## Facing the Global Financial Cycle: What Role for Policy Responses? (wpiea2021034-print-pdf) ## Source details **Canonical URL:** [Facing the Global Financial Cycle: What Role for Policy Responses? (wpiea2021034-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021034-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021034-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021034-print-pdf.pdf.json) --- ### Key findings - A common global factor closely tracks changes in global financial conditions and explains cross-country co-movements in (private) capital inflows; this factor is highly correlated with the VIX and the Miranda‑Agrippino and Rey (MAR) principal component. - Large heterogeneity in country sensitivity to the same global factor: - Some countries experience falls in capital inflows when global conditions tighten, while others do not. - Policy levers that reduce sensitivity to the Global Financial Cycle (GFCy): - Higher levels of capital controls and macroprudential policies both dampen sensitivity to the global factor (effect strongest during inflow surge episodes). - Monetary and exchange rate policies can also be deployed to influence exposure (effectiveness depends on country-specific conditions). - Results apply beyond pre-2008 surge episodes and are relevant for disruptions following the Covid-19 pandemic. ### Methodology and econometric framework - Time-varying factor model: CF_t = Λ_t F_t + E_t, with Λ_t (N×1) time-varying factor loadings and F_t a 1×1 common factor. - Factor dynamics: - F_t = ρ_1 F_{t−1} + .... + ρ_q F_{t−q} + u_t, u_t = e^{h_t/2} φ_t, φ_t ∼ N(0,1). - h_t = h_{t−1} + η_t, η_t ∼ N(0,σ^2_η). - Idiosyncratic errors: - E_t = Θ_1 E_{t−1} + .... + Θ_p E_{t−p} + χ_t, χ_t ∼ N(0, Ω_χ) with Ω_χ diagonal. - Time-varying loadings: - Λ_t = I_N Λ_{t−1} + ε_t, ε_t ∼ N(0, Ω_ε) with Ω_ε diagonal. - Estimation: - Gibbs Sampling with lags p = 1, q = 2. - 100,000 draws taken; first 80,000 discarded as burn-in. - Priors following Del Negro and Otrok (2008); convergence checked via visual diagnostics. - Sample: 53 countries (for the factor model using all available-and-complete series). - Fixed-loadings alternative: model reduces to fixed Λ; results broadly consistent on heterogeneity but prior fixed-loadings regressions showed no strong evidence for domestic policy effects. ### Data, sample, and normalization - Capital flows: private gross inflows focus for EMDEs; flows expressed as % share of group GDP; quarterly over 1980Q1:2019Q1 for factor model; annual data from 1996 to 2017 for capital flow regressions. - Aggregate inflows = FDI + Portfolio Investments + Other Investments. - Data source: IMF’s Balance of Payment (BOP) database. - Adjustments: - Series adjusted following Forbes and Warnock (2012). - Capital flows normalized by trend GDP using Hodrick‑Prescott (HP) Filter on annual GDP (Broner et al. (2013) approach). - Series standardized before estimation (Choi (2017)). - Final regression country sample (restricted by explanatory variables): Argentina, Bulgaria, Brazil, Chile, China, Colombia, Costa Rica, Czech Republic, Hungary, Indonesia, India, Israel, Korea, Sri Lanka, Morocco, Mexico, Pakistan, Peru, Philippines, Poland, Romania, Russia, Thailand, Turkey, Ukraine, Uruguay and South Africa. ### Pull variables, policy measures, and indices - Fundamentals included: GDP growth; Current account; Trade integration; Financial integration; CPI inflation; Institutional quality (ICRG Political Risk Index); Gross debt to GDP; IMF Financial Development Index; Currency peg dummy; Gross external assets and liabilities. - Policy variables and constructed indices: - Resistance Index (RI): - EMP_it = %∆e_it + ρ_{i,t} ∆FX_it, RI_it = 1 − %∆e_it / EMP_it, bounded between 0 and 1. - Hypothesis: higher RI (greater resistance to exchange rate appreciation) reduces sensitivity to GFCy. - Sterilization Index: - Regression: ∆DC_it/MB_{i,t−12} = α + β_i ∆NFX_it/MB_{i,t−12} + φ_i ∆GDP_it/GDP_{i,t−12} + u_it. - Interpretation: β_i = −1 full sterilization; β_i = 0 no sterilization; β_i multiplied by −1 in main regressions so an increase corresponds to greater sterilization. - Hypothesis: ambiguous net effect on sensitivity. - Monetary policy: short-term rate. - Fiscal policy: deviation of government expenditures to GDP from backward-looking rolling trend (HP filter parameter = 1000; first 5 years when available used to construct trend). Guidance: debt-to-GDP ratio of 40% suggested as prudential limit for EMDEs. - Capital controls: Fernández et al. (2016) index (indicator per asset category; study uses aggregate and disaggregated inflows/outflows and residency). - Macroprudential policy index: Cerutti et al. (2017) — sum across 12 instruments (0–1 per instrument). - WaMaFi Index (Warranted Macro-Financial Adjustments Index): - Aligns with IMF Institutional View states: S = {Adeq.Res., Overh., Overv., Adeq.Res.&Overv., Overv.&Overh., Adeq.Res.&Overh.} plus residual CFMs state. - Country-year policy response coded binary then weighted (e.g., I_{s,i,t} = 1 × |%∆e_{s,i,t} / ∆cf_{i,t}|; sterilization weight multiplies by (− β̂_{sterilization,i,t})). - Weighted indices mapped to scores 0–5; WaMaFi index ∈ [0,5]; higher = greater relative intensity of warranted adjustments. - Empirical note: average WaMaFi index (normalized to 1) tracks average gross inflows (to GDP), notably pre-2008. - Hypothesis: higher WaMaFi reduces sensitivity to the GFCy. ### Main empirical strategy - Time-varying factor loadings estimated first; dependent variable in policy regressions is λi,t = estimated posterior 50th percentile. - Panel fixed-effect regressions: - λi,t = ai + δEpisodei,t + (β′1 + γ′ × Episodei,t) Xi,t−1 + β′2 Zi,t−1 + Tt + ei,t - Episodei,t dummy = 1 in years when country i experiences globally-driven episodes (surges), classified at 1%, 5%, 10% significance levels. - X = policy variables; Z = fundamentals; interactions Episode × policy included; independent variables lagged one year. - Robustness: include alternative WaMaFi excluding CFMs; disaggregate capital controls by inflow/outflow and residency; test complementarity between CFMs and warranted macroeconomic adjustments. ### Baseline regression results — selected empirical findings - Episodei,t: periods of capital inflows surges associated with higher average sensitivity to the GFCy (coefficient without interactions statistically insignificant). - Policy variables (interactions with Episode): - Capital controls and macroprudential policies are effective at reducing estimated sensitivity during inflow surge episodes: - Macropru. I. × Episode S.: consistently negative and significant (e.g., −0.007 ∗∗∗ in main specifications). - Cap. Controls × Episode S.: negative and significant in baseline (e.g., -0.026 ∗∗). - WaMaFi Index × Episode S.: negative and significant (e.g., −0.033 ∗∗∗), indicating warranted macro-financial adjustments reduce sensitivity during surges. - Fiscal policy: - In high debt countries, positive deviation of public expenditures from trend associated with higher capital flows (increases sensitivity). - In low debt countries, higher deviation associated with lowered sensitivity during inflow surges. - Exchange rate management and sterilization: - Resistance index associated with lowered sensitivity during non-episode times; muted/insignificant effects during surges. - Higher sterilization tends to increase sensitivity (not always statistically significant; interaction terms sensitive to surge classification). - Fundamentals: - Trade integration and financial integration associated with higher sensitivity to the GFCy. - Net foreign assets (NFA) positively associated with sensitivity (interpretation: higher NFA may attract more foreign investment when global conditions ease). ### Disaggregation insights on capital controls - Decomposing capital controls: - Only controls on capital outflows remain statistically significant (especially using most stringent 1% surge classification). - Residency-based results: - Statistically significant effects for controls on non-resident inflows and outflow controls; evidence highlights relevance of controls targeted at non-residents. - Interpretation caveat: measures count presence/absence of restrictions (not intensity); historical imposition of outflow controls could explain lower exposure to future inflow surges. ### Complementarity between CFMs and warranted macroeconomic adjustments - Extended specifications include interactions between WaMaFi (without CFMs) and CFMs (capital controls or macroprudential policies), and their triple interaction with Episode S. - Evidence indicates complementarity, especially for macroprudential policies: - WaMaFi × Macropru. I. × Episode S. coefficients negative and significant (e.g., −0.006 ∗∗ in reported tables), suggesting greater impact of warranted macroeconomic adjustments when CFMs are present. - Results counter a pure-substitution view; CFMs and warranted macroeconomic adjustments act as complements in reducing exposure during surges. ### Robustness checks - WaMaFi excluding CFMs component: - Re-estimations remain highly significant across models; quantitative results in line with baseline, indicating findings not driven exclusively by CFM component. - Fixed-loadings alternative: - Confirms high cross-country heterogeneity in exposure; initial between-estimator analyses (averages) gave limited evidence for domestic policy role—time-varying approach recovers policy relevance during surge episodes. ### Key summary statistics (selected from Table 1) - λ (factor loading): μ 0.57, σ 0.41, Min -0.17, Max 1.38 - Short r.: μ 0.07, σ 0.08, Min 0.00, Max 0.86 - Cap. Controls: μ 0.48, σ 0.32, Min 0.00, Max 1.00 - Macropru. I.: μ 3.10, σ 2.14, Min 0.00, Max 10.00 - Resistance I.: μ 0.55, σ 0.24, Min 0.01, Max 1.00 - Sterilization I.: μ 0.83, σ 0.23, Min -0.01, Max 1.23 - WaMaFi I.: μ 0.41, σ 0.39, Min 0.00, Max 1.00 - WaMaFi I. (Macro): μ 0.36, σ 0.39, Min 0.00, Max 1.00 - CPI: μ 0.06, σ 0.07, Min -0.02, Max 0.55 - Debt: μ 0.45, σ 0.22, Min 0.04, Max 1.52 - Fin. Integr.: μ 1.23, σ 0.83, Min 0.31, Max 5.88 - GDP growth: μ 0.04, σ 0.03, Min -0.15, Max 0.14 ### Bayesian Model Averaging (fixed-loadings between-estimator highlights) - BMA posterior inclusion probabilities (PIP) identify robust correlates of fixed (time-invariant) sensitivities: - CPI inflation: PIP = 0.83 (increases sensitivity). - Fiscal policy in low debt countries (G. Exp. (low Debt)): PIP = 0.70 (increases sensitivity). - Trade integration: PIP = 0.59 (increases sensitivity). - Other variables (short-term rate; sterilization; institutional quality; current account) show moderate PIPs. - Capital controls, macroprudential index, and WaMaFi index: relatively low PIPs in the between-estimator, but signs consistent with policy reducing sensitivity. ### Policy implications and takeaways - During capital inflow surge episodes, domestic policy tools can reduce exposure to the GFCy: - Capital controls and macroprudential regulations are effective in lowering sensitivity to global financial conditions during surges. - Warranted macroeconomic and financial adjustments (WaMaFi) reduce sensitivity; effects robust when CFMs excluded. - CFMs and warranted macroeconomic adjustments are complementary rather than pure substitutes. - Monetary and exchange rate policies matter, but effectiveness depends on country circumstances (e.g., sterilization can have ambiguous effects). - Structural fundamentals (trade/financial integration, CPI inflation, fiscal stance) materially shape sensitivity and should guide policy calibration. - Practical implication: while policymakers cannot stop global liquidity swings, they can deploy a combination of conventional (macroprudential, monetary, fiscal) and less-conventional (targeted capital account measures) tools to dampen transmission and preserve domestic stability. ### Conclusion and research outlook - Gross capital inflows to EMDEs can be described by a common factor mimicking the GFCy; country sensitivities are heterogeneous and time-varying. - Novel WaMaFi index tracks warranted macro-financial adjustments per IMF Institutional View and correlates with inflow dynamics. - Time-varying loadings focused on surge episodes reveal policy effectiveness that fixed‑loadings averages can obscure. - COVID-19 related EME outflows motivate future work to assess which policy actions (including CFMs) helped manage outflows; this study provides a conceptual and empirical framework for such investigation. *Source: wpiea2021034-print-pdf* ### 2.1 Global Financial Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 ### Facing the Global Financial Cycle: What Role for Policy Responses? ### Key findings - A common global factor closely tracks changes in global financial conditions and explains cross-country co-movements in (private) capital inflows; this factor is highly correlated with the VIX and the Miranda-Agrippino and Rey (MAR) principal component. - There is a large degree of heterogeneity in the sensitivity of each country to the same global factor: - Some countries experience falls in capital inflows when global conditions tighten, while others do not. - Higher levels of capital controls and macroprudential policies both dampen the sensitivity to the global factor. - Monetary and exchange rate policies can also be successfully deployed to influence exposure to changes in global financial conditions. - The results extend beyond the surge that preceded the 2008 global financial crisis and are relevant for disruptions following the Covid-19 pandemic. ### Methodology and data - Introduces a time-varying factor model where country sensitivity (factor loadings) to the Global Financial Cycle (GFCy) is allowed to change over time — the study is, to the authors’ knowledge, the first to introduce this time dimension for this question. - Focuses on episodes of surges of gross capital flows, departing from between estimators (averages over the sample period). - Data note: flows are expressed as % share of group GDP; data are quarterly over the period 1980Q1:2019Q1. - Empirical strategy includes: - Extraction of a common factor for private capital inflows and comparison with established measures of the GFCy (VIX, MAR factor). - Regression-based analyses to evaluate the role of domestic fundamentals and explicit policy levers (capital controls, macroprudential policies, sterilization, resistance to exchange market pressures, and a novel macroeconomic/financial adjustments index). ### Contributions relative to literature - Methodological innovation: allows factor loadings to vary over time, enabling investigation of sensitivity during surge episodes. - Policy focus: integrates measures of policy actions (degree of sterilization, resistance to exchange market pressures, novel WaMaFi index) into the GFCy literature. - Empirical findings differ from some prior studies: - Contrasts with Davis et al. (2019) where measures of capital controls were not statistically significant predictors of sensitivity; here, capital controls and macroprudential policies reduce sensitivity. - Extends the literature that uses latent factor models (Cerutti et al. (2019b), Davis et al. (2019), Miranda‑Agrippino and Rey (MAR)) by emphasizing policy role around surge episodes. ### Context from the introduction and literature review - Historical pattern: international purchases of domestic assets (“gross” capital inflows) rose significantly from the mid-1990s to the first half of the 2000s for both EMDEs and AEs; inflows dropped sharply in 2008, rebounded in 2009, and fell again in late 2011. - The GFCy perspective: changes in center countries’ monetary policy affect global banks’ leverage and risk‑taking, which in turn affects cross-border capital flows and domestic credit growth (Bruno and Shin (2013); Rey (2015)). - Debate on push vs. pull factors remains active; relative importance may differ by type of flow (portfolio flows, FDI) and over time. - Prior mixed evidence on the effectiveness of capital flows management measures: - Some studies find limited impact of capital controls on sharp flow episodes, while others find capital controls reduce volume and probability of extreme episodes. - Evidence on macroprudential policies (MPPs) is mixed: some studies find MPPs affect portfolio flows or financial fragility, others find little medium-term effect on flows; some evidence suggests MPPs reduce sensitivity of GDP growth to VIX and flows shocks. ### Policy implications - Capital controls and macroprudential policies can be part of the toolkit to reduce countries’ sensitivity to adverse swings in the Global Financial Cycle. - Domestic monetary and exchange rate policies remain relevant tools to influence exposure, though effectiveness may depend on country-specific conditions (e.g., UIP deviations). - Complementary measures introduced in the paper (sterilization, resistance to exchange market pressures, WaMaFi index) offer additional dimensions for policy evaluation around surge episodes. *Nicoletta Batini and Luigi Durand, January 17, 2021. JEL codes: F30, F32, F38, F42.* ### 3.1 Capital Flow Data ### 3.1 Capital Flow Data ### Data focus - Focus on private gross inflows from a sample of EMDEs. - Rationale: in EMDEs net inflows are largely driven by gross inflows; using gross or net flows is likely to lead to similar estimation results. - Advantage of gross flows: better capture the impact of changes in foreign investors’ flows on domestic macroeconomic and financial stability (see Miranda-Agrippino and Rey (2020b)). ### Definitions and aggregation - Data source: IMF’s Balance of Payment (BOP) database. - Gross flows: - Gross inflows: transactions by non-residents in domestic financial assets. - Gross outflows: transactions by residents in foreign financial assets. - Net inflows = gross inflows − gross outflows. - Aggregate capital inflows computed as the sum of: - Foreign Direct Investments (FDI) flow (involves a controlling stake in a corporation of at least 10%). - Portfolio Investments (PI) flows (bonds and equity holdings flows that do not lead to controlling rights). - Other Investments (OI) flows (residual category recording transactions including loans, deposits and trade credits; broadly associated with banking flows). ### Sample period, adjustments, and normalization - Estimation period: annual data from 1996 to 2017. - Country sample selection: - As exhaustive as possible, excluding series with gaps (require all three series to be available) or where capital flows are consistently unavailable through the sample period. - Data adjustments: - All data series adjusted following Forbes and Warnock (2012). - Capital flows series normalized by trend GDP, where trend GDP is extracted using an Hodrick-Prescott (HP) Filter of annual GDP series (as suggested by Broner et al. (2013)). - Series then standardized before model estimation (as recommended by Choi (2017)) to address non stationarity in factor models. - Methodological note: - HP filter used for detrending for consistency with previous literature; alternative methodologies such as the “Hamilton filter” (Hamilton (2018)) are noted as suitable and having benefits (e.g., avoiding potential issues associated with the HP filter, including introduction of spurious dynamics). ### Final regression sample (countries) - The final sample, restricted by availability of explanatory variables, is composed of: - Argentina, Bulgaria, Brazil, Chile, China, Colombia, Costa Rica, Czech Republic, Hungary, Indonesia, India, Israel, Korea, Sri Lanka, Morocco, Mexico, Pakistan, Peru, Philippines, Poland, Romania, Russia, Thailand, Turkey, Ukraine, Uruguay and South Africa. *Source: wpiea2021034-print-pdf - 3.1 Capital Flow Data* ### 3.2 Pull Variables ### 3.2 Pull Variables ### 3.2.1 Fundamentals - Group of included fundamentals: - GDP growth - Current account - Trade integration (sum of exports and imports over GDP) - Financial integration (sum of foreign assets and liabilities over GDP) - CPI inflation - Institutional quality: International Country Risk Guide (ICRG) Political Risk Index (Higher values indicate greater institutional quality) - Gross debt to GDP (measure of financial risk) - IMF Financial Development Index (Sahay et al. (2015)) — summarizes depth, access, and efficiency of financial markets and institutions - Currency peg dummy (Shambaugh (2004); = 1 if peg, 0 otherwise) - Gross external assets and liabilities (following Davis et al. (2019)) - Rationale and key arguments: - GDP growth: neo-classical model predicts capital flows to higher-growth countries. - Current account: driver of net flows; in EMDEs current account often driven by gross inflows, making it relevant here. - CPI inflation: captures price stability; price stability matters for foreign capital inflows by limiting currency risk premium and potentially promoting financial stability (Borio and Lowe (2002)); relationship with financial stability remains disputed (see Blot et al. (2015)). - Institutional quality: institutional investors are sensitive to domestic institutional quality (Ahlquist (2006)); summary index keeps model parsimonious given broad underlying concepts. - Financial development: prior literature finds significant relationship between capital flows and financial development (Nispi Landi and Schiavone (2018)); greater financial development can reduce sensitivity of domestic financial conditions to the GFCy (Arregui et al. (2018)). - Currency peg and gross external positions are included for their potential roles in capital flows sensitivity to the GFCy. ### 3.2.2 Policy Variables - Overview: second group focuses on policy actions. The paper builds on existing indexes and constructs new ones. - Resistance Index (RI) - Define Exchange Market Pressure (EMP) index: - EMP_it = %∆e_it + ρ_{i,t} ∆FX_it - %∆e_it is quarter-on-quarter percentage change in the exchange rate, expressed so that an increase corresponds to an appreciation of the domestic currency (with respect to the USD). - ∆FX_it is quarter-on-quarter change in foreign reserves, measured in billions of USD. - ρ_{i,t} is a conversion factor measuring the change in the exchange rate associated with a $1 billion USD intervention (monthly conversion factor estimated by Patnaik et al. (2017)). - The EMP Index falls when the exchange rate is either depreciating or the domestic central bank is reducing international reserves. - Patnaik et al. (2017) conversion factor reflects currency market liquidity: more developed markets have smaller ρ_{i,t}. - Resistance Index: - RI_it = 1 − %∆e_it / EMP_it - Index bounded between 0 and 1 (as in Cardarelli et al. (2010)). - Interpretation: larger RI implies greater resistance (exchange rate prevented from reacting as it would absent reserves intervention). - Hypothesis 1: - Countries that resist nominal exchange rate appreciation (higher RI) should be less sensitive to the GFCy, because capital inflows associate with real exchange rate appreciation (see Combes et al. (2012)), and resistance may reduce “one-way” bet positions from intermediaries (IMF (2011); Bruno and Shin (2014)). - Sterilization Index - Regression used to measure degree of sterilization (Aizenman and Glick (2009)): - ∆DC_it/MB_{i,t−12} = α + β_i ∆NFX_it/MB_{i,t−12} + φ_i ∆GDP_it/GDP_{i,t−12} + u_it - NFX denotes net foreign assets expressed in domestic currency. - DC_it ≡ MB_it − NFX_it - MB: Monetary Base (currency issued + deposits of commercial banks with the central bank). - Interpretation of β_i: - β_i = −1 implies full monetary sterilization (monetary base unchanged). - β_i = 0 implies no sterilization. - β_i < −1 implies monetary tightening (e.g., to counter inflationary pressures). - β_i > 0 implies expansionary monetary policy. - Regression run using monthly data (quarterly for GDP growth) in a rolling fashion; collected β_i coefficients included as regressor (in main regressions β_i multiplied by −1 so that an increase corresponds to greater sterilization). - Hypothesis 2: - Ambiguous effects: sterilization can prevent domestic interest rates from falling (which may beget further inflows and increase sensitivity to common factor) but also reduce credit growth and real exchange appreciation (which may reduce sensitivity). - Monetary and Fiscal Policy measures - Monetary policy: short-term rate. - Fiscal policy: deviation of government expenditures to GDP with respect to its backward-looking rolling trend (HP filter parameter = 1000; first 5 years when available used to construct trend). - Notes on fiscal measure: - Simple and widely available but may not always capture fiscal stance (e.g., persistent expansionary policy affects trend). A more precise alternative would be deviation from debt-stabilizing primary balance but requires more data. - Hypothesis 3: - Fiscal consolidation limits capital inflows’ absorption, limiting exchange rate appreciation, but can improve creditworthiness and raise domestic returns, potentially increasing sensitivity to global financial conditions. Net effect ambiguous and may depend on level of gross debt to GDP. Guidance: a debt-to-GDP ratio of 40% suggested as prudential limit for EMDEs to maintain fiscal sustainability (Abbas et al. (2019)). - Capital Controls & Macroprudential Policies - Capital account restrictions: index from Fernández et al. (2016) — indicator variable (0 or 1) per asset category and time; study uses aggregate measure and distinguishes inflows-only and outflows-only sub-indices. - Macroprudential policy index: Cerutti et al. (2017) — based on 12 instruments (e.g., capital buffer, leverage ratio, loan-to-value); each instrument coded 0-1, index is sum across instruments. - Caveat: neither index captures intensity of regulation (only presence/absence), ensuring objectivity but missing intensive margin. - Hypothesis 4: - Higher capital account and macroprudential restrictions, if effective, should reduce exposure to the GFCy; however, challenges exist including endogeneity and inability of many indexes to capture intensity changes (exceptions like Pasricha et al. (2018) have limited samples). - Warranted Macro-Financial Adjustments Index (WaMaFi Index) - Purpose: reflect prescriptions of the IMF Institutional View (IV) on capital flows management (Arora et al. (2013)); capture whether countries implement warranted macroeconomic and financial policy adjustments as first line of defense against capital flow disruptions. - Conceptual framework: - Venn Diagram states: “Economy Overheating”, “Reserves Adequate”, “Exchange Rate Overvalued”. - Six possible IV states: - S = {Adeq.Res., Overh., Overv., Adeq.Res.&Overv., Overv.&Overh., Adeq.Res.&Overh.} - Plus a residual state that allows use of CFMs. - Construction: - For each country-year and IV state, code policy response as binary (0-1) depending on whether observed policy is in accordance with IV for that state. - Binary variables weighted to reflect intensity of policy action: - Example exchange rate weight: - I_{s,i,t} = 1 × |%∆e_{s,i,t} / ∆cf_{i,t}| - where cf_{i,t} denotes capital flows (to GDP). - Sterilization when reserves accumulate enters multiplicatively: - I_{s,i,t} = 1 × |%∆FX_{s,i,t} / ∆cf_{i,t}| × (− β̂_{sterilization,i,t}) - where β_{sterilization,i,t} is the estimated sterilization coefficient; larger weight for more aggressive sterilization. - For each of the six states (plus residual CFM state), distribution of weighted indices across sample mapped into sample quantiles; each weighted index mapped to a score between 0 and 5. - WaMaFi index for each country-year takes value between 0 and 5; 0 when country does not follow IV; higher values indicate greater relative intensity of warranted adjustments compared to other countries. - If two codifications compatible with IV exist for a given state (e.g., exchange rate change vs foreign reserves change), include only the highest computed score. - Empirical note: - Figure 3 (described) shows average WaMaFi index (normalized to 1) tracks average gross inflows (to GDP) well, notably in run-up to 2008 GFC; alternative WaMaFi excluding CFM component closely follows baseline. - Hypothesis 5: - Higher WaMaFi score (positive and larger) should reduce countries’ sensitivity to the GFCy, because a score of zero reflects no warranted macroeconomic adjustments aligned with IV while positive scores reflect warranted adjustments of monetary and exchange rate policies (and higher values indicate greater relative intensity). - Additional methodological notes and definitions - Net foreign assets (NFX): central bank’s claims on nonresidents less liabilities to nonresidents; in most countries liability component negligible so NFX equated with net official international reserves. - Monetary Base (MB): constructed from central bank balance sheet summing currency issued (held in banks and outside banks) plus deposits of commercial banks with the central bank. - EMP concept origins: Girton and Roper (1977); variants by Eichengreen et al. (1996), Kaminsky et al. (1998), Sachs et al. (1996), Weymark (1995). Patnaik et al. (2017) EMP index allows cross-country comparisons and captures periods with fixed/modest exchange rate changes. - Regression timing: sterilization regression run in rolling fashion, one month added at a time; country-specific β_i collected for main regressions (multiplied by −1 for interpretability). *Source: wpiea2021034-print-pdf — Section 3.2 Pull Variables* ### 4.1 Econometric Framework ### 4.1 Econometric Framework ### Model specification - Capital flows modeled as: - CF_t = Λ_t F_t + E_t - CF_t is an N×1 vector stacking each country’s capital flows series, cf_{i,t}. - Λ_t is an N×1 vector stacking each country’s factor loadings, λ_{i,t}. - F_t is a 1×1 common factor. - E_t is an N×1 vector of idiosyncratic elements. - Interpretation: - F_t describes the systematic source of variation among capital flow series. - λ_{i,t} determines the strength of the association between country i flows and the common component (higher value implies stronger relationship). ### Factor dynamics and volatility - Factor evolution: - F_t = ρ_1 F_{t−1} + .... + ρ_q F_{t−q} + u_t - u_t = e^{h_t/2} φ_t and φ_t ∼ N(0,1). - Log-volatility: - h_t = h_{t−1} + η_t with η_t ∼ N(0,σ^2_η). ### Idiosyncratic errors - Error terms follow an AR(p) process: - E_t = Θ_1 E_{t−1} + .... + Θ_p E_{t−p} + χ_t - Θ_i are diagonal matrices. - χ_t ∼ N(0_{N×1}, Ω_χ) with - Ω_χ = diag(σ^2_{1,χ}, σ^2_{2,χ}, ..., σ^2_{N,χ}). ### Time-varying factor loadings - Loadings follow a driftless random walk: - Λ_t = I_N Λ_{t−1} + ε_t - ε_t ∼ N(0_{N×1}, Ω_ε) with - Ω_ε = diag(σ^2_{1,ε}, σ^2_{2,ε}, ..., σ^2_{N,ε}). - Disturbances χ_t and ε_t are assumed independent of each other. ### Estimation method and implementation details - Gibbs Sampling: - lags p = 1, q = 2 - 100,000 draws taken; first 80,000 discarded as burn-in. - Priors: fairly loose priors following Del Negro and Otrok (2008) (see Appendix for details). - Convergence checked via visual diagnostics including recursive mean plots of non-time varying parameters (e.g., AR coefficients). - Sample: - Use all available-and-complete capital inflows series; sample of 53 countries in total. ### Fixed loadings alternative and comparison - By fixing Λ_t over time, the model reduces to a fixed-loadings factor model (similar to Neely and Rapach (2011); Cerutti et al. (2019a)). - Main findings from fixed-loadings analysis (broadly consistent with preceding literature): - i) there is a high degree of cross-country heterogeneity in terms of exposure to the estimated global factor; - ii) there is a high degree of cross-country heterogeneity in terms of the share of variance that can be accounted for by the global factor; - iii) when performing a regression of the estimated factor loadings on the sample averages of the variables of interest there is no strong evidence in favor of a role for domestic policies. *Italic: Source: wpiea2021034-print-pdf - 4.1 Econometric Framework* ### 6.1 Baseline Regression Results ### 6.1 Baseline Regression Results ### Econometric framework - Estimate panel fixed effect regressions of the type: λi,t = ai + δEpisodei,t + (β′1 + γ′ × Episodei,t) Xi,t−1 + β′2 Zi,t−1 + Tt + ei,t - Episodei,t is a dummy equal to 1 in years when country i experiences globally-driven episodes, and 0 otherwise. - Include country fixed effects (ai) and time fixed effects (Tt). - X contains “policy variables” of interest; Z contains “fundamental” variables. - Interaction term between Episodei,t and policy variables is included; primary interest is vector coefficient γ. - All episodes classified as global at the 1%, 5% and 10% significance levels are included. - Dependent variable λi,t corresponds to the estimated posterior 50th percentile from the time varying factor loading model. - Independent variables are lagged one period (one year) to reduce endogeneity. ### Main empirical findings (Table 7) - Episodei,t: periods of capital inflows surges are associated with a higher average sensitivity to the GFCy, ceteris paribus (coefficient, absent interactions, is statistically insignificant). - Policy variables: - Use of both capital controls and macroprudential policies are effective at reducing estimated sensitivity, but only during episodes of capital inflows surges. - This result contrasts with Davis et al. (2019) and supports the view that international financial integration matters for domestic autonomy and resilience to global forces (credit to Rey (2015)). - CFMs emerge as an effective tool to influence risk spillovers related to U.S. monetary policy shocks; limits credit growth and leverage during booms reduce transmission of the GFCy. - Figure 10 shows the association is robust and not driven by outlier observations. - WaMaFi Index: - During periods of capital inflows surges, a higher score on the WaMaFi Index is significantly associated with a reduction in sensitivity to the GFCy. - Fiscal policy: - In high debt countries, positive deviation of public expenditures from trend is economically significantly associated with higher capital flows (positive effect on sensitivity). - In low debt countries, higher deviation of public expenditures from trend is associated with lowered sensitivity to the GFCy during inflows surges. - Exchange rate management and sterilization: - Resistance to exchange rate appreciation is associated with lowered sensitivity during non-episode times, but effect is muted during surges and coefficients are insignificant. - Higher sterilization tends to increase capital inflows sensitivity, though not statistically significant; coefficients with interaction terms flip sign at 5% and 10% surge classifications. - Fundamentals: - Both trade and financial integration are associated with higher sensitivity to the GFCy. - Positive association with net foreign assets; interpretation: higher net foreign assets may imply higher net foreign asset returns, a stabilizing force against global income shocks. ### Notes - Explanatory variables included with a lag; interpretation: countries with higher levels of capital controls and macroprudential policies display less sensitivity to the GFCy during a capital inflow surge. --- ### 6.2 Disaggregation of Capital Controls Measures ### Motivation and approach - Decompose aggregate capital controls measure into: - i) sub-indices targeting inflows and outflows - ii) sub-indices targeting inflows and outflows based on residency - Re-estimate equation (5) with these disaggregated components. - Caveat: sample size adjusted depending on data availability on disaggregated components. ### Key results (Table 8, Tables 9 and 10) - Total inflows vs total outflows controls: - Only controls on capital outflows remain statistically significant, and only when using the most stringent (1%) classification of globally-driven surges. - Residency-based disaggregation: - Results confirm statistical significance of capital controls on outflows. - Highlight relevance of capital controls on non-residents inflows. - Comparison to other literature: - Gupta and Masetti (2018) find, on average, countries do not change pre-existing levels of capital control restrictions during surge episodes except loosening controls on resident outflows; our measure differs (counts restrictions in place vs tightening/loosening). - Authors note alternative interpretations: results could capture effects of previously imposed outflow controls (e.g., imposed during past outflows episodes) making countries less exposed to future inflow surges. --- ### 6.3 WaMaFi Index without CFMs component ### Purpose - Re-estimate regression removing the CFMs component from the WaMaFi Index to emphasize conventional macroeconomic and financial adjustments and test robustness. ### Findings (Table 11) - Results are highly significant across three models. - Quantitatively in line with previous findings, indicating results are not exclusively driven by the use of CFMs. --- ### 6.4 Complementarity of CFMs and Warranted Macroeconomic Adjustments ### Extended specification - Modify equation (5) to include interaction between WaMaFi Index (without CFMs) and CFMs during surges: λi,t = ai + δEpisodei,t + (β′1 + γ′ × Episodei,t) ˜Xi,t−1 + α1 CFM i,t−1 + α2 WaMaFi i,t−1 + (θ1 + ζ1 × Episodei,t) CFM i,t−1 × WaMaFi i,t−1 + β′2 Zi,t−1 + Tt + ei,t - ˜X excludes CFM and WaMaFi components; CFM is either capital controls or macroprudential regulations. - Interest in ζ1: ζ1 < 0 would suggest impact of warranted macroeconomic adjustments on sensitivity to the GFCy during surges increases with degree of CFMs. ### Findings (Tables 12–14) - Evidence suggests complementarity between macroeconomic policy levers and CFMs, especially for macroprudential policies. - Results counter the concern that CFMs are used as substitutes for warranted macroeconomic adjustment; instead CFMs and other policy levers complement each other in reducing sensitivity to the GFCy. --- ### 7 Conclusion ### Summary of contributions - Document that gross capital inflows in emerging and developing countries can be described by a common factor that mimics measures of the Global Financial Cycle (GFCy). - Estimate country-specific sensitivities to this global factor and find significant heterogeneity. - Develop a novel index of “warranted” macroeconomic adjustments (WaMaFi) aligned with the IMF Institutional View of CFMs, tracking evolution of gross capital inflows to emerging markets. - Introduce a model allowing time variation in country sensitivities, focusing on globally-driven capital inflow surges. ### Main policy-relevant findings - Countries with higher capital controls and, more significantly, higher macroprudential regulations and those that enact warranted macroeconomic and financial adjustments are less exposed to the GFCy during inflow surges. - Findings contrast with related research that does not find significant roles for capital controls and macroprudential regulations. - Implication: domestic policymakers, though unable to stop swings in global liquidity, can use conventional and less-conventional tools to counteract exposure and help preserve domestic economic and financial stability. ### Notes on COVID-19 context and future research - Recent EME outflows during the Covid-19 pandemic raise related questions on domestic policy roles; central banks and fiscal policy played major roles according to recent research. - Results here provide a conceptual framework to explore which policy actions helped manage EME outflows, including capital controls; further investigation is left for future research. *Source: wpiea2021034-print-pdf - 6.1 Baseline Regression Results* ### 8.1 Tables ### 8.1 Tables ### Summary statistics (Table 1) - Sample statistics used for baseline regression reported in Table 7. - Means (μ), standard deviations (σ), minimum (Min) and maximum (Max): - λ (fact. loading): μ 0.57, σ 0.41, Min -0.17, Max 1.38 - Short r.: μ 0.07, σ 0.08, Min 0.00, Max 0.86 - Cap. Controls: μ 0.48, σ 0.32, Min 0.00, Max 1.00 - Macropru. I.: μ 3.10, σ 2.14, Min 0.00, Max 10.00 - Resistance I.: μ 0.55, σ 0.24, Min 0.01, Max 1.00 - Sterilization I.: μ 0.83, σ 0.23, Min -0.01, Max 1.23 - G. Exp. (low Debt): μ -0.02, σ 0.07, Min -0.47, Max 0.25 - G. Exp. (high Debt): μ -0.01, σ 0.07, Min -0.40, Max 0.26 - WaMaFi I.: μ 0.41, σ 0.39, Min 0.00, Max 1.00 - WaMaFi I. (Macro): μ 0.36, σ 0.39, Min 0.00, Max 1.00 - CPI: μ 0.06, σ 0.07, Min -0.02, Max 0.55 - Peg: μ 0.20, σ 0.40, Min 0.00, Max 1.00 - Inst. Q.: μ 0.67, σ 0.08, Min 0.44, Max 0.84 - Debt: μ 0.45, σ 0.22, Min 0.04, Max 1.52 - Fin. Devlpt. I.: μ 0.41, σ 0.15, Min 0.12, Max 0.86 - NFA: μ -0.48, σ 0.27, Min -1.54, Max 0.04 - CA: μ 0.00, σ 0.06, Min -0.17, Max 0.21 - Trade Integr.: μ 0.70, σ 0.35, Min 0.19, Max 2.08 - Fin. Integr.: μ 1.23, σ 0.83, Min 0.31, Max 5.88 - GDP growth: μ 0.04, σ 0.03, Min -0.15, Max 0.14 - Notes: CA = Current Account; G. Exp. = Government Expenditures; NFA = Net Foreign Assets (Other assets and liabilities). ### Bayesian Model Averaging for Gross Inflows (Table 2) - Sample: Observations 27. - Reported per variable: Coef., t−stat, Post-Inclusion Probability (PIP). - Selected entries (Coef. / t−stat / PIP): - Cap. Controls: -0.02 / -0.20 / 0.10 - Short Rate: -1.88 / -0.67 / 0.42 - Macropru. I.: 0.00 / 0.08 / 0.08 - Resistance I.: 0.12 / 0.40 / 0.20 - Sterilization: 0.16 / 0.55 / 0.31 - G. Exp. (low Debt): 5.33 / 1.31 / 0.70 - G. Exp. (high Debt): -0.32 / -0.24 / 0.14 - Peg: 0.08 / 0.38 / 0.19 - WaMaFi I.: -0.06 / -0.08 / 0.08 - CPI: 6.99 / 1.46 / 0.83 - Inst. Q.: 0.20 / 0.49 / 0.26 - Debt: -0.06 / -0.24 / 0.14 - Fin. Devlpt. I.: 0.10 / 0.36 / 0.17 - NFA: 0.14 / 0.37 / 0.18 - CA: 0.63 / 0.42 / 0.22 - Trade Int.: 0.27 / 1.02 / 0.59 - Fin. Int.: 0.03 / 0.39 / 0.19 - GDP Growth: 0.22 / 0.14 / 0.10 - Notes: Variables with the highest PIPs are considered the most robust correlates. All independent variables are expressed as sample means. ### Codification of capital inflows surges (Table 3) - Binary codification (1 = surge detected, 0 = no surge) for each country-year in sample. - Countries and binary year strings provided for 1996–2008 and 2009–2018 panels (examples): - ARG (1996–2008): 0100000000110 - BGR (1996–2008): 0101000110110 - BRA (1996–2008): 0000000000110 - CHL (1996–2008): 0001000000011 - CHN (1996–2008): 0100000011000 - ... (table continues for listed countries) - Notes: 1 indicates a capital inflow surge as defined in the paper; 0 indicates no surge detected. ### Surges in Private Inflows: CLogLog regression analysis (Table 4) - Dependent variable: Binary indicator = 1 whenever there is a surge in gross inflows. - Observations: 422 (both columns). - Selected coefficient estimates and significance (Episode S. specifications): - GFCy: 0.469 ∗∗∗ (t = 3.19) in Column (1); 0.594 ∗∗ (t = 2.24) in Column (2) - Commodity I.: 0.282 (1.55); 0.365 ∗ (1.88) - Inst. Q.: 13.93 ∗∗∗ (3.76); 17.31 ∗∗∗ (3.46) - NFA: 3.484 ∗∗∗ (2.75); 3.308 ∗∗ (2.50) - CPI: -8.003 ∗∗ (-2.09); -6.971 (-1.63) - Peg: 0.959 ∗∗ (2.16); 1.005 ∗∗ (1.99) - WaMaFi I.: 0.575 (1.30); 0.635 (1.24) - Interaction highlights: - GFCy×Cap. Controls: -2.704 ∗∗ (t = -2.01) - GFCy×G. Exp. (low Debt): -6.163 ∗∗ (t = -2.27) - Notes: - t-statistics in parentheses. - ∗ p <0.10, ∗∗ p <0.05, ∗∗∗ p <0.01. - Robust standard errors and fixed effects by country. All domestic independent variables are lagged by one year. Variables standardized. ### Summary statistics by Episode S. states (Table 5) - Means (μ) and standard deviations (σ) for Episode S. = 0 and Episode S. = 1 across Low, Medium, High categories. - Select entries (Episode S. = 0, μ): - Low: λ 0.161; Cap. C. 0.672; WaMaFi I. 0.294; Macropru I. 4.068; Short r. 0.067; G. Exp. (L. D.) 0.001; G. Exp. (H. D.) -0.009; Resist. I. 0.522; Stz. 0.884 - Medium: λ 0.491; Cap. C. 0.307; WaMaFi I. 0.347; Macropru I. 2.440; Short r. 0.061; G. Exp. (L. D.) -0.023; G. Exp. (H. D.) -0.007; Resist. I. 0.542; Stz. 0.803 - High: λ 1.052; Cap. C. 0.440; WaMaFi I. 0.431; Macropru I. 2.819; Short r. 0.091; G. Exp. (L. D.) -0.006; G. Exp. (H. D.) -0.003; Resist. I. 0.581; Stz. 0.824 - Select entries (Episode S. = 1, μ): - Low: λ 0.152; Cap. C. 0.642; WaMaFi I. 0.282; Macropru I. 4.727; Short r. 0.060; G. Exp. (L. D.) -0.031; G. Exp. (H. D.) 0.005; Resist. I. 0.561; Stz. 0.890 - Medium: λ 0.543; Cap. C. 0.593; WaMaFi I. 0.295; Macropru I. 2.143; Short r. 0.059; G. Exp. (L. D.) -0.065; G. Exp. (H. D.) -0.018; Resist. I. 0.588; Stz. 0.810 - High: λ 1.097; Cap. C. 0.306; WaMaFi I. 0.629; Macropru I. 2.143; Short r. 0.073; G. Exp. (L. D.) -0.001; G. Exp. (H. D.) -0.058; Resist. I. 0.575; Stz. 0.759 - Notes: WaMaFi I. excludes CFMs. Surges selected at a 5% significance level. ### Sensitivity to GFCy during surges in private inflows (Table 6 and continuation) - Panel regression, dependent variable: Estimated 50th percentile of the time-varying factor loadings. - Observations: 423 (columns vary by inclusion threshold of Episode S.). - Selected coefficients and interactions: - Episode S.: 0.016 / 0.019 / 0.022 / 0.029 ∗ (t-statistics reported) - Episode S. × Short r.: 0.160 / 0.237 ∗∗∗ / 0.225 ∗∗∗ / 0.128 ∗∗∗ - Cap. Controls × Episode S.: -0.026 ∗∗ (column 1) and other columns reported - Macropru. I. × Episode S.: -0.007 ∗∗∗ / -0.005 ∗∗∗ / -0.005 ∗∗∗ / -0.003 ∗∗ - Episode S. × G. Exp. (low Debt): -0.279 ∗∗∗ / -0.119 ∗ / -0.124 ∗ / -0.132 ∗ - G. Exp. (high Debt): 0.147 ∗∗ / 0.144 ∗∗ / 0.140 ∗∗ / 0.150 ∗∗ - WaMaFi I. × Episode S.: -0.033 ∗∗∗ / -0.022 ∗∗ / -0.019 ∗∗ / -0.009 - NFA: 0.060 ∗∗ / 0.062 ∗∗ / 0.062 ∗∗ / 0.059 ∗∗ - Trade Integr.: 0.093 ∗ / 0.098 ∗ / 0.098 ∗ / 0.096 ∗ - Fin. Integr.: 0.022 ∗∗ / 0.022 ∗∗ / 0.023 ∗∗ / 0.023 ∗∗ - Model fit: R2a = 0.313 / 0.320 / 0.321 / 0.314 across columns. - Notes: Robust standard errors, fixed effects by country and time, domestic variables lagged by one year. ### Distinguishing inflows and outflows controls (Tables 7–9) - Table 7: Inflows vs outflows controls (overall): - Cap. C. (Outf.): coefficients approx -0.022 to -0.023; Cap. C. (Outf.) × Episode S.: -0.033 ∗∗ in column 1. - Cap. C. (Inf.): coefficients approx -0.019 to -0.017; Cap. C. (Inf.) × Episode S.: non-significant. - Observations: 423; R2a ~ 0.304–0.319 depending on specification. - Table 8: Non-residents inflows and outflows controls: - Cap. C. (Outf. N.Res.): -0.001 to -0.005; Cap. C. (Outf. N.Res.) × Episode S.: -0.024 ∗ (column 1). - Cap. C. (Inf. N.Res.): -0.021 to -0.023; Cap. C. (Inf. N.Res.) × Episode S.: -0.020 ∗ / -0.022 ∗∗ / -0.019 ∗ across columns. - Observations: 410–413; R2a ~ 0.319–0.332. - Table 9: Residents inflows and outflows controls: - Cap. C. (Outf. Res.): -0.015; Cap. C. (Outf. Res.) × Episode S.: -0.032 ∗∗ (column 1). - Cap. C. (Inf. Res.): -0.019 to -0.020; Cap. C. (Inf. Res.) × Episode S.: mostly non-significant. - Observations: 417–420; R2a ~ 0.305–0.330. - Notes: Dependent variable: Estimated 50th percentile of time-varying factor loadings. Robust standard errors, fixed effects, domestic variables lagged one year. ### WaMaFi Index and interactions (Tables 10–13) - Table 10: WaMaFi I. (excluding CFMs component) - WaMaFi I.: 0.008 / 0.006 / 0.006 (t-statistics ~1.465, 1.140, 1.052) - WaMaFi I. × Episode S.: -0.031 ∗∗∗; -0.021 ∗∗; -0.017 ∗∗ (t-statistics: -2.840, -2.241, -2.172) - Observations 423; R2a 0.310–0.318. - Table 11: Interaction WaMaFi × Macropru I. - Macropru. I.: 0.000 (non-significant) - WaMaFi I.: -0.004 (non-significant) - WaMaFi I. × Macropru. I.: 0.004 ∗ / 0.004 ∗∗ / 0.004 ∗∗ (t ~ 2.010, 2.087, 2.173) - WaMaFi I. × Macropru. I. × Episode S.: -0.006 ∗∗; -0.007 ∗∗∗; -0.006 ∗∗∗ (t-statistics reported) - Observations 423; R2a 0.300–0.316. - Table 12: Interaction WaMaFi I. and Non-Residents controls - Cap. C. (Outf. N.Res.) × Episode S. interaction with WaMaFi I.: -0.041 ∗∗∗ / -0.018 ∗ / -0.015 (reported) - WaMaFi I. × Cap. C. (Inf. N.Res.) × Episode S.: -0.049 ∗∗∗ / -0.032 ∗∗ / -0.027 ∗ - Observations 410–413; R2a 0.321–0.332. - Table 13: Interaction WaMaFi I. and Residents controls - WaMaFi I. × Cap. C. (Outf. Res.) × Episode S.: -0.043 ∗∗∗; -0.027 ∗∗; -0.024 ∗∗ - WaMaFi I. × Cap. C. (Inf. Res.) × Episode S.: -0.038 ∗∗∗; -0.019; -0.017 - Observations 417–420; R2a 0.305–0.329. - Notes: All regressions use WaMaFi I. without CFMs component when specified. Robust standard errors, fixed effects, domestic variables lagged one year. ### Figures: Capital flows (Figures 11 and 11a) - Figure 11 and Figure 11a plot standardized series of capital flows to GDP (solid lines) and the 50th percentile of the estimated common factor (dashed lines) for listed countries over 2000–2010 (and beyond for continuation). - Countries plotted include: ISR, ZAF, ARG, BRA, CHL, COL, BGR, PER, IND, IDN, KOR, PHL, THA, RUS, HUN, CHN, CRI, CZE, LKA, MAR, MEX, PAK, POL, ROU, TUR, UKR, URY. - Notes: Solid lines represent the (standardized) series of capital flows to GDP and dashed lines represent the 50th percentile of the estimated common factor. ### WaMaFi Index: Data details (Section 8.3) - States used to build WaMaFi Index (six possible states, excluding residual CFMs state): - S = {Adeq.Res., Overh., Overv., Adeq.Res.&Overv., Overv.&Overh., Adeq.Res.&Overh.} - Measurement rules: - Adequate Reserves: - Use Reserves over ARA EM metric (as described in IMF (2016)). - ARA EM metric components: (i) export income; (ii) broad money; (iii) short-term debt; (iv) other liabilities. - Reserves in the range of 100-150 percent of the composite metric considered adequate for precautionary purposes. - If ARA metrics not available, use import cover ratio with foreign reserves covering three months of total imports (annual imports divided by four). - Overheating: - If GDP growth exceeds its potential as computed by IMF World Economic Outlook (WEO) database. - If data missing, perform HP decomposition of GDP growth series and take trend as potential GDP. - Overvaluation: - Use EQCHANGE dataset (CEPII) on effective exchange rates and misalignments (Couharde et al. (2018)). - Methodology relies on BEER estimates to define equilibrium real effective exchange rate; misalignment = observed REER − estimated equilibrium. - CFMs (residual state): - Quantified using yearly changes in total capital inflow restrictions (Fernández et al. (2016)) and macroprudential tightening/loosening actions (Alam et al. (2019)). - Note: The macroprudential measure used to codify CFMs captures tightening/loosening actions and differs from the macroprudential regressor used elsewhere (which reflects number of regulations in place). - Motivation: explicitly capture policy changes. _Italic: Source — wpiea2021034-print-pdf (Tables and Figures sections)._ ### 8.4 Factor Model with Time-Invariant Factor Loadings ### 8.4 Factor Model with Time-Invariant Factor Loadings ### Estimation Results - Factor loadings represent the correlation between country capital flows series and the estimated global factor. - Estimated factor loadings display significant cross-country heterogeneity, with some countries having very low exposure. - Figure 12: Estimated Factor Loadings (Notes: Estimated Factor Loadings with 86% confidence bands.) - Variance decomposition for each country in a 1 factor model (with series in deviation from their time series mean) is given by: - V(̃cf i,t) = 1≈ λ̂2 i V(cf i,t) + V(ε̂ i,t)/V(cf i,t) - R2 i ≡ 1 − V(ε̂ i,t)/V(cf i,t) - Figure 13: Variance Decomposition (Notes: Estimated Variance Decomposition from dynamic factor model) confirms high cross-country heterogeneity in exposure to the GFCy. ### What Drives the Sensitivity of Capital Flows to the GFCy? - Objective: regress λi on sample averages of explanatory variables to study domestic policy actions and economic fundamentals influencing exposure to the first common factor. - Small sample size and model uncertainty make a simple between estimator unsatisfactory. - Methodology: Bayesian Model Averaging (BMA) over all subsets of k regressors (total 2^k models). Model Mj: Λ = αι + Xj × βj + σϵ where ϵ ∼ N(0, I). Posterior probabilities combine marginal likelihoods l_Λ(Mj) and priors. - Priors used: non-informative on error-variance and Gaussian informative on parameters (algorithm following De Luca and Magnus (2011)). Summary of BMA results (Table 2 referenced in source): - Posterior Inclusion Probabilities (PIPs) identify most robust correlates with estimated sensitivities; PIP is the sum of posterior model probabilities for all models including the regressor. - Variables with highest PIPs: - CPI inflation: PIP = 0.83 (increases sensitivity to the GFCy) - Fiscal policy in low debt countries: PIP = 0.70 (increases sensitivity to the GFCy) - Level of trade integration: PIP = 0.59 (increases sensitivity to the GFCy) - All three belong to the “Fundamentals” category. - Moderate PIPs observed for: - Short term rate - Degree of sterilization - Degree of institutional quality - Degree of trade integration (noting trade integration also listed among highest PIPs) - Current account - Notable signs and interpretations: - Resistance to appreciate: positive coefficient — potentially endogenous (countries resisting appreciation are also more exposed). - Short term rate: sensitivity surprisingly decreases with higher short term rate — consistent with CPI inflation coefficient. - Capital controls, macroprudential index, and WaMaFi index: relatively low PIPs; associated coefficients point in the “expected” direction (higher levels associated with lower sensitivity), but lack robustness. - Net foreign assets (NFA, specifically “other assets” category): positive sign — higher NFA associated with higher exposure to the GFCy; interpreted via Adler and Garcia-Macia (2018) as higher NFAs acting as insurance against negative domestic shocks, encouraging foreign investor entry when global financial constraints relax. - Overall assessment: - The exercise yields relatively low PIPs for most covariates and is only suggestive of potential relationships. - Two major limitations: (1) it uses averages of variables over sample years, and (2) it is based on fixed sensitivities over time. *Italic: Source — 8.4 Factor Model with Time-Invariant Factor Loadings, wpiea2021034-print-pdf.* --- _Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021034-print-pdf.pdf_