over the sample, and never have any year with controls on above 20% (below 60%) of subcategories. Those neither “Wall” or “Open” are classified as “Gate”. As expected, a “Wall” country imposes capital controls persistently and pervasively, while a “Gate” imposes capital controls from time to time over a limited set of assets. 7 KC frequently break their “walls” (e.g. Myanmar, Bangladesh and Vietnam), while countries low in average KC also shut their “gates” periodically (e.g. Uganda and Israel). Figure 2: Scatter Plots of KC Measures Notes: The figure shows capital controls index for all 78 countries. An index of 0 (1) indicates the most open (restrictive) capital accounts. Red dots represent EMDEs, and blue dots represent AEs. To compare with country factors other than capital controls, with particular focus on the policy side, we further analyze the roles of exchange rate regime (EA) and foreign exchange intervention (FXI). We show their correlations with our main KC in Table 1, and refer the readers to Table A1 for statistics on each country. Table 1 confirms that our baseline KC has strong negative correlations with alternative measures of financial openness, both de facto and de jure. We also find our overall capital restriction index is highly correlated with restriction on capital inflow and outflow. On the other hand, we find that countries with less flexible exchange rate regimes have higher 8 capital restrictions, more frequent foreign exchange interventions, and greater trade exposure to the United States. Moreover, countries with more restrictive capital control policies have significantly lower degree of trade integration with the US. On the other hand, FXI tends to be a substitute for KC, while there is little statistical significance, unless one investigates the de facto capital controls. We find no strong relationship between capital flow management and the degree of dollar utilization. Table 1: Cross-correlation Among Country Factors KC KC FOpen FOpen KS US Variables KC EA PEG FXI DOL Inflow Outflow Chinn-Ito (Ha) PEG TR KC 1.00 KC 0.98* 1.00 Inflow KC 0.99* 0.93* 1.00 Outflow FOpen -0.93* -0.91* -0.92* 1.00 Chinn-Ito FOpen -0.67* -0.69* -0.64* 0.75* 1.00 (Ha) EA -0.59* -0.58* -0.59* 0.58* 0.53* 1.00 PEG 0.40* 0.34* 0.44* -0.39* -0.16 -0.54* 1.00 KSPEG 0.42* 0.37* 0.46* -0.37* -0.16 -0.66* 0.93* 1.00 FXI -0.05 -0.11 0.01 0.14 0.30* -0.36* 0.23 0.41* 1.00 USTR -0.29* -0.28* -0.28* 0.35* 0.35* 0.48* -0.19 -0.18 0.01 1.00 DOL 0.03 0.05 0.01 -0.13 -0.39* -0.41* 0.19 0.32* 0.07 -0.36* 1.00 Notes: (1) KC is the average capital control index, a larger value indicates greater stringency. KC Inflow (Outflow) is the average index for restrictions on capital inflow (outflow). (2) FOpen (Chinn-Ito) is the Chinn-Ito index on capial account openness proposed by Chinn & Ito (2006); FOpen (Ha) is a de-facto financial openness measure defined as the sum of international assets and liabilities in percent of GDP (Ha et al. 2019), winsorized at 99% and standardized to 0-1. (4) EA is the average value of monthly classification from Ilzetzki et al. (2021), a higher value indicates greater exchange rate flexibility; PEG and KSPEG are the average of the peg indicator from Shambaugh (2004) and Klein & Shambaugh (2008) respectively, a higher value indicates greater fixity. (5) FXI is the standard deviation of a broad foreign exchange interventions proxy defined by Adler et al. (2021) divided by the sum of standard deviations of this proxy and log change in the exchange rates against USD. (6) USTR is the updated trade integration measure in Miniane & Rogers (2007). It is total imports from and exports to the U.S. divided by annual GDP denominated in USD. (7) DOL is the deposit dollarization measure provided by Christiano et al. (2021), and supplemented by the dollar asset share measure in Adrian & Xie (2020). (8) * indicates significance at 90%. Our large sample of countries allows us to conduct separate analysis of AEs and EMDEs, a crucial distinction emphasized by Kalemli-Özcan (2019). Table 2 highlights the much greater volatility of EMDE exchange rates and asset returns, and also more unstable inflation and output growth compared to AEs. It also shows the much more volatile interest rates, exchange rates and prices for countries with higher KC, more floating EA regimes or fewer foreign exchange interventions. 9 Table 2: Macroeconomic Volatility By Country Groups By Economic Region By KC By EA By FXI Partial Full U.S. AE EMDE Low High Fixers Floaters Low High σ∆S 7.96 2.82 / 2.36 3.41 2.19 3.33 2.40 3.19 3.47 2.35 [3.68] [2.61] / [2.33] [3.05] [2.16] [3.06] [2.29] [2.93] [3.20] [2.21] σR 15.65 10.41 1.56 3.12 14.46 3.49 13.39 6.01 13.1 13.6 8.32 [7.34] [5.21] [1.56] [2.42] [9.63] [2.62] [7.80] [3.63] [6.79] [8.04] [4.24] σ∆IP 2.96 2.96 0.65 2.79 3.21 3.07 2.84 3.09 2.82 2.58 3.33 [2.57] [2.57] [0.65] [2.32] [2.98] [2.58] [2.56] [2.75] [2.39] [2.28] [3.01] σ∆P 10.17 0.85 0.27 0.29 1.26 0.33 1.13 0.48 1.10 1.09 0.78 [1.99] [0.53] [0.27] [0.27] [0.95] [0.28] [0.78] [0.37] [0.69] [0.72] [0.53] N 78 44 1 27 17 22 22 22 22 17 17 Notes: (1) The table reports the standard deviations among all countries in each group indicated by each column. The bracket below each cell is the average standard deviation for each country’s time series. (2) Partial stands for all 78 “Partial analysis” countries and Full stands for the 44 “Full analysis” countries. All the other columns except US report statistics for some subsets of the “Full analysis” sample. (3) Low (High) KC refers to countries with average KC index below (above) the median. Fixers (Floaters) refers to countries with average EA below (above) the median. Low (High) FXI refers to countries with FXI below (above) the median. (4) ∆S is monthly change in the log of exchange rate per USD, and ∆IP (P) is monthly change in log of IP (CPI). N counts the number of countries in each group. 3 Baseline SVAR Results 3.1 U.S. Model with the Global Factor We first estimate an “aggregate open economy” VAR including U.S. variables and the global factor constructed by Miranda-Agrippino & Rey (2020). The global factor is a sort of summary statistic of the fortitude of global asset prices. Movements in the factor thus provide an aggregate view of U.S. monetary policy transmission. As shown in Figure 3, the impulse responses of U.S. variables following a U.S. monetary policy shock are conventional, and they suggest an information component in both the NS and forward guidance shocks as expected. Furthermore, we observe an immediate drop in the global factor following a U.S. monetary tightening. This is in line with Miranda-Agrippino & Rey (2020)’s finding that U.S. monetary policy is an important driver of the Global Financial Cycle. Hereafter, we suppress domestic responses to save space. 10 Figure 3: Responses of U.S. Variables and the Global Factor Notes: (1) The figure presents impulse response functions (IRFs) to a U.S. monetary policy shocks (normalized to have 100bp hikes on U.S. one-year Treasury yields), estimated by Cholesky VAR with a dummy for ZLB period (2009M1-2015M12) in the deterministic component. (2) Each column presents VAR model using the indicated shock. (3) Dashed lines show 68% confidence intervals (also for all the following figures). 3.2 Panel Responses by Capital Control Stringency We now present our baseline results, comparing impulse responses across “country groups” based on average capital controls. Specifically, we divide countries into three groups, covering 20% bandwidth equally spaced in the overall capital control distribution. That is, the top 20%, the 40%-60%, and the bottom 20%, respectively. Figure 4 compares responses of foreign economies with different positions in the capital controls distribution of our 44 “full analysis” countries. Specifically, we include the first principal component (PC1) of R*, S, and IP* extracted from countries falling into the top, middle, and bottom 20% 11 groups, respectively, sorted by their average capital control index over 1995-2019. In Figure B2, we present results for the broader sample of 78 countries, and in Figure B3, we present median impulse responses for countries within each group. As seen from the first three rows of Figure 4, the impact of U.S. monetary policy shocks on foreign interest rates is generally larger for countries with lower capital controls. For the group of low KC countries, there is a small yet pronounced rise in interest rates in the near term, while for the middle and high KC groups, interest rate movements are largely independent of U.S. rates, with high KC countries even moving in the opposite direction on impact. These findings provide some evidence supporting the prediction that for a particular foreign country, greater capital account stringency can insulate domestic financial conditions from U.S. monetary policy shocks. However, this additional buffering effect of capital controls is relatively small in economic magnitude, and is not firmly robust to different shock measures (especially for the forward guidance shock). This is reconfirmed by Figure B2 and Figure B3 showing that the responses in the high KC group are noticeably smaller than those in the low and middle KC ones, but again, the differences are not large and appear to be somewhat sensitive to the measures of U.S. MP shocks. We show in section 5 that the somewhat mixed relationships between R* responses and capital controls can be partly explained by the distinctive nature of advanced economies and emerging market economies in their exposure to the U.S. and macroeconomic and financial market volatility. 12 Figure 4: Impulse Responses to U.S. MP Shocks by KC Percentiles, Category 1 Panel (PC1) Notes: (1) This figure shows impulse response functions (IRFs) of foreign interest rates (R*) and exchange rates (S) from 9 separate models - 3 shocks × 3 country groups by capital controls. (2) Each VAR model includes six US variables and three PC1 series on R*, S, and foreign industrial production (IP*). (3) Subplots from left to right are IRFs of PC1 from countries in the bottom 20%, 40%-60%, and top 20% KC percentiles, with sample size of 9, 10 and 9 countries respectively. (4) The first three rows show IRFs of R* (PC1), and the last three show IRFs of S (PC1). We depict the responses of exchange rates in the last three rows of Figure 4. We observe a strong depreciation of foreign currencies following the Fed tightening in low (and middle) KC countries. On the contrary, the exchange rate responses are largely muted in high KC countries. Compared with the results from the interest rate responses, these findings on exchange rates are rather robust, as can be seen from the bottom three rows of the two Appendix Figures. 3.3 Country-Specific Responses and Capital Control Stringency We now turn to more granular analyses based on our country-by-country VAR estimates. In Figure 5 we display maximum impulse responses against the average KC for each country. The 13 scatter plots generally suggest that capital controls are more effective in EMDEs (red-colored) than AEs (blue-colored) in the sense that the negative relationship between the strength of U.S. monetary spillovers and the degree of capital controls is clearer in EMDEs and are broadly consistent across different types of U.S. MP shocks. Figure 5: Impulse Responses to U.S. MP Shocks against Capital Controls, Country-specific Results Notes: (1) This figure shows scatter plots of maximum response of foreign interest rates (R*) or exchange rates (S) and average capital controls over 1995-2019, based on country-specific SVAR models for 44 countries. (2) Advanced economies (AEs) are marked in blue while emerging market and developing economies (EMDEs) in red. (3) Dash lines are OLS fits of maximum impulse responses on average KC using AE and EMDE samples, respectively. 4 Relevance of Exchange Rate Regimes and FX Interventions In this section, we relate our SVAR results to two other policy factors, exchange rate regime (EA) and foreign exchange intervention (FXI), as well as their interactions with capital controls. 14 Figure 6: Impulse responses to U.S. MP Shocks by exchange rate flexibility Notes: This figure shows impulse responses of foreign interest rates (R*) and exchange rates (S), based on the first principal component for each variable, across three country groups with different degree of foreign exchange flexibility – from left to right, “Fixed”, “Crawling”, and “Floating” countries, respectively. Figure 6 shows R* and S responses for countries with different degrees of exchange rate flexibility. We find no significant differences in the response of R* among countries with different exchange rate flexibility despite the differences in the response of S, which is consistent with the definition of EA itself. In contrast, there appears to be some significant relationship between R* responses and foreign exchange interventions. 15 Figure 7: Impulse responses to U.S. MP Shocks by foreign exchange interventions Note: This figure shows impulse responses of foreign interest rates (R*) and exchange rates (S), based on the first principal components for each variable, across three country groups with different degree of foreign exchange interventions (FXI)–from top to bottom, “bottom 20%”, “40-60%”, and “top 20%” FXI countries, respectively. The number of countries is 7 in each group. As shown in Figure 7, the degree of responsiveness of R* to U.S. monetary shocks decreases systematically as the degree of FXI strengthens. While the relationships between S responses and FXI are mixed, this result is consistent with theoretical findings by Itskhoki & Mukhin (2022) suggesting that capital controls are less efficient when foreign exchange interventions are already implemented, combined with optimal monetary policy. In addition, it also partly points to the “inefficacy” of exchange rate flexibility we documented above: a country’s interventions to FX market developments (in response to a U.S. monetary policy shock) can assist in offsetting the spillovers, rather than its currency regime per se. In Figure 8 and Figure 9, we display scatter plots of country-specific IRFs against EA and FXI, respectively. On the one hand, we again find little difference between exchange rate regimes and the degree of U.S. spillovers. On the other hand, the efficacy of FXI (in reducing the spillovers) seems 16 more obvious in EMDEs according to R* responses, while the evidence on S responses is somewhat mixed. Figure 8: Maximum Responses of R* and S to U.S. MP Shocks vs. Exchange Rate Regimes Note: This figure shows maximum responses of foreign interest rates (R*) or exchange rates (S) and average exchange rate regime (EA) over 1995-2019, based on country-specific SVAR models for 44 countries. Figure 9: Maximum Responses of R* and S to U.S. MP Shocks vs. Foreign Exchange Interventions Note: This figure shows maximum responses of foreign interest rates (R*) or exchange rates (S) and average foreign exchange intervention (FXI) over 1995-2019, based on country-specific SVAR models for 44 countries. This highlights a trade-off for policy makers in EMDEs between stabilizing domestic markets in the face of external shocks and potential distortionary effects due to policy interventions. In this context, countries are expected to have an optimal choice of policy tools, combining multiple tools at the same time, with interactions among the policy tools reinforcing their benefits while dampening 17 some negative consequences compared to when a single instrument is used. This issue of integrated policy framework has recently been discussed widely Das, Gopinath & Kalemli-Özcan (2022), and formally modeled by Basu, Boz, Gopinath, Roch & Unsal (2020). We discuss these issues further in section 5. 5 Regional Divergence and KC-EA Interaction So far, we have found mixed results on the relationship between country characteristics such as capital controls and U.S. monetary spillovers. Motivated by existing work on the divergence between responses of emerging markets and advanced economies to external shocks such as Kalemli-Özcan (2019), Hoek et al. (2021), Ahmed et al. (2021) and De Leo et al. (2022), we examine the relationship between KC and U.S. monetary policy shocks separately for EMDEs and AEs. Table 3: U.S. Monetary Policy Transmission Across KC and Income Groups Max. R* Responses Max. S Responses Group NS BRW SS-FG NS BRW SS-FG Panel A: Full Sample All 1.37 1.07 0.49 0.23 0.06 0.09 [-0.47, 4.79] [-0.70, 4.12] [-0.03, 0.14] [0.02, 0.45] [-0.03, 0.19] [0.01, 0.17] Low KC 1.22 1.20 0.21 0.23 0.07 0.09 [0.13, 2.25] [0.44, 2.31] [0.00, 0.06] [0.11, 0.36] [0.00, 0.14] [0.05, 0.14] High KC 1.52 0.87 0.75 0.23 0.05 0.06 [-0.08, 4.12] [-0.03, 3.14] [0.00, 0.19] [0.08, 0.39] [0.00, 0.13] [0.02, 0.12] Panel B: AE Sub-sample AE All 0.88 1.03 0.17 0.25 0.08 0.10 [0.14, 1.70] [0.47, 1.68] [-0.04, 0.44] [0.13, 0.37] [0.01, 0.14] [0.06, 0.15] Low KC 1.15 1.06 0.18 0.25 0.08 0.09 [0.22, 1.70] [0.55, 1.68] [-0.04, 0.48] [0.14, 0.37] [0.01, 0.15] [0.05, 0.15] High KC 0.58 1.02 0.15 0.24 0.08 0.10 [-0.01, 1.69] [0.36, 1.58] [-0.04, 0.42] [0.12, 0.37] [0.00, 0.14] [0.06, 0.15] Panel C: EMDE Sub-sample EMDE All 3.04 1.17 0.94 0.20 0.04 0.05 [-0.09, 6.45] [-0.34, 4.37] [0.03, 2.56] [0.05, 0.34] [-0.01, 0.11] [0.02, 0.10] Low KC 4.31 1.37 1.55 0.14 0.03 0.05 [1.19, 9.64] [-1.24, 6.23] [0.18, 3.20] [0.03, 0.28] [-0.02, 0.11] [0.02, 0.11] High KC 1.36 1.01 0.52 0.23 0.04 0.06 [-0.33, 3.86] [-0.01, 2.89] [-0.07, 1.81] [0.07, 0.37] [0.00, 0.11] [0.02, 0.09] Notes: (1) This table reports the median of maximum IRFs over 1-30 months for countries in each sub-group as indicated by the Group column; (2) Statistics in the brackets show the lower and upper bound for 68% confidence band; (3) “High (Low) KC” indicates countries with average KC above (below) median. As indicated by Panel B and C of Table 3, we find that the negative relationship between impulse responses and capital controls becomes stronger and robust once we split the sample into AE 18 and EMDE. Within AE or within EMDE, the magnitude of U.S. spillovers to foreign interest rates systematically decreases in capital account stringency. This decline is much larger for EMDEs. In the meantime, we find that EMDEs exhibit greater R* responses to U.S. monetary policy shocks. Since EMDEs typically apply more restrictive capital controls (Table A1), this implies that full-sample results (section 3, Panel A of Table 3) can understate the efficacy of capital controls. On the other hand, we find in the last three columns in Table 3 that the overall negative correlations between S responses and capital controls disappear once we consider the AE-EMDE division. The reason seems to be that countries with higher capital controls (mostly EMDEs) maintain more fixity in their exchange rates. Thus, for one thing the negative relationship between KC and S responses we see in Panel A could be due to the fact that EMDE countries have higher KC while at the same time have smaller responses in S since they are fixed. In addition, it is unsurprising to see this relationship vanish once we control for the AE-EMDE distinction. What confounds our analysis of S responses here is likely to be the exchange rate regimes. Table 4 justifies this argument, by investigating impulse responses within EA groupings. As evident in this table, the negative relationship between S responses and capital controls becomes rather robust once we control for the degree of exchange rate flexibility. Meanwhile, it is unsurprising to see that the negative relationship between R* responses and KC does not persist when looking only at “Fixers” or “Floaters”, since we show in Table 3 that what really matters in the case of R* responses is the AE-EMDE dimension. Simply controlling for EA here does not disentangle the true relationship between R* responses and KC. 19 Table 4: U.S. Monetary Transmission Across EA and KC Groups Max. R* Responses Max. S Responses Group NS BRW SS-FG NS BRW SS-FG Panel A: Fixer Sub-sample Fixer 1.34 1.07 0.23 0.23 0.05 0.08 [0.04, 2.74] [0.36, 2.64] [-0.09, 0.79] [0.09, 0.35] [0.00, 0.13] [0.02, 0.12] Low KC 1.15 1.09 0.12 0.23 0.07 0.09 [0.02, 1.92] [0.54, 2.44] [-0.12, 0.48] [0.10, 0.35] [0.00, 0.14] [0.05, 0.13] High KC 1.52 1.07 0.51 0.17 0.04 0.04 [0.13, 4.04] [0.05, 3.14] [-0.06, 1.54] [0.04, 0.27] [-0.02, 0.10] [0.01, 0.09] Panel B: Floater Sub-sample Floater 1.43 1.00 0.80 0.24 0.07 0.10 [0.12, 3.41] [-0.11, 2.73] [0.17, 1.81] [0.09, 0.43] [0.01, 0.14] [0.04, 0.15] Low KC 1.28 1.15 0.35 0.22 0.09 0.10 [0.22, 2.48] [0.18, 2.43] [0.02, 0.79] [0.11, 0.38] [0.02, 0.14] [0.04, 0.15] High KC 1.59 0.71 1.25 0.27 0.05 0.09 [-0.34, 4.80] [-0.20, 2.97] [0.25, 2.65] [0.09, 0.49] [0.00, 0.14] [0.04, 0.16] Notes: (1) This table reports the median of maximum IRFs over 1-30 months for countries in each sub-group as indicated by the Group column; (2) Statistics in the brackets show the lower and upper bound for 68% confidence band; (3) “Floater (Fixer)” indicates countries with average EA above (below) median. The above analyses suggest that the divergence between AEs and EMDEs is central to our analysis. In Table 5, we look more closely into countries within the EMDE, exploiting the rich regional diversity in our sample. In this way, we provide a more granular view on the evaluation of capital control effectiveness. First, we point out that there exist some salient outliers whose deviations from the general picture are not explained by our country factors in a systematic way and might be due to unobserved country heterogeneity (institutional background, historical reasons, etc.). As suggested by the relative R* responses, among countries in the 25 lowest degree of KC, the interest rate responses are smaller and less significant for Netherlands, Belgium, France, and Zambia. Furthermore, within the group of 25 countries with the highest degree of KC, we find countries—Brazil, Mexico, Angola, the Russian Federation, and Kazakhstan in particular—that exhibit relatively strong R* responses. On the other hand, the S responses of these countries are still in line with our previous results, even though they behave counter-intuitively in the case of R* responses. Second, we confirm in Table 5 that our analyses still apply when regional economic features are considered. For example, among the seven largest EMDEs (EM7)–i.e., Brazil, China, India, Malaysia, Mexico, Türkiye, and Russia; the seven countries considered to be substantial in economic scales compared to other EMDEs–for those that use capital controls more pervasively like China, India, 20 and Malaysia, the responses of interest rates are weaker than the median. Meanwhile in Türkiye, Russia, Mexico, and Brazil, where capital control stringency is relatively lower (KC below median), the sizes of the R* responses are all higher than those in the median EMDE. These arguments also apply to the exchange rate responses. In the East Asia and Pacific (EAP) area, where all the countries have above-median capital restrictions, the responses of interest rates and exchange rates are all below the median. A similar pattern is observed in the economic regions of Middle East and North Africa (MNA) and South Asia (SAR), where the responses of interest rates and exchange rates were mostly below median. On the other hand, in the Latin America and Caribbean (LAC) region, where most of the countries have below-median capital controls, the responses of interest rates are relatively sizeable in most cases. Likewise in the Europe and Central Asia (ECA) region, the same pattern exists in particular with respect to the responses of exchange rates. Table 5: Relative KC and Impulse Responses By EMDE Regions Region Country KC R∗ S Region Country KC R∗ S EAP Indonesia 0.13 -0.03 0.20 ECA Bulgaria -0.34 1.17 0.46 Thailand 0.22 0.01 0.06 Hungary -0.28 0.04 0.11 Myanmar 0.27 -0.17 -0.32 Kyrgyz Rep. -0.25 0.33 0.45 Malaysia 0.30 -0.18 0.01 Romania -0.20 1.08 0.65 Philippines 0.34 -0.07 0.14 Türkiye -0.06 1.03 1.73 Vietnam 0.37 -0.18 -0.42 Kazakhstan -0.03 0.32 0.04 China 0.44 -0.16 -0.43 Russian Federation 0.10 1.51 2.00 LAC Uruguay -0.51 0.72 0.36 Poland 0.21 0.07 -0.07 Peru -0.51 0.17 -0.00 Ukraine 0.28 0.26 0.34 Costa Rica -0.47 -0.07 -0.83 MNA Egypt, Arab Rep. -0.34 -0.10 -0.33 Nicaragua -0.47 0.08 -0.67 Kuwait -0.16 -0.18 -0.29 Paraguay -0.41 0.27 -0.13 Morocco 0.25 -0.10 0.14 Bolivia -0.34 -0.03 -0.32 Algeria 0.34 -0.03 0.02 Ecuador -0.12 0.01 3.16 SSA Zambia -0.51 2.00 0.82 Chile -0.12 0.01 -0.00 Uganda -0.41 -0.15 -0.32 Jamaica -0.04 -0.05 -0.81 Mauritius -0.40 0.02 0.51 Argentina -0.01 2.71 -0.22 Nigeria -0.29 0.12 1.19 Mexico 0.09 0.83 0.09 Kenya -0.19 -0.02 -0.37 Brazil 0.11 0.09 0.53 Ghana 0.01 0.12 -0.08 Colombia 0.12 -0.04 0.36 South Africa 0.13 -0.04 1.76 SAR Pakistan 0.21 0.05 -0.29 Eswatini 0.32 -0.11 1.81 Bangladesh 0.31 -0.16 -0.56 Angola 0.35 5.13 2.76 India 0.45 -0.11 -0.00 Tanzania 0.41 -0.12 -0.44 Notes: (1) KC is the relative KC, i.e. KC minus the median KC among EMDE countries; R∗ (S) stands for standardized R* (S) impulse responses, i.e. subtracted by the median and divided by standard deviation. Hence it’s straightforward to evaluate the pattern by the “ −” signs, since negative (positive) numbers suggest values below (above) the median. (2) We take the median over the three MP shocks in reporting IRFs; (3) EAP: East Asia and Pacific, ECA: European and central Asia, LAC: Latin America and Caribbean, MNA: Middle East and North Africa, SAR: South Asia, and SSA: Sub-Saharan Africa. 21 6 Robustness 6.1 Different Types of Capital Controls? In this section we check the robustness of our key results using alternative measures of capital controls or financial openness. In Figure 10, we regress interest rate responses (both point estimates, and the 68% confidence band) at each horizon, on different capital control measures, controlling for all the other country factors including EA, FXI, an indicator for EM, the deposit dollarization (DOL) measure constructed by Christiano et al. (2021), and the U.S. trade share (USTR) as defined in Miniane & Rogers (2007).15 As seen from the graph, a higher capital restriction (or lower financial openness) suggests a lower response to U.S. monetary policy shocks, and this conclusion holds for a variety of capital control measures. 6.2 Endogenous Capital Controls? Conclusions concerning the relationship between capital controls and U.S. spillovers rely on the assumption that capital controls are exogenous. If countries enact more stringent controls in response to undesirable external shocks, the statistical findings will be biased (downward for high KC groups in particular). Table 6 reports countries’ transition probabilities among KC quintiles. Clearly, there is a strong tendency that KC stances maintain rather than jumping around from time to time. This evidence of hysteresis in capital controls provides some support for the exogeneity assumption. Table 6: Transition Probabilities Among KC Quintiles N qt = 1 N qt = 2 N qt = 3 N qt = 4 N qt = 5 N qt+1 = 1 0.9375 0.0740 0.0105 0.0041 0.0022 N qt+1 = 2 0.0473 0.8603 0.1477 0.0062 0.0043 N qt+1 = 3 0.0152 0.0603 0.7996 0.0372 0.0043 N qt+1 = 4 0.0000 0.0055 0.0380 0.8822 0.1041 N qt+1 = 5 0.0000 0.0000 0.0042 0.0702 0.8850 Note: Markov transition matrix of countries moving from the ith capital control quintile at time t to the j th quintile at time t + 1, estimated by maximum likelihood from annual data of 99 countries. To address the potential endogeneity in capital controls, we estimate state-dependent impulse 15 Data resource: the Direction of Trade Statistics (DOTS). 22 Figure 10: Capital Controls and Interest Rate Responses at Each Horizon Notes: (1) each point reported in the graph comes from a multivariate OLS regression. The dependent variable is the median or upper/lower bound impulse response at corresponding horizon, the independent variables are (i) capital control measures as labeled on the left of the figure (ii) a variety of other country factors including EA, FXI, EM, DOL and USTR. (2) KC Inflow (Outflow) is the average capital restriction index solely for capital inflow (outflow) as in FKRSU. KC 95-13 is an earlier version of the KC index constructed by FKRSU, while contains a smaller set of assets (10 categories) and ends at 2013. (3) The Chinn-Ito financial openness index is from Chinn & Ito (2006). (4) The de-jure financial openness is from Ha et al. (2019). 23 responses exploiting sudden changes in countries’ capital control status. First, we identify sudden jumps in capital controls based on the concept proposed by Klein (2012). In particular, for each country i in year t, we define i “Shutting Gates” if (i) in the last five years, its average capital control was lower than 0.1, and its maximal control was lower than 0.2; (ii) the capital control index at t was greater than 0.2. Similarly, we define i “Breaking Walls” if (i) in the last five years, its average capital control was greater than 0.7, and its minimal control was lower than 0.6; (ii) the capital control index at t was lower than 0.6.16 Figure 11 highlights the “Shutting Gates” (in sand) and “Breaking Walls” (in blue) events for each country that we identify to had experienced staggered capital control changes. As can be seen from the graphs, we identify in total 30 countries to have sudden jumps in KC status. These countries had a quite low (high) level of capital restrictions for a relatively long period, and then suddenly closed (opened) their capital accounts, in most of the cases, once and for all. (a) Wall Breakers (b) Gate Shutters Figure 11: Shutting Gates and Breaking Walls: Country-wise Study Note: the thick black line shows countries’ overall capital restriction index; sand (blue) dots show the timing of “shutting gate” (“breaking wall”) Adopting these country-wise sudden changes in countries capital control status, we then estimate a state-dependent local projection based on regime shifts in capital controls: P re P ost Yit+h = αih + (1 − zit )βh M P St−1 + zit βh M P St−1 + (1 − zit )Πh (L)Yi,t−1 + zit Γh (L)Yi,t−1 + uit , h = 0, . . . , H (1) 16 These cutoffs follow Fernández, Klein, Rebucci, Schindler & Uribe (2016)’s definition of “Open” and “Walls”, yet is extended to investigate time variations in countries’ Open-Gate-Wall status. We look at five year window as a baseline, which is both to guarantee a persistent enough “pre-shock” status, and also to allow for enough variation given our sample period. However, the results are quite robust to different window, and various cutoffs, which are the two dimensions to determine how “staggered” the shock is, and its “magnitude” respectively. 24 Where Y is the vector of the variables in our baseline VAR model, and the incidence function zit = 1 if country i was after a Gate-shutting / Wall-breaking at time t, and 0 otherwise. M P S indicates three types of monetary policy shocks we consider in this paper.17 Figure 12 shows the estimation results. As evident in the state-dependent impulse responses, countries’ interest responses to U.S. monetary policy shocks increases (decreases) dramatically after a wall-breaking / gate-shutting, and the differences are strong and statistically significant. (a) R* Response Pre vs Post Wall Breaking (b) R* Response Pre vs Post Gate Shutting Figure 12: Impact of Staggered Change in KC on U.S. Monetary Spillovers 6.3 Different Types of Interest Rates? Our baseline choice of interest rates in the main text is meant to enlarge the sample size as possible. As can be seen from Table A2, the foreign interest rate variables cover various types, which might have distinctive behaviors following a U.S. monetary policy shocks. In this section, we follow De Leo et al. (2022) to look at the responses of policy rate in particular.18 As shown in Figure 13, the effect of capital controls on mitigating U.S. monetary spillover is even stronger than we document in the main text, highlighting the stronger consequence of capital controls on guaranteeing monetary independence. 17 This specification is a simplified version of Auerbach & Gorodnichenko (2012), who use a continuous probability density predicting economic booms and recessions. 18 The main data source is BIS policy rate database. For countries whose data is unavailable in BIS, we supplement from IFS and Bloomberg. The more restrictive data requirement makes us lose several countries in our sample. 25 Figure 13: Policy Rate Responses to U.S. Monetary Policy Shocks by KC Percentiles 6.4 Are U.S. Monetary Policy Spillovers Important? Table 7 shows the maximum variance shares of the monetary policy shock for foreign interest rate and exchange rate respectively, over all forecasting horizons. It can be seen that the variance share of U.S. monetary policy is typically lower than 10%. The figures are larger for exchange rate responses than for interest rates. There is some evidence that the share of volatility explained by U.S. MP shocks is lower for high capital control countries, while again this is somewhat sensitive to the choice of shock measures. The minor figures shown in the variance decomposition results cast doubt on the importance of U.S. monetary spillovers in a broad context. Our main analysis has been focusing on evaluating the efficacy of countries’ policies on reducing this spillover effect, while it may impose little power on economic condition for the foreign economy in the first place. This thus calls for considerations of other dimensions of international policies like liquidity effects on the financial market, coordination between the public and private sectors and so forth, in addition to isolating the effects of U.S. monetary policy shocks. 26 Table 7: Maximum Share of Variance Explained By U.S. MP Shocks (in %) Foreign interest rates Exchange rates Low 20% 40-60% Top 20% Low 20% 40-60% Top 20% NS 2.20 2.08 2.02 16.60 9.30 4.36 [0.40, 12.45] [0.25, 10.14] [0.41, 8.16] [3.35, 37.89] [0.90, 27.55] [1.13, 14.95] BRW 5.56 8.02 4.96 3.32 5.05 8.73 [1.00, 18.09] [0.86, 21.49] [0.87, 15.18] [0.45, 16.25] [0.54, 18.72] [0.75, 28.90] SS-FG 2.06 1.49 3.84 11.53 7.50 4.95 [0.51, 7.72] [0.25, 7.33] [0.82, 11.61] [1.40, 25.18] [0.47, 20.22] [0.71, 19.17] Notes: (1) The table shows the maximum share of U.S. monetary policy shocks in a forecast error variance decomposition over 30-month horizons. (2) The lower (upper) bound of 68% confidence bands are shown in the brackets. 7 Conclusion Emerging markets and developing economies (EMDE) exhibit significantly greater volatility in asset returns and output growth than advanced economies. Across countries, there are significant commonalities in these risky asset returns and flows, as documented by the influential literature on the Global Financial Cycle (GFC). The effects on EMDEs are particularly strong and occur independently of the exchange rate regime (Rey 2015, Miranda-Agrippino & Rey 2020, Kalemli- Özcan 2019, Miranda-Agrippino & Rey forthcoming). If these effects are furthermore independent of countries’ financial openness, the classic Trilemma becomes Obstfeld (2022)’s “Lemma”: countries can do nothing to decouple from the GFC. Under the view that U.S. monetary policy is the key driver of the GFC, countries may inherit U.S. monetary policy no matter what they do on exchange rates or capital controls policy. Empirical results in this paper suggest that countries with more stringent capital controls exhibit smaller responses of interest rates and exchange rates to U.S. monetary policy shocks and that this result holds more firmly for EMDEs than advanced economies. In contrast, we find only weak evidence that the degree of exchange rate regime flexibility affects U.S. spillovers to foreign interest rates and exchange rates. Our findings indicate that (i) the global financial system lies between the classic Trilemma and Obstfeld’s “Lemma”, and (ii) policy tools like capital controls are still powerful under this backdrop, for EMDEs in particular. 27 References Acosta, M. & Saia, J. (2020), ‘Estimating the effects of monetary policy via high frequency factors’, Unpublished working paper. Columbia University . Adler, G., Chang, K. S., Mano, R. & Shao, Y. (2021), Foreign exchange intervention: A dataset of public data and proxies, International Monetary Fund. Adrian, M. T. & Xie, P. (2020), The Non-US Bank Demand for US Dollar Assets, International Monetary Fund. Ahmed, S., Akıncı, Ö. & Queraltó, A. (2021), U.S. Monetary Policy Spillovers to Emerging Markets: Both Shocks and Vulnerabilities Matter, International Finance Discussion Papers 1321, Board of Governors of the Federal Reserve System (U.S.). Arteta, C., Kamin, S. B. & Ruch, F. U. (2022), How Do Rising U.S. Interest Rates Affect Emerging and Developing Economies? It Depends, Policy Research Working Paper Series 10258, World Bank, Washington DC. Auerbach, A. J. & Gorodnichenko, Y. (2012), Fiscal multipliers in recession and expansion, in ‘Fiscal policy after the financial crisis’, University of Chicago Press, pp. 63–98. Basu, S. S., Boz, E., Gopinath, G., Roch, F. & Unsal, F. D. (2020), A conceptual model for the integrated policy framework, Working paper no. 2020/121, International Monetary Fund. Bauer, M. D. & Swanson, E. T. (2020), The Fed’s Response to Economic News Explains the “Fed Information Effect", CESifo Working Paper Series 8151, CESifo. Boehm, C. E. & Kroner, T. N. (2020), The US, Economic News, and the Global Financial Cycle, Working Papers 677, Research Seminar in International Economics, University of Michigan. Brandao-Marques, L., Gelos, R. G., Harjes, T., Sahay, R. & Xue, Y. (2020), Monetary policy trans- mission in emerging markets and developing economies, Working paper no. 2020/035, International Monetary Fund. Bu, C., Rogers, J. & Wu, W. (2021), ‘A unified measure of fed monetary policy shocks’, Journal of Monetary Economics 118, 331–349. Camara, S. (2021), ‘Spillovers of us interest rates: Monetary policy & information effects’, arXiv preprint arXiv:2111.08631 . 28 Cerutti, E., Claessens, S. & Rose, A. K. (2019), ‘How important is the global financial cycle? evidence from capital flows’, IMF Economic Review 67(1), 24–60. Chamon, M. & Garcia, M. (2016), ‘Capital controls in brazil: effective?’, Journal of International Money and Finance 61, 163–187. Chinn, M. D. & Ito, H. (2002), Capital account liberalization, institutions and financial development: Cross country evidence, Working Paper 8967, National Bureau of Economic Research. Chinn, M. D. & Ito, H. (2006), ‘What matters for financial development? capital controls, institutions, and interactions’, Journal of Development Economics 81(1), 163–192. Christiano, L., Dalgic, H. & Nurbekyan, A. (2021), Financial dollarization in emerging markets: Efficient risk sharing or prescription for disaster?, National Bureau of Economic Research. . . (2022), Preemptive policies and risk-off shocks in Das, M., Gopinath, G. & Kalemli-Özcan, S emerging markets, Technical report, National Bureau of Economic Research. . . (2022), Monetary policy cyclicality in emerging De Leo, P., Gopinath, G. & Kalemli-Özcan, S economies, Working Paper 30458, National Bureau of Economic Research. Dooley, M. P. (1996), ‘A survey on controls over international capital transactions’, IMF Staff Papers 43, 639–687. Edison, H. J., Klein, M. W., Ricci, L. A. & Sløk, T. (2004), ‘Capital account liberalization and economic performance: Survey and synthesis’, IMF Staff Papers 51(2), 220–256. Edwards, S. (2001), Capital mobility and economic performance: Are emerging economies different?, Working Paper 8076, National Bureau of Economic Research. Eichengreen, B. (2002), ‘Capital account liberalization: What do cross-country studies tell us?’, World Bank Economic Review 15(3), 341–365. Fernández, A., Klein, M. W., Rebucci, A., Schindler, M. & Uribe, M. (2016), ‘Capital control measures: A new dataset’, IMF Economic Review 64(3), 548–574. Georgiadis, G. (2016), ‘Determinants of global spillovers from us monetary policy’, Journal of International Money and Finance . Gilchrist, S., Wei, B., Yue, V. Z. & Zakrajšek, E. (2021), The Term Structure of the Excess Bond Premium: Measures and Implications, Policy Hub 94155, Federal Reserve Bank of Atlanta. 29 Grilli, V. & Milesi-Ferretti, G. (1995), ‘Economic effects and structural determinants of capital controls’, IMF Staff Papers 42, 517–551. Gurkayanak, R., Sack, B. & Swanson, E. (2005), ‘Do actions speak louder than words? the response of asset prices to monetary policy actions and statements’, International Journal of Central Banking 1, 55–93. Ha, J., Kose, M. A. & Ohnsorge, F. (2019), Inflation in emerging and developing economies: Evolution, drivers, and policies, World Bank Publications. Ha, J., Kose, M. A. & Ohnsorge, F. (2021), One-Stop Source: A Global Database of Inflation, Policy Research Working Paper Series 9737, World Bank, Washington DC. Habib, M. M. & Venditti, F. (2019), The global capital flows cycle: structural drivers and transmission channels, Working Paper Series 2280, European Central Bank. Hoek, J., Kamin, S. B. & Yoldas, E. (2021), Are Rising U.S. Interest Rates Destabilizing for Emerging Market Economies?, FEDS Notes 2021-06-23-2, Board of Governors of the Federal Reserve System (U.S.). Ilzetzki, E., Reinhart, C. M. & Rogoff, K. S. (2021), Rethinking exchange rate regimes, Working Paper 29347, National Bureau of Economic Research. Itskhoki, O. & Mukhin, D. (2022), ‘Optimal exchange rate policy’, working paper. Jarociński, M. & Karadi, P. (2020), ‘Deconstructing monetary policy surprises—the role of information shocks’, American Economic Journal: Macroeconomics 12(2), 1–43. . . (2019), U.S. Monetary Policy and International Risk Spillovers, Working Paper Kalemli-Özcan, S 26297, National Bureau of Economic Research. Klein, M. W. (2012), Capital controls: Gates versus walls, Working Paper 18526, National Bureau of Economic Research. Klein, M. W. & Shambaugh, J. C. (2008), ‘The dynamics of exchange rate regimes: Fixes, floats, and flips’, Journal of International Economics 75(1), 70–92. Klein, M. W. & Shambaugh, J. C. (2015), ‘Rounding the corners of the policy trilemma: sources of monetary policy autonomy’, American Economic Journal: Macroeconomics 7(4), 33–66. 30 Landi, V. N. & Schiavone, A. (2021), ‘The effectiveness of capital controls’, Open Economies Review 32(1), 183–211. Miniane, J. & Rogers, J. (2007), ‘Capital controls and the international transmission of u.s. money shocks’, Journal of Money, Credit and Banking 39, 1003–10036. Miranda-Agrippino, S. (2016), Unsurprising Shocks: Information, Premia, and the Monetary Trans- mission, Working Paper 626, Bank of England. Miranda-Agrippino, S. & Rey, H. (2020), ‘US monetary policy and the global financial cycle’, Review of Economic Studies 87(6), 2754–2776. Miranda-Agrippino, S. & Rey, H. (forthcoming), The global financial cycle, in ‘Handbook of Interna- tional Economics’, Vol. 5. Nakamura, E. & Steinsson, J. (2018), ‘High-frequency identification of monetary non-neutrality: the information effect’, Quarterly Journal of Economics 133(3), 1283–1330. Obstfeld, M. (2022), Discussion of “Global Footprints of Monetary Policies” by Miranda-Agrippino, Nenova, and Rey, Power Point Presentation, ASSA Annual Meeting. Ramey, V. A. (2016), ‘Macroeconomic shocks and their propagation’, Handbook of macroeconomics 2, 71–162. Reinhart, C. M. & Rogoff, K. S. (2004), ‘The modern history of exchange rate arrangements: a reinterpretation’, Quarterly Journal of economics 119(1), 1–48. Rey, H. (2015), Dilemma not Trilemma: The Global Financial Cycle and Monetary Policy Indepen- dence, Working Paper 21162, National Bureau of Economic Research. Rodrik, D. (1998), ‘Who needs capital account convertibility?’, Princeton Essays in International Finance 207, 55–65. Rogers, J., Sun, B. & Wu, W. (2022), U.S. monetary policy and the global financial cycle, working paper, Fudan University. Shambaugh, J. C. (2004), ‘The effect of fixed exchange rates on monetary policy’, Quarterly Journal of Economics 119(1), 301–352. Sims, C. A. (1980), ‘Macroeconomics and reality’, Econometrica pp. 1–48. 31 Swanson, E. T. (2021), ‘Measuring the effects of federal reserve forward guidance and asset purchases on financial markets’, Journal of Monetary Economics 118, 32–53. 32 Appendix A Data Sources and Detailed Descriptions A.1 Construction of Country Factors In this section, we describe in detail the data collection of our country factors, and alternative measures we use for robustness. First, we display the evolution of our main KC index for each country in Figure A1. In particular, we look at countries in one of the three categories {“Open”, “Gate”, “Wall”} as defined in Klein (2012). Figure A1 shows that the dynamics in our KC measure maps well into the Open/Gate/Wall classification. Figure A1: Evolution of Capital Restrictions By Open/Gate/Wall Categories Notes: (1) The figure plots time series of capital restriction index for countries in one of the three groups – “Open” (“Wall”) countries as those who have capital controls on less than 10% (more than 70 %) of their transaction subcategories over the sample period, and do not have any years in which controls are on more than 20% (less than 60 %) of their transaction subcategories. “Gate” countries are those neither “Wall” nor “Open”. (2) We further split “Gate” countries into two groups – “Gate-Shutters” are those who have lower KC for years prior to 2009 than the later period, and “Wall-Breakers” are the reverse. In addition to different variations of our main capital control index (FKRSU), we also use alternative measures of financial openness retrieved from a variety of resources to check robustness. First, we use the Chinn-Ito index proposed by Chinn & Ito (2006), which is a de jure financial openness measure based on the binary dummy variables that codify the tabulation of restrictions on cross-border financial transactions reported in the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER).19 We also consider a de facto financial openness measure 19 The updated series covers 1970-2020 for 182 countries and is downloaded from Chinn-Ito’s website. 33 (FOpen Ha) defined as the sum of international assets and liabilities in percent of GDP calculated by Ha et al. (2019). The major proxy for exchange rate regime we use is the de facto measure proposed by Ilzetzki et al. (2021). They classify countries into six categories: 1. Fixed; 2. Crawling pegs; 3. Managed floating; 4. Free floating; 5. Free falling; 6. Dual market in which parallel market data is missing. Since we focus on U.S. transmission, countries pegged with a non-USD currency are reclassified as 20 Free floating. In cross-sectional analyses, we average the EA index over our sample period, and label countries with the overall average ranging [1,2] as “Fixed” countries, (2,3] as “Crawling” and (3,5) as “Floating”.21 To check robustness, we also use the indicators for pegs by Shambaugh (2004) (PEG) and Klein & Shambaugh (2008) (KSPEG).22 These two alternate measures take the value of 1 if the country’s currency is pegged with a certain base, and is 0 otherwise (hence are essentially opposite to our measure of EA).23 Compared with the basic measure of Shambaugh (2004), which restricts the “pegged” sample to countries who were pegged for at least two consecutive years, Klein & Shambaugh (2008) pay special attention to the durability of exchange rate regimes, hence include single year pegs as pegs, but does not include discrete devaluations. For foreign exchange intervention, we follow Adler et al. (2021) by calculating the degree of foreign exchange intervention for each country as the ratio of the standard deviation of their FXI proxy to the sum of standard deviations of the proxy and log change in the binomial exchange rates.24 For the measure of dollarization, we mainly use the deposit dollarization constructed by Christiano et al. (2021). For five countries (Australia, Bangladesh, Brazil, Hong Kong, Mainland China, and India) that are not included in their dataset, we follow Adrian & Xie (2020) to construct a U.S. asset share measure based on the Locational Banking Statistics and IMF’s Other Depository Corporation Survey data. For the U.S. trade share measure, we follow Miniane & Rogers (2007) to compute it as the sum of total imports from and exports to U.S., divided by nominal GDP (all in dollars). The trade data is sourced from the Direction of Trade Statistics. We summarize these factors for each country in Table A1. (Panel (a) for the 44 “Full analysis” countries and Panel (b) for the additional 34 “Partial analysis” countries). 20 The reclassification is based the finer classification ranging 1-14 (see Table V in Reinhart & Rogoff (2004)) and is on monthly basis. In particular, For monthly observations lower than 4 (include pre-announced pegs, horizontal bands narrower than or equal to +/-2%, and de facto pegs) while the peg currencies are not USD, we change them to 13 — freely floating. Note that since we base our three-way groupings in the main text on average indicators over the whole sample, this reclassfication is not extreme — e.g. European countries are still allowed to have some fixity (see the positive values in the sixth to eighth columns in Table A1). 21 Less than 5% of the observations fall into the Free falling category, and no countries in our sample have average EA greater than 4, so the Floating group actually only consists of countries with average EA ranging (3,4]. 22 They’re both updated to April 2019 by Jay Shambaugh and can be downloaded from https://iiep.gwu.edu/jay-c- shambaugh/data/. 23 We recode the measures to 0 if the base currency is not US dollar. 24 This is analogy to their equation (1) by replacing their nominal effective exchange rate with binomial exchange rate against USD in our context. Note that Adler et al. (2021) define foreign exchange intervention as “any transaction that changes a central bank’s foreign currency position”. In the numerator, we use their “broad” proxy that encompasses foreign currency operations conducted by the central banks both in spot markets and with derivatives. This proxy adjusts for deviating factors like valuation changes, portfolio reallocation etc., that might lead to bias in coarse proxies primarily based on changes in the stock of international reserves. 34 Table A1: Summary Statistics of Country Factors Table A1.a. Full Analysis Countries FOpen FOpen KS Country EM KC Group Chinn EA PEG FXI DOL USTR (Ha) PEG -Ito Netherlands AE 0.000 Open 1.000 0.983 4.000 0.000 0.000 NA 0.047 0.144 Japan AE 0.002 Open 0.988 0.445 4.000 0.000 0.000 NA 0.017 0.031 Peru EME 0.007 Open 0.983 0.247 2.380 0.083 0.208 0.381 0.617 0.077 U.K. AE 0.011 Open 1.000 1.000 3.440 0.000 0.000 0.000 0.111 0.042 Italy AE 0.025 Open 1.000 0.621 3.847 0.000 0.000 NA 0.016 0.331 Spain AE 0.029 Open 0.986 0.761 4.000 0.000 0.000 NA 0.017 0.331 Ireland AE 0.048 Open 0.998 1.000 3.853 0.000 0.000 NA 0.089 0.123 Norway AE 0.052 Open 0.976 0.811 3.000 0.000 0.042 0.170 0.041 0.029 Canada AE 0.056 Open 1.000 0.730 3.407 0.000 0.042 0.039 0.129 0.324 Denmark AE 0.058 Open 1.000 0.900 3.680 0.000 0.000 0.403 0.064 0.028 Latvia AE 0.064 Open 0.980 0.531 3.287 0.000 0.042 0.351 0.405 0.009 Belgium AE 0.067 Open 0.964 1.000 4.000 0.000 0.000 NA 0.038 0.331 France AE 0.072 Open 1.000 0.906 4.000 0.000 0.000 NA 0.027 0.174 Sweden AE 0.086 Open 0.993 0.917 2.617 0.000 0.000 0.203 0.048 0.036 Uganda EME 0.091 Gate 0.918 0.138 2.841 0.000 0.042 0.199 0.290 0.007 New Zealand AE 0.101 Open 1.000 0.581 3.000 0.000 0.000 0.168 NA 0.051 Israel AE 0.126 Gate 0.828 0.423 3.000 0.000 0.000 0.317 0.200 0.109 Singapore AE 0.140 Open 0.971 1.000 2.700 0.125 0.250 0.685 0.001 0.270 Austria AE 0.146 Open 1.000 0.863 4.000 0.000 0.000 NA 0.018 0.319 Finland AE 0.153 Open 1.000 0.896 4.000 0.000 0.000 NA 0.026 0.173 Portugal AE 0.173 Gate 0.993 0.876 4.000 0.000 0.000 NA 0.068 0.331 Greece AE 0.175 Open 0.851 0.615 4.000 0.000 0.000 0.175 0.150 0.331 Switzerland AE 0.192 Gate 1.000 1.000 2.727 0.208 0.375 0.446 0.408 0.058 Germany AE 0.194 Gate 1.000 0.848 4.000 0.000 0.000 NA 0.013 0.082 Hungary EME 0.237 Gate 0.826 0.706 2.410 0.000 0.042 0.306 0.208 0.032 Australia AE 0.268 Gate 0.808 0.603 4.000 0.000 0.000 0.089 0.065 0.033 Czech Rep. AE 0.291 Gate 0.866 0.412 3.081 0.000 0.042 0.352 0.110 0.025 Romania EME 0.318 Gate 0.703 0.181 2.827 0.000 0.000 0.297 0.365 0.015 Korea AE 0.349 Gate 0.542 0.226 2.930 0.000 0.000 0.231 0.021 0.083 Slovenia AE 0.372 Gate 0.705 0.429 3.187 0.000 0.083 0.245 0.181 0.331 Chile EME 0.390 Gate 0.651 0.511 3.000 0.000 0.000 0.151 0.113 0.085 Cyprus AE 0.451 Gate 0.651 0.971 2.920 0.292 0.292 0.405 0.145 0.012 Türkiye EME 0.456 Gate 0.287 0.143 3.720 0.000 0.000 0.097 0.415 0.022 Argentina EME 0.509 Gate 0.404 0.233 2.333 0.375 0.375 0.100 0.281 0.026 Mexico EME 0.603 Gate 0.680 0.207 3.251 0.000 0.042 0.078 0.089 0.328 Russia EME 0.614 Gate 0.456 0.321 2.982 0.000 0.042 0.196 0.277 0.014 Brazil EME 0.625 Gate 0.291 0.139 3.113 0.000 0.000 0.092 0.055 0.032 Colombia EME 0.621 Gate 0.331 0.166 3.000 0.000 0.000 0.077 NA 0.089 Indonesia EME 0.644 Gate 0.628 0.164 2.537 0.000 0.083 0.133 0.179 0.031 Poland EME 0.724 Gate 0.428 0.239 3.000 0.000 0.000 0.202 0.140 0.014 Malaysia EME 0.813 Wall 0.424 0.531 2.403 0.292 0.333 0.532 0.054 0.112 Bangladesh EME 0.825 Gate 0.174 0.000 1.683 0.625 0.542 0.213 NA 0.031 Algeria EME 0.850 Wall 0.164 0.191 2.001 0.000 0.125 0.469 0.140 0.068 India EME 0.967 Wall 0.164 0.051 2.413 0.083 0.167 0.203 0.042 0.029 Mean — 0.296 — 0.764 0.557 3.195 0.047 0.071 0.243 0.140 0.117 Sd — 0.276 — 0.281 0.327 0.637 0.125 0.129 0.154 0.140 0.119 Notes: (1) KC is the average capital controls, and larger values indicate greater stringency. Countries are displayed in ascending order of KC. (2) Group is defined as in Table 2 of FKRSU. (3) FinOpen (Chinn-Ito) is the de jure financial openness measure by Chinn & Ito (2006). FinOpen (Ha) is the de facto financial openness measure by Ha et al. (2019), winsorized at 99% and standardized to 0-1. (4) EA is the average value of monthly classification from Ilzetzki et al. (2021). A low value indicates fixity, and is reclassified as flexible if the de facto peg is not USD. (5) KSPEG is the annual indicator of “pegs” by Klein & Shambaugh (2008) respectively. (6) FXI is the ratio of the standard deviation of 35 et al. (2021)) to the sum of standard deviations of the a broad foreign exchange interventions proxy (defined in Adler proxy and log change in the binomial exchange rates against USD. (7) DOL is the deposit dollarization measure by Christiano et al. (2021). For Australia, Bangladesh, Brazil, Hong Kong, Mainland China, and India, we use the US asset share measure following Adrian & Xie (2020). (8) USTR is the US trade share measure following Miniane & Rogers (2007). Table A1: Summary Statistics of Country Factors (Continued) Table A1.b. Partial Analysis Countries FOpen FOpen KS Country EM KC Group Chinn EA PEG FXI DOL USTR (Ha) PEG -Ito Zambia EME 0.000 Open 0.923 0.519 3.712 0.000 0.000 0.200 0.384 0.010 Uruguay EME 0.003 Open 0.951 0.415 2.801 0.000 0.083 0.158 0.769 0.018 Hong Kong AE 0.021 Open 1.000 NA 1.000 1.000 1.000 0.926 0.428 0.279 Costa Rica EME 0.031 Open 0.783 0.203 1.473 0.250 0.250 0.321 0.443 0.245 Nicaragua EME 0.044 Open 0.953 0.453 2.000 0.000 0.000 0.721 0.698 0.158 Paraguay EME 0.104 Open 0.578 0.418 2.811 0.042 0.042 0.128 0.510 0.029 Mauritius EME 0.115 Open 0.791 0.722 1.761 0.000 0.042 0.253 0.396 0.045 Qatar EME 0.122 Open 1.000 0.971 1.000 1.000 1.000 1.000 0.284 0.029 Bolivia EME 0.173 Gate 0.596 0.291 1.553 0.542 0.583 0.725 0.614 0.072 Egypt EME 0.174 Open 0.665 0.176 1.757 0.333 0.542 0.164 0.246 0.026 Bulgaria EME 0.179 Gate 0.633 0.469 2.880 0.333 0.333 0.468 0.462 0.018 Nigeria EME 0.225 Gate 0.253 0.087 2.577 0.458 0.542 0.265 0.129 0.069 Kyrgyz Rep. EME 0.266 Gate 0.649 0.416 2.430 0.167 0.250 NA 0.568 0.028 Kenya EME 0.321 Gate 0.672 0.091 2.031 0.208 0.208 0.176 0.151 0.011 Kuwait EME 0.341 Gate 0.736 0.769 1.320 0.833 0.833 0.705 0.112 0.014 Ecuador EME 0.381 Gate 0.629 0.168 1.583 NA NA 1.000 NA 0.185 Iceland AE 0.466 Gate 0.521 0.687 2.767 0.000 0.042 0.184 0.110 0.046 Jamaica EME 0.479 Gate 0.788 0.461 2.000 0.000 0.125 0.445 0.340 0.212 Kazakhstan EME 0.481 Gate 0.166 0.368 2.301 0.250 0.333 0.287 0.477 0.016 Ghana EME 0.511 Gate 0.121 0.152 2.853 0.125 0.208 0.237 0.289 0.025 Lebanon EME 0.612 Gate 0.688 0.819 1.000 1.000 1.000 1.000 0.627 0.031 South Africa EME 0.647 Gate 0.174 0.407 3.805 0.000 0.000 0.056 0.033 0.035 Saudi Arabia EME 0.666 Gate 0.736 0.489 1.000 1.000 1.000 1.000 0.151 0.079 Pakistan EME 0.727 Wall 0.157 0.047 2.000 0.208 0.292 0.211 0.113 0.028 Thailand EME 0.734 Gate 0.305 0.349 2.833 0.083 0.125 0.369 0.018 0.114 Morocco EME 0.761 Wall 0.174 0.226 1.773 0.667 0.708 0.279 0.037 0.035 Myanmar EME 0.783 Gate 0.007 0.256 2.632 0.000 0.083 0.028 NA 0.007 Ukraine EME 0.796 Wall 0.082 0.351 1.913 0.417 0.458 0.193 0.389 0.023 Eswatini EME 0.835 Wall 0.174 0.226 NA 1.000 1.000 NA NA 0.026 Philippines EME 0.855 Wall 0.412 0.238 2.081 0.000 0.125 0.322 0.247 0.114 Angola EME 0.862 Wall 0.072 0.386 3.683 0.125 0.250 0.211 0.584 0.131 Vietnam EME 0.887 Gate 0.285 0.214 2.000 0.750 0.750 0.449 NA 0.129 Tanzania EME 0.922 Wall 0.168 0.179 2.000 0.208 0.292 0.256 0.332 0.007 China EME 0.957 Wall 0.164 0.147 1.000 0.708 0.708 0.479 0.699 0.060 Mean — 0.456 — 0.500 0.369 2.132 0.354 0.400 0.413 0.354 0.069 Sd — 0.318 — 0.315 0.225 0.787 0.365 0.346 0.303 0.217 0.072 A.2 Foreign Variables Data resources and descriptions for the foreign variables for each country are listed in Table A2. 36 Table A2: Data Resource and Description of Foreign Variables Table A2.a. Full Analysis Countries Country EXR Interest Rate Price IP Algeria IFS Lending rate, IFS CPI, IFS IFS∗ Angola IFS Lending rate, IFS HCPI, WB2 NA Argentina IFS Policy rate, BIS CPI, CBA3 Manu., INDEC4 Australia IFS Lending rate, IFS CPI, IFS IFS∗ Austria IFS 10Y Gov bond, FRED CPI, IFS OECD Bangladesh IFS Lending rate, IFS HCPI, WB IFS∗ Belgium IFS 10Y Gov bond, FRED CPI, IFS OECD Brazil IFS Deposit rate, IFS CPI, IFS IFS Canada IFS 10Y Gov bond, FRED CPI, IFS OECD Chile IFS Deposit rate, IFS CPI, IFS OECD Colombia IFS Lending rate, IFS CPI, IFS OECD Cyprus IFS1 Lending rate, Central bank of Cyprus HCPI, WB IFS Czechia IFS1 Lending rate, IFS CPI, IFS IFS Denmark IFS 10Y Gov bond, FRED CPI, IFS OECD Finland IFS 10Y Gov bond, FRED CPI, IFS OECD France IFS 10Y Gov bond, FRED CPI, IFS OECD Germany IFS 10Y Gov bond, FRED CPI, IFS OECD Greece IFS 10Y Gov bond, IFS CPI, IFS OECD Hungary IFS 3M Tbills, FRED CPI, IFS IFS India IFS Lending rate, IFS CPI, IFS OECD Indonesia IFS Lending rate, IFS CPI, IFS Manu., Haver5 Ireland IFS 10Y Gov bond, FRED CPI, IFS OECD Israel IFS Tbills, IFS CPI, IFS OECD Italy IFS 3-12M Tbills, IFS CPI, IFS OECD Japan IFS 10Y Gov bond, FRED CPI, IFS OECD Korea, Rep. IFS Call money rate, IFS CPI, IFS OECD Latvia IFS1 Refinancing rate, Bank of Latvia CPI, IFS IFS Malaysia IFS Lending rate, IFS HCPI, WB Manu., Haver Mexico IFS 90-day Tbills, IFS CPI, IFS Manu., Haver Netherlands IFS 10Y Gov bond, FRED CPI, IFS OECD New Zealand IFS 10Y Gov bond, FRED CPI, IFS IFS∗ Norway IFS 10Y Gov bond, FRED CPI, IFS OECD Peru IFS Policy rate, BIS CPI, IFS Manu., Haver Poland IFS Interbank deposit, IFS CPI, IFS OECD Portugal IFS 10Y Gov bond, FRED CPI, IFS OECD Romania IFS Lending rate, IFS CPI, IFS IFS Russian Federation IFS Interbank loan, IFS CPI, IFS OECD Singapore IFS Lending rate, IFS CPI, IFS Manu., Haver Slovenia IFS1 Interbank deposit, Slovenia central bank CPI, IFS OECD Spain IFS 1Y Tbills, IFS CPI, IFS OECD Sweden IFS 10Y Gov bond, FRED CPI, IFS OECD Switzerland IFS 10Y Gov bond, FRED CPI, IFS IFS∗ Türkiye IFS 3M Deposit, IFS CPI, IFS OECD Uganda IFS 91-day Tbills, IFS CPI, IFS IFS and Manu., Haver6 United Kingdom IFS 10Y Gov bond, FRED CPI, IFS OECD Notes: (1) Exchange rates for Cyprus, Czech Rep., Latvia and Slovenia end at the time when the domestic currency was completely replaced by euro. For each of these countries, we fill the missings by EUR per USD multiplied by the exchange rate to euro at the time when the series discontinues, thus essentially assume that the exchange rate between the country’s currency and euro is fixed post the ending point; (2) We use the headline consumer price index from the Inflation Database constructed by Ha et al. (2021) for countries whose CPI data is missing in IFS; (3) Central Bank of Argentina; (4) Manufacturing Industrial Production Index from https://www.indec.gob.ar/; (5) We use IP index for manufacturer sector from Haver Analytics due to data availability; (6) IP data of Uganda comes from IFS before 2004, and from Haver after that; (7) ∗ stands for data where only quarterly frequency is available so linear interpolation is implemented to get monthly observations. 37 Table A2: Data Resource and Description of Foreign Variables (Continued) Table A2.b. Partial Analysis Countries Country Exchange Rate Interest Rate Consumer Price Index Production Bolivia IFS 10Y Gov bond, FRED CPI, IFS NA Bulgaria IFS Lending rate, IFS CPI, IFS NA China IFS Lending rate, IFS CPI, IFS NA Costa Rica IFS Lending rate, IFS CPI, IFS NA Ecuador IFS Saving rate, IFS CPI, IFS NA Egypt, Arab Rep. IFS Lending rate, IFS CPI, IFS NA Eswatini IFS 91-day Tbills, IFS HCPI, WB NA Ghana IFS Tbills, Haver HCPI, WB NA Hong Kong SAR, China IFS Policy rate, IFS CPI, IFS NA Iceland IFS 10Y Gov bond, FRED CPI, IFS NA Jamaica IFS Lending rate, IFS CPI, IFS NA Kazakhstan IFS Refinancing rate, IFS HCPI, WB NA Kenya IFS 3M Tbills, IFS HCPI, WB NA Kuwait IFS Lending rate, IFS HCPI, WB NA Kyrgyz Rep. IFS 3M Tbills, IFS HCPI, WB NA Lebanon IFS 3M Tbills, IFS NA NA Mauritius IFS Lending rate, IFS CPI, IFS NA Morocco IFS Interbank lending, IFS HCPI, WB NA Myanmar IFS Lending rate, IFS HCPI, WB NA Nicaragua IFS Lending rate, IFS NA NA Nigeria IFS Tbills, IFS CPI, IFS NA Pakistan IFS Call money rate, IFS CPI, IFS NA Paraguay IFS Policy rate, IFS CPI, IFS NA Philippines IFS Lending rate, IFS HCPI, WB NA Qatar IFS Deposit facility, IFS CPI, IFS NA Saudi Arabia IFS Repo, Haver CPI, IFS NA South Africa IFS 91-day Tbills, IFS CPI, IFS NA Tanzania IFS 3M Tbills, IFS CPI, IFS NA Thailand IFS Overnight interbank lending, IFS CPI, IFS NA Ukraine IFS Lending rate, IFS CPI, IFS NA Uruguay IFS Lending rate, IFS CPI, IFS NA Vietnam IFS Lending rate, IFS CPI, IFS NA Zambia IFS Tbills, IFS CPI, IFS NA 38 B Additional Results B.1 Smaller Sample and Median IRFs Figure B2: Responses to U.S. MP Shocks by KC Percentiles, Category 1 and 2 Panel (PC1) 39 Figure B3: Responses to U.S. MP Shocks by KC Percentiles, Category 2 Panel (Median) 40 Figure B4: Responses to U.S. MP Shocks by EA Groupings, Category 2 Panel (Median) 41 Figure B5: Responses to U.S. MP Shocks by FXI Percentiles, Category 2 Panel (Median) B.2 Local Projection Results We estimate local projection coefficients in a panel setting, controlling for all lags of variables in the VAR, and also country-specific trend to control for predetermined dynamics. 42 Figure B6: R* (PC1) Responses by Local Projection: Comparing 3 Shocks Note: Impulse responses are estimated by local projections, controlling for 3 lags of both domestic variables and first principal component of foreign variables in each KC percentile group. We also include a dummy of ZLB period and time trend in the deterministic component. B.3 Sign Restriction Identification and Central Bank Information Shocks We see in the main text that the potency of capital controls relies on the particular monetary policy shocks we are using. One reason for this might be that the “Fed information component” contained in each shock is different. This leads to the question how much of the effects we identified are due to a pure monetary policy shock (JKMP) and how much to a central bank information shock (CBI). In this section, we adopt the sign restriction identification method in Jarociński & Karadi (2020), and compare impulse responses across groups under the two types of shocks.25 As can be seen from Figure B7, there is a significant drop in real output and inflation, and a worsening of domestic credit conditions after a pure monetary policy shock, but it is the opposite for the CB information shock, consistent with Jarociński & Karadi (2020). On the other hand, neither of the two types of shocks induces significantly different responses of foreign interest rates and exchange rates across capital control groups. There is some minor evidence that capital controls insulate countries more from information shocks, though the effect is not strong. 25 Specifically, we use the monetary policy shock and CB information shock extracted by the “median rotation” approach as facilitated by Jarociński & Karadi (2020), and estimate the model under the sign restriction that (i) on “regular days”, the high-frequency movement of stock prices and interest rates within tight window around FOMC announcement move in opposite direction, and (ii) on “information days”, they move in the same direction. 43 Figure B7: HFI with Sign Restrictions: MP vs CBI Shocks Across KC Groupings Notes: (1) This figure shows impulse responses estimated by the high-frequency identification with sign restriction method as in Jarociński & Karadi (2020), in which we include the same variable as Jarociński & Karadi (2020) except adding the PC1 series of foreign interest rates (R*(PC1)) and exchange rates against USD (S(PC1)); (2) A pure monetary shock (JKMP) and a central bank information component (CBI) are separately identified, and we report IRFs are the three country groups every two columns. 44