## wpiea2026060-source-pdf ## Source details **Canonical URL:** [wpiea2026060-source-pdf](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026060-source-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026060-source-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026060-source-pdf.pdf.json) --- ### 1. Introduction and paper structure - Organized to analyze capital flows to emerging market economies (EMEs) through empirical and theoretical lenses. - Major parts: - "2. Drivers of Capital Flows to EMEs: A Simple Analytical Framework" (page 13) - "3. Capital Flows to Emerging Economies: Stylized Facts" (page 14) with data, correlations, Dynamic Factor Model (DFM) identification, extensions, and robustness. - "4. Capital Flows and Country Spread Dynamics: A Structural Multi-Country Model" (page 45) with model, calibration, evaluation, impulse responses, and counterfactuals. - "5. Conclusions" (page 63) - Appendices and supplementary material: Data Appendix (A), Additional Empirical Results (B), Theoretical Model (C). ### 3.1 Data — measurement, sample, and descriptive dynamics - High-frequency proxy for net capital inflows: - KI_t = M_t − X_t + R_t − R_{t−1} (equation (3)). - Seasonal removal: ∆KI_t = 11 ∑_{k=0} KI_{t−k} − 11 ∑_{k=0} KI_{t−k−1} (equation (4)). - Deflated using U.S. Producer Price Index – All Commodities (PPIACO). - Spread measure: - J.P. Morgan Emerging Market Bond Index Global (EMBI Global) per country. - Sample and period: - Sample period: begins in 1997:2 and ends in 2022:7. - Countries: Argentina, Brazil, China, Colombia, Ecuador, Malaysia, Mexico, Panama, Philippines, Poland, South Africa and Turkey. - Sample accounts for 55 percent of median gross inflows to a larger pool of 85 EMs over 2017 and 2021. - Descriptive episodes: - Late 1990s crises, mid-2000s inflows, 2008 GFC spike, 2012–2013 Taper Tantrum effects, early-2020 Covid shock. - Sample ends mid-2022; does not capture most of the U.S. monetary tightening after that. - Within-country correlation (Table 1): - Median correlation ρ(s,f) = -0.11 for 1997:1-2022:7. - Range: -0.20 (China) to 0.00 (Argentina and Panama). - Across-country comovement (cohesion measure, Figure 3): - Country spreads correlation ≈ 0.6 across EMEs (noted as close to double that of capital flows). - Spreads correlation increased from around 0.4 (early 2000s) to above 0.9 during the Global Financial Crisis; similar spike in 2020. - Capital flows display much lower and less time-varying comovement. - Empirical strategy takeaway: - Descriptive statistics point to an important role for supply shocks and common drivers that affect spreads more than capital flows. ### 3.4 Disentangling drivers — DFM specification and identification - Two-step empirical approach: - Step 1: identify common and idiosyncratic drivers using Dynamic Factor Model (DFM). - Step 2: identify supply vs demand disturbances via sign and zero restrictions. - DFM specification: - Xt = βFt + εt - Ft = γFt−1 + ηt - Ft is 2×1 with Fk_t (capital flows) and Fs_t (country spreads), AR(1). - Identification: βN+1:2×N,1 = 0 and β1:N,2 = 0 so one factor is for capital flows and one for spreads. - Estimated common factors: patterns and correlation - Fs_t (spreads): five spikes overlapping global market turmoil; contemporaneous correlation with VIX ≈ 0.6 (noted 0.57). - Fk_t (capital flows): different pattern, lower common volatility; strong decline in 2020 and partial recovery in 2021. - Correlation between factors: −0.58 (strong negative). - Structural sign/zero restrictions: - Common supply shock εS,G_t: moves Fs_t and Fk_t in opposite directions. - Common demand shock εD,G_t: moves Fs_t and Fk_t in same direction. - Idiosyncratic supply εS,i_t and demand εD,i_t move country i spreads and flows with opposite or same signs respectively and do not affect Fs_t or Fk_t. ### 3.4 Variance decompositions — quantitative results (baseline DFM) - Country spreads (one-year ahead variance decomposition; median shares): - Common drivers (εS,G_t + εD,G_t): 64 percent (median). - Idiosyncratic drivers (εS,I_t + εD,I_t): 36 percent (median). - Within common drivers: common supply εS,G_t explains 64 percent of Fs_t variability; common demand explains 36 percent of Fs_t variability. - Combining common and idiosyncratic disturbances: supply forces explain 59 percent of spreads dynamics; demand 41 percent. - Country heterogeneity examples: Argentina and Ecuador: 9 and 28 percent of spread variance due to global factors; Mexico and Colombia: 90 and 75 percent due to global factors. - Net capital flows (one-year ahead variance decomposition; median shares): - Idiosyncratic drivers (εS,I_t + εD,I_t): 90 percent (median). - Common drivers (εS,G_t + εD,G_t): 10 percent (median). - Within capital flows combining common and idiosyncratic: supply forces explain a median of 51 percent; demand explains 49 percent. - Country heterogeneity: Brazil exception with 51 percent global vs 49 percent idiosyncratic for net capital flows; other countries have global share below 28 percent. - Historical decomposition (Figure 6): - Supply shocks dominate common-factor variability during global crises (late 1990s, GFC, Covid). - Demand shocks relatively more important mid-2010s, helping keep spreads low while flows receded. ### 3.6 Robustness — alternative measures, aggregation, and samples - Alternative measures and datasets produce consistent qualitative results: - Brazil monthly BoP disaggregated proxy: correlation between flows and spread −0.25 (comparable to baseline −0.12). - Quarterly corporate-debt-based data: median correlation −0.20. - Debt issuance and yields (selected EMEs): median correlation −0.25. - IMF quarterly BoP net inflows: median correlation −0.28. - Koepke and Paetzold monthly flows with EMBI: median correlation −0.27. - Quarterly balanced-panel DFM (1999Q1:2022Q2): - Main conclusions hold: common credit supply shocks predominate for spreads; idiosyncratic shocks dominate capital flows. - Temporal aggregation reduces contribution of common shocks to capital flows. - Gross flows analysis (quarterly, 1999:Q1-2022:Q2, ten economies): - Contemporaneous correlations (Table 8): - Net flows correlation with EMBI at quarterly frequency: −0.33 (vs −0.14 monthly). - Gross inflows median correlation noted as −0.43 (table median). - Common factor importance (Table 9): - Inflows: median Adjusted R2 = 0.31. - Outflows: median Adjusted R2 = 0.06. - Interpretation: common factor explains gross inflows more than outflows; gross measures show greater comovement than net flows. - Portfolio investment and FDI: - Disaggregated FDI and portfolio flows driven mostly by supply and idiosyncratic shocks; FDI largely idiosyncratic. - Subsample with open financial accounts: - Median correlation between flows and spreads: −0.12 (full sample median −0.11). - Pre-Covid sample (ending 2019:12): - Dynamics and variance decompositions comparable; correlation between factors with and without Covid = 0.98. ### 4.1 The Model — two-EME small-open-economy RBC setup - Households: GHH preferences, utility (equation (6)) with parameters β, γ, ω. - Budget constraint (equation (7)), No-Ponzi constraint (equation (8)). - Production: y_t = A_t k_t^{α} h_t^{1−α} (equation (9)); capital accumulation (equation (10)). - Capital flows: kf_t = d_t − d_{t−1} (equation (11)). - Shocks and processes: - Productivity ln A_t = ρ_A ln A_{t−1} + ε_At with Σ_A allowing cross-country covariance σ_{A12} ≠ 0. - Country interest rate R_t = R^*_t + ε_{r t} with Σ_R allowing σ_{12} ≠ 0 (spread correlation targeted ≈ 0.66). - International interest rate ĤR^*_t = ζ ĤR^*_{t−1} + ε^*_t. - Baseline calibration (quarterly, Brazil and Mexico, 1997:Q1–2019:4) — selected parameter values (Table 10): - γ = 2 - ω = 1.455 - δ = 0.025 - α = 0.32 - R^* Annual Interest rate in SS: 1.04 - β = 0.9901 - ζ = 0.92 - σ_{R^*} = 0.00183 - d (debt in SS): Brazil 0.38, Mexico 3.9 - φ (capital adjustment cost): Brazil 0.00386, Mexico 0.00767 - ψ (Portfolio adjustment costs): Brazil 0.000000147, Mexico 0.0000005 - ρ_A: Brazil 0.6355, Mexico 0.765 - σ_{A,ii}: Brazil 0.00653, Mexico 0.00505 - σ_{A,ij}: 0.3, 0.3 - σ_{R,ii}: Brazil 0.0083, Mexico 0.0033 - σ_{R,ij}: 0.707, 0.707 - Model evaluation (Table 11 — selected moments): - std(y_t): Brazil 1.73 vs 1.73; Mexico 1.68 vs 1.68 - std(i_t): Brazil 5.49 vs 5.46; Mexico 4.40 vs 4.40 - std(tby_t): Brazil 2.50 vs 2.50; Mexico 1.63 vs 1.25 - std(kf_t / y_t): Brazil 3.16 vs 0.73; Mexico 1.14 vs 0.60 - std(R^*_t): 0.45 vs 0.45 - std(s_t): Brazil 0.83 vs 0.83; Mexico 0.32 vs 0.32 - corr(s_t, kf_t / y_t): Brazil −0.21 vs −0.10; Mexico −0.03 vs −0.11 - cross-country corr(s_BR, s_MEX): 0.65 vs 0.65; corr(kf_BR, kf_MEX): 0.35 vs 0.34 - Assessment: - Model reproduces high comovement of spreads and lower comovement of net capital flows. - Limitations: net capital flows in the model are less volatile and more persistent than data. ### 4.5–4.6 Impulse responses and synchronization counterfactuals - Impulse responses (selected qualitative findings): - Domestic TFP shock: immediate increase in capital flows, protracted decrease thereafter; spread falls protractedly. - Domestic spread shock (increase): raises country interest rate, improves current account via capital outflows; investment falls. - International interest rate shock (increase): shifts credit supply upward, raises country spread gradually, induces capital outflows. - Cross-country transmission: an increase in Mexico’s spread raises Brazilian spread and induces Brazilian capital outflows due to spread correlation. - Counterfactual experiments (Table 12 — corr results Data | Baseline | Exp.1 | Exp.2 | Exp.3): - Exp.1: set σ_{A,ij} = 0 (no correlated TFP shocks). - Exp.2: set σ_{R,ij} = 0 (no correlated spread shocks). - Exp.3: set σ_{R,ij} = σ_{A,ij}. - Results: - corr(y_BR, y_MEX): Data 0.30 | Baseline 0.30 | Exp.1 0.00 | Exp.2 0.30 | Exp.3 0.30 - corr(s_BR, s_MEX): Data 0.65 | Baseline 0.65 | Exp.1 0.65 | Exp.2 0.00 | Exp.3 0.30 - corr(kf_BR, kf_MEX): Data 0.35 | Baseline 0.34 | Exp.1 0.08 | Exp.2 0.28 | Exp.3 0.33 - Interpretation: - Correlated TFP shocks are key for income and capital-flow synchronization. - Correlated spread shocks drive spread synchronization (Global Financial Cycle) with limited effects on capital-flow synchronization. ### 4.7 Model extensions — estimated rates, credit risk, and global solution methods - Estimated interest rate processes (Uribe and Yue, 2006): - Country interest rate function includes lagged and contemporaneous R^*, output, investment, trade balance terms plus ε_{r_t}. - Augmented model yields similar main findings: business cycle synchronization explains over half of capital-flow comovement; correlated spread shocks explain little of capital-flow comovement. - Unconditional correlations in augmented simulations (Table 13): corr(y_BR, y_Mex): 0.30 baseline; corr(s_BR, s_MEX): 0.65 baseline; corr(kf_BR, kf_MEX): baseline 0.39. - Credit risk shocks (stochastic elasticity of interest rate to debt): - Replace PAC with Internal Debt Elastic interest-rate device; ψ_t = ψ + ε_{ψ_t}. - Mechanism amplifies capital-flow volatility via slope shifts in credit supply and household internalization of debt effects. - Calibration matches capital-flow volatility but: - Induces excess volatility in investment. - Cross-country comovement of capital flows falls (from 0.35 to 0.03). - Country-level correlation between spreads and flows remains similar. - Solving with global methods (FiPIt fixed-point global solution): - Global solution raises capital-flow volatility substantially (Brazil and Mexico volatilities double relative to baseline). - Trade-off: cross-country correlation in capital flows increases to 0.63, above data 0.35. - Extension conclusions: - Modifications can improve match to some moments (capital-flow volatility) but entail trade-offs (excess investment volatility or excessive cross-country comovement). ### Executive summary of substantive findings (empirical and theoretical synthesis) - Stylized empirical facts: - Within-country correlation between spreads and capital flows is negative albeit low (median −0.11). - Cross-country comovement in spreads is much stronger than that of capital flows (spreads correlation ≈ 0.6 vs flows much lower). - DFM decomposition: - Common factor in spreads accounts for 64 percent of spreads' variability (median). - Two thirds of common-spread variability traced to supply shocks; supply shocks account for 41 percent of the common-spread variability (reported as majority portion identified). - Common factor in capital flow volumes accounts for less than 10 percent of their volatility (median common drivers 10 percent); idiosyncratic shocks explain 90 percent (median), roughly split between supply and demand origins. - Extensions: - Including small advanced economies separates a global factor from EME-specific factor; baseline EME factors highly correlated with EME-specific factors in larger sample. - Regional factors (Asia, Latin America) materially increase the share of capital-flow variance attributable to common forces (variance share nearly quadruples when controlling for regional drivers). - U.S. monetary policy shocks impact common factors in spreads and volumes. - Structural model insights: - Calibrated two-EME SOE/RBC model reproduces moderate cross-country correlation of net capital flows and low negative within-country correlation between spreads and flows. - Counterfactuals: - Turning off correlated TFP shocks reduces capital-flow correlation to about one fourth of baseline but does not affect spread correlation. - Turning off joint spread shocks reduces spread synchronization markedly but leaves capital-flow and income correlations largely unchanged. - Interpretation: - Global Financial Cycle through common credit supply shocks is an important source of EME spread fluctuations. - Common TFP shocks are key for synchronization in net capital flows. - Overall key takeaways: - Fluctuations in sovereign spreads across EMEs are largely driven by common credit supply shocks. - Fluctuations in net capital inflows are predominantly country-specific and attributable to idiosyncratic credit supply and demand drivers in roughly equal measure. - Allowing for regional components substantially increases the share of capital-flow variation attributable to common forces. - Gross flow analysis indicates greater commonality in gross measures—especially gross inflows—than in net financing. - A parsimonious small-open-economy framework with correlated productivity and borrowing-condition shocks can jointly rationalize observed patterns in prices and quantities across EMEs. *Source: wpiea2026060-source-pdf (IMF Working Paper — content unit provided).* ### 1. Introduction ........................................................................................................ ### 1. Introduction ### Overview of the paper's structure and themes - The paper begins with "1. Introduction" on page 6. - It is organized to analyze capital flows to emerging market economies (EMEs) through empirical and theoretical lenses, with major parts including: - "2. Drivers of Capital Flows to EMEs: A Simple Analytical Framework" (page 13) - "3. Capital Flows to Emerging Economies: Stylized Facts" (page 14), with subsections: - 3.1 Data (page 14) - 3.2 Correlation Between Spreads and Capital Flows within EMEs (page 21) - 3.3 Correlation Between Spreads and Capital Flows Across EMEs (page 22) - 3.4 Disentangling the Drivers of Capital Flows to EMEs (page 24), including: - Estimating Common Factors (page 24) - The Role of Supply and Demand Drivers (page 27) - 3.5 Extensions (page 32), including: - Adding Advanced Economies (page 32) - Regional EME Factors (page 33) - Impact of U.S. Monetary Policy (page 36) - 3.6 Robustness (page 37), including: - Alternative Measures of Capital Flows and Country Spreads (page 39) - Portfolio Investment and FDI (page 43) - Economies with Open Financial Account (page 44) - "4. Capital Flows and Country Spread Dynamics: A Structural Multi-Country Model" (page 45), with subsections: - 4.1 The Model (page 47) - 4.2 Driving Forces (page 49) - 4.3 Calibration (page 51) - 4.4 Model Evaluation (page 52) - 4.5 Impulse Responses (page 54) - 4.6 Assessing the Determinants of Synchronization (page 56) - 4.7 Extensions of the Theoretical Model (page 58), including: - Estimated Interest Rate Processes (page 59) - Credit Risk Shocks (page 60) - Solving the Model with Global Methods (page 62) - "5. Conclusions" (page 63) ### Empirical and supplementary material indicated in the Introduction - Appendices and supplementary sections listed after the main text include: - A. Data Appendix (page 71), with A.1 Empirical Analysis (page 71) - B. Additional Empirical Results (page 73), with subsections including: - B.1 Correlation between Capital Flows and Country Spreads within EMEs (page 73) - B.2 Correlation between Capital Flows Across EMEs (page 73) - B.3 Dynamic Factors (page 76), including: - B.3.1 Non-Cumulated Factors (page 76) - B.3.2 Alternative Variance Decomposition (page 76) - B.3.3 Extensions of the Baseline DFM (page 77) - B.3.4 Alternative DFM Specification (page 78) - B.3.5 Regional Factors (page 79) - B.4 Alternative Empirical Specifications (page 81), including: - B.4.1 Alternative Datasets (page 81) - B.4.2 Net FDI and Portfolio Flows (page 81) - B.4.3 Pre-COVID-19 Sample (page 82) - C. Theoretical Model (page 85), including: - C.1 Simple Two-Period Analytical Model (page 85), with: - C.1.1 Demand for Credit without Uncertainty (page 85) - C.1.2 Equilibrium in Credit Markets with Uncertainty (page 86) - C.2 Two-Country Small Open Economy Model (page 89), with: - C.2.1 Equilibrium Conditions (page 89) - C.2.2 Additional Results from the Theoretical Model (page 89) - C.3 Extensions of the Theoretical Model (page 92), including: - C.3.1 Estimated Interest Rate Processes (page 92) - C.3.2 Credit Risk Shocks (page 94) - C.3.3 Solution with Global Methods (page 94) *Source: wpiea2026060-source-pdf (IMF Working Paper — table of contents provided)* ### References ............................................................................................................. ### wpiea2026060-source-pdf - References ### Equations and Formal Elements - Equations listed in the content unit: - Equation 1 - Equation 2 - Equation 3 - Equation 4 - Equation 5 - Equation 6 - Equation 7 - Equation 8 - Equation 9 - Equation 10 - Equation 11 - Equation 12 - Equation 13 - Equation 14 - Equation 15 ### Figures (enumeration and focus) - Figures enumerated in the content unit: - Figure 1. Equilibrium in International Credit Markets - Figure 2. Capital Flows and Country Spreads in EMEs - Figure 3. Dynamic Correlation of Country Spreads and Capital Flows Across EMEs - Figure 4. Common Factor for Country Spreads and Capital Flows - Figure 5. Common Factor of Country Spreads and the VIX - Figure 6. Historical Decomposition of the Common Factors - Figure 7. Global and EME-Specific Factors - Figure 8. IRFs of both Factors to Contractionary U.S. Monetary Policy Shocks - Figure 9. The Case of Brazil - Appendix/Online figures: Figure A.1 through Figure A.6 (topics include country-level flows and spreads, common and regional factors, COVID-19 exclusion, IRFs in Mexico) ### Tables (enumeration and focus) - Tables enumerated in the content unit: - Table 1 through Table 13 (core analysis: correlations, shares of variance, calibrated parameters, empirical and theoretical moments, unconditional moments, counterfactuals) - Appendix tables: Table A.1, Table A.2, Table A.3 through Table A.29 (coverage includes lists of systemic sudden stop episodes, one-year ahead variance decompositions, alternative DFM specifications, alternative measures, calibrated parameters, and empirical/theoretical moments across variants) ### Glossary (definitions and abbreviations) - Terms and abbreviations provided: - Advanced Economies (AE) - Balance of Payments (BoP) - Dynamic Factor Model (DFM) - Emerging Markets (EM) - Emerging Market Bond Index (EMBI) - Emerging Market Economies (EME) - Foreign Direct Investment (FDI) - Global Financial Crisis (GFC) - Implied Volatility Index (VIX) - Impulse Response Functions (IRF) - Large-Scale Asset Purchases (LASP) - Net Foreign Asset (NFA) - Portfolio Adjustment Costs (PAC) - Real Business Cycle (RBC) - State Owned Enterprises (SOE) - Standard Deviation (S. D.) - Total Factor Productivity (TFP) - Treasury Bills (T-Bill) ### Executive summary of substantive findings and quantitative results - Stylized empirical facts: - The within-country correlation between spreads and capital flows is negative albeit low. - The degree of cross-country comovement in spreads is much stronger than that of capital flows. - Dynamic Factor Model (DFM) quantitative decomposition: - A common factor in spreads accounts for 64 percent of spreads' variability. - Two thirds of that common-spread variability can be traced back to supply shocks. - Supply shocks account for 41 percent of the common-spread variability (reported as the majority portion identified). - A common factor in capital flow volumes accounts for less than 10 percent of their volatility. - 90 percent of capital flow volatility (within the volume common factor context) is driven by idiosyncratic shocks, which are roughly evenly split between supply and demand origins. - Extensions and additional findings: - Augmenting the sample with small advanced economies differentiates a global factor from an EME-specific factor; baseline EME factors are highly correlated with EME-specific factors from the larger sample. - Global shocks are equally relevant—if not more—than common-EME shocks for external drivers of capital flows and spreads. - Regional analysis finds two distinct regional factors for Asia and Latin America; regional factors materially increase the share of capital-flow variance attributable to common forces (variance share nearly quadruples when controlling for regional drivers). - Shocks to U.S. monetary policy—conventional and unconventional—impact the common factors in spreads and volumes of capital flows to EMEs. - Robustness and sensitivity checks: - Results hold across alternative measures of capital flows and sovereign borrowing conditions (including corporate debt flows and spreads). - Analysis replicated using gross inflows and outflows; gross measures show more apparent commonality—particularly gross inflows—than net financing. - Additional robustness: focusing on net portfolio and FDI flows, restricting to EMEs with more open financial accounts, and re-estimating on pre-COVID sample ending in 2019. - Structural model (two-EME SOE/RBC calibration to Brazil and Mexico) insights: - The calibrated structural model reproduces: - Moderate cross-country correlation of net capital flows. - Low and negative within-country correlation between spreads and capital flows. - Counterfactual experiments: - Turning off correlation of TFP shocks across countries reduces capital-flow correlation to about one fourth of the baseline calibrated data match but does not affect spread correlation. - Turning off joint shocks to spreads (reducing presence of a Global Financial Cycle) leaves capital-flow and income correlations largely unchanged but considerably reduces synchronization in spreads. - Interpretation: - A Global Financial Cycle that materializes through common credit supply shocks is an important source of EME spread fluctuations. - Common TFP shocks are key to understanding synchronization in net capital flows to EMEs. - Overall key takeaways (concise): - Fluctuations in sovereign spreads across EMEs are largely driven by common credit supply shocks. - Fluctuations in net capital inflows are predominantly country-specific and attributable to idiosyncratic credit supply and credit demand drivers in roughly equal measure. - Allowing for regional components substantially increases the share of capital-flow variation attributable to common forces. - Gross flow analysis indicates greater commonality in gross measures—especially gross inflows—than in net financing. - A parsimonious small-open-economy framework with correlated productivity and borrowing-condition shocks can jointly rationalize the observed patterns in prices and quantities across EMEs. ### Structure of the full analysis (as organized in the content unit) - Section 1: Introduction (motivation, dataset emphasizing joint observation of volumes and risk/spreads, empirical strategy) - Section 2: Drivers of Capital Flows to EMEs: A Simple Analytical Framework (two-period SOE model; demand and supply drivers; idiosyncratic vs common shocks; illustrative Figures) - Section 3: Capital Flows to Emerging Economies: Stylized Facts (data description; correlations within and across countries; formal identification with DFM; extensions for AEs, regional factors, and US monetary policy; robustness) - Section 4: Structural Model Results (two-EME calibrated SOE/RBC model; counterfactuals and extensions) - Section 5: Conclusion *Source: wpiea2026060-source-pdf - References* ### 3.1 Data ### 3.1 Data ### High-frequency proxy for capital inflows - Monthly balance of payments data for EMEs are often unavailable; to work at monthly frequency the analysis follows Calvo et al. (2008) and constructs a proxy for net capital inflows: - KI_t = M_t − X_t + R_t − R_{t−1} (equation (3)), where KI_t denotes capital inflows in period t, X_t denotes exports, M_t denotes imports, and R_t is the stock of international reserves. - The proxy nets out the trade balance from changes in foreign reserves and therefore does not include net factor income and current transfers. - The proxy abstracts from errors and omissions that could add volatility to capital flow measures. - To remove seasonal movements the authors use cumulative annual flows for each month and then take the first difference of this measure: - ∆KI_t = 11 ∑_{k=0} KI_{t−k} − 11 ∑_{k=0} KI_{t−k−1} (equation (4)). - The capital flows measure is deflated using the U.S. Producer Price Index – All Commodities (PPIACO from FRED) to obtain a volume measure of net capital flows. ### Choice of price variable and spread measure - Country spreads are used instead of country interest rates because spreads reflect issues specific to emerging economies rather than global interestrate drivers. - Spread measure: J.P. Morgan Emerging Market Bond Index Global (EMBI Global) for each country (an arithmetic, market-capitalization weighted average of US-dollar denominated bond spreads issued by sovereign and quasi-sovereign entities over U.S. Treasury bonds of similar duration). ### Sample, period, and countries - Sample period: begins in 1997:2 and ends in 2022:7. - Baseline analysis includes the COVID-19 period (an extension excluding it is considered elsewhere in the chapter). - The sample consists of 12 countries with continuous monthly data on capital flows and country spreads: Argentina, Brazil, China, Colombia, Ecuador, Malaysia, Mexico, Panama, Philippines, Poland, South Africa and Turkey. - This sample accounts for 55 percent of median gross inflows to a larger pool of 85 EMs over the years 2017 and 2021. - Aggregate series construction: capital flows for each country are defined as the cumulative trade deficit plus the change in international reserves at monthly frequency (equation (4)); the aggregate series is computed as the monthly median of the country series and then standardized. Medians are employed to prevent China from driving dynamics. ### Descriptive dynamics highlighted - Historical episodes noted: - Asian and Russian crises (late 1990s): significant increase in country spreads and capital outflows. - Mid-2000s: country spreads declined and capital inflows increased. - Global Financial Crisis (2008): sharp yet transitory increase in country spreads and a fall in net capital flows; capital inflows resumed vigorously in 2010. - Decline in capital flows starting in 2012 coinciding with the peak in commodity prices; related volatility around the Taper Tantrum in 2013 and U.S. rate expectations. - Strong but brief decline in capital flows in early 2020 associated with the Covid outbreak. - The sample ends in mid-2022 and therefore does not capture most of the increase in the U.S. monetary stance; no major volatility is observed at the start of that tightening cycle within the sample. ### Correlation between spreads and capital flows — within countries - Table 1 results (contemporaneous correlation between country spreads and capital flows, country-level, period 1997:1-2022:7): - Median correlation ρ(s,f) = -0.11. - Range in full sample correlations: -0.20 (China) to 0.00 (Argentina and Panama). - No country exhibits a strictly positive correlation in the sample. - Robustness: correlation remains similar when excluding Sudden Stop episodes (second row of Table 1). - Interpretation: negative and relatively low correlations suggest a predominance of credit supply shocks driving the negative relation between spreads and capital flows in most sample countries (to be explored formally in subsequent sections). - Sudden Stop definition (following Calvo et al., 2008): - i) At least one observation where year-on-year decline in capital flows is at least two standard deviations below its sample mean (unpredicted decline). - ii) Sudden stop phase ends when annual change in capital flows surmounts one standard deviation below its sample mean. - iii) Onset also ascertained by the first time the annual change drops one standard deviation below the mean. - First and second moments of the capital flow series are calculated with an expanding window with a minimum of 24 months and a start date fixed at January 1997. ### Correlation and comovement — across EMEs - Methodology: cohesion measure (5-year rolling correlation across countries at different frequencies) following Croux et al. (2001); points computed with 5-year backward rolling window. - Main findings from cohesion analysis (Figure 3): - Comovement in prices (spreads) is much stronger than comovement in volumes (capital flows). - Country spreads display a high correlation of 0.6 across EMEs, close to double that of capital flows. - Comovement is time-varying for spreads and aligns with global financial stress episodes (e.g., correlation increased from around 0.4 in the early 2000s to above 0.9 during the Global Financial Crisis; similar increase in 2020 at the start of the COVID pandemic). - Capital flows show much lower and less time-varying comovement. - Both correlations are stable across business cycle frequencies, indicating the relationships are not driven solely by high-frequency transients. - Component analysis note: change in foreign reserves is slightly more correlated across countries than the trade balance, but both components display significantly lower correlation than country spreads (see Section B.2 noted by authors). ### Takeaway for empirical strategy - Descriptive statistics point to an important role for supply shocks and common drivers that affect spreads more than capital flows. - The chapter will proceed to a formal econometric approach, jointly observing volumes and prices of capital flows across EMEs following the conceptual framework in Section 2. *Source: IMF — Section 3.1 ("Data") of the provided chapter (period and figures as stated).* ### 3.4 Disentangling the Drivers of Capital Flows to EMEs ### 3.4 Disentangling the Drivers of Capital Flows to EMEs ### Methodology: overview and dynamic factor model (DFM) - Two-step approach: - Step 1: Pin down common and idiosyncratic drivers of capital flows and country spreads. - Step 2: Identify supply and demand disturbances behind those drivers using sign and zero restrictions. - Dynamic Factor Model (DFM) specification: - Xt = βFt + εt - Ft = γFt−1 + ηt - Xt is the (2×N)×1 matrix of capital flows and country spreads for N countries. - β is the (2×N)×2 matrix of factor loadings. - Ft is 2×1 with two common factors: Fk_t (capital flows) and Fs_t (country spreads), assumed AR(1). - εt is (2×N)×1 idiosyncratic shocks; γ is 2×2 persistence; ηt is (2×1) disturbances to common factors. - Identification of separate common factors: - Imposed βN+1:2×N,1 = 0 and β1:N,2 = 0 so the first factor is identified using only capital flows and the second only spreads. - Presentation convention: - Factors presented in cumulative terms from start of sample; units relate to standard deviations. - For spreads, first difference of monthly average is standardized. For capital flows (already in first difference) only standardized. ### Estimated common factors: patterns and correlation - Common factor for country spreads (Fs_t): - Displays five clear spikes overlapping world capital market turmoil: Russian Crisis (late 90s), spike in late 2001 (Argentinian sovereign default), the Global Financial Crisis (GFC), a milder spike around 2015 (lift off of the FED’s Funds Rate), and onset of the Covid pandemic in 2020. - Tracked closely by the U.S. CBOE Implied Volatility Index (VIX); contemporaneous correlation of 0.6 (correlation reported as 0.57 in Figure 5 note). - Common factor for capital flows (Fk_t): - Different pattern: initially fell following late 90s crises until early 2000s, then reversed to gradually increase through the 2000s. - Increased volatility and clear acceleration in the two years before the GFC, fell abruptly during the GFC, recovered with a spike in 2012, remained high until late 2014, decreased sharply amid US policy normalization until 2016, then reverted at lower levels, and displayed a strong decline in 2020 with the Covid outbreak before recovering in 2021. - Correlation between the two common factors: - Strong negative correlation of −0.58, suggesting supply forces dominate among common drivers but demand forces and idiosyncratic shocks also matter. ### Identification of supply vs. demand and common vs. idiosyncratic shocks - Imposed structural identification through sign and zero restrictions: - Common supply shock (εS,G_t): moves Fs_t and Fk_t in opposite directions. - Common demand shock (εD,G_t): moves Fs_t and Fk_t in the same direction. - Idiosyncratic supply shock in country i (εS,i_t): moves country i spreads and capital flows in opposite directions, without affecting Fs_t or Fk_t. - Idiosyncratic demand shock in country i (εD,i_t): moves country i spreads and capital flows in the same direction, without affecting Fs_t or Fk_t. - Interpretation examples: - Global supply force: US monetary tightening → investors redirect to US → spreads and capital flows move in opposite directions. - Global demand force: commodity-price boom → simultaneous demand boost across commodity-exporting EMEs → spreads and capital flows move in same direction. - Country-specific supply: political uncertainty reduces foreign supply → spreads and capital flows move in opposite directions. - Country-specific demand: expansionary fiscal policy increases demand for capital → spreads rise while capital inflows finance public spending. ### Variance decompositions: country spreads (Table 2) — high-level findings - Overall median shares (one-year ahead variance decomposition of country spreads): - Common drivers (εS,G_t + εD,G_t): 64 percent (median). - Idiosyncratic drivers (εS,I_t + εD,I_t): 36 percent (median). - Supply vs. demand within common drivers (median): - Common supply εS,G_t explains 64 percent of the variability in the common factor Fs_t. - Common demand εD,G_t explains 36 percent of Fs_t variability. - When common and idiosyncratic disturbances are combined: - Supply forces explain 59 percent of spreads dynamics; demand disturbances explain 41 percent. - Country heterogeneity examples from Table 2 (shares of variance explained by common shocks): - Argentina and Ecuador: relatively low variance in spreads associated to global factors, 9 and 28 percent respectively. - Mexico and Colombia: 90 and 75 percent of spread variance accounted for by global factors. (Note: Table 2 contains country-level numeric entries as in the source.) ### Variance decompositions: net capital flows (Table 3) — high-level findings - Overall median shares (one-year ahead variance decomposition of net capital flows): - Idiosyncratic drivers (εS,I_t + εD,I_t): 90 percent (median). - Common drivers (εS,G_t + εD,G_t): 10 percent (median). - Supply vs. demand within common drivers for capital flows: - Common supply and demand split but common drivers are a small share overall; supply shocks drive most of the variability in the estimated common factor Fk_t. - When combining common and idiosyncratic disturbances: - Supply forces explain a median of 51 percent of capital flows dynamics; demand explains the remaining 49 percent. - Country heterogeneity examples from Table 3: - Brazil is an exception: share of global disturbances in net capital flows 51 percent vs. idiosyncratic 49 percent. - For remaining countries, share of capital flow dynamics accounted for by common disturbances is below 28 percent at most. (Note: Table 3 contains country-level numeric entries as in the source.) ### Historical decomposition of common factors (Figure 6) - Supply shocks explain most variability in common factors during global crises: - Asian Crisis (late 90s), the years surrounding the GFC, and the Covid outbreak in 2020. - Demand shocks become relatively more important toward the mid-2010s: - Contributed to keeping spreads low while capital flows were receding from EMEs. ### Key takeaways from baseline analysis - Spreads in EMEs: predominately driven by common/global factors (median 64 percent), with supply forces relatively more preponderant. - Net capital flows to EMEs: predominately driven by idiosyncratic/country-specific disturbances (median 90 percent). - Supply forces are relatively more important for spreads; for capital flows, supply and demand disturbances are nearly equal in relevance when combined across common and idiosyncratic shocks. *Source: IMF Working Paper — Section 3.4 "Disentangling the Drivers of Capital Flows to EMEs" (content unit provided).* ### 3.6 Robustness ### 3.6 Robustness ### Alternative Measures of Capital Flows and Country Spreads - Baseline uses sovereign spread based on public bonds and proxy for capital flows including public and private flows; sovereign spreads are highly correlated with corporate spreads but do not coincide. - Disaggregated monthly BoP data for Brazil (BoP M): - Proxy: Portfolio Investment (net incurrence of liabilities) and sovereign spread (EMBI). - Revised correlation between capital flows and country spread: -0.25 (comparable to baseline -0.12). - Quarterly corporate-debt-based database (CFP, Caballero et al., 2019): - Net corporate debt flows defined as difference in stock of corporate debt; country spread via External Financial Index. - Median correlation between capital flows and country spreads: -0.20 (baseline median correlation -0.11). - Quarterly corporate debt issuance and yields (Brazil, Colombia, Mexico, Turkey; 1994:1-2019:1; Debt Issuance): - Yield: weighted average of individual bond yields; issuance: sum of bonds per quarter (weights = share of every bond on total bond issued in that quarter). - Consideration: only bonds issued in international currency to partially address domestic vs international issuance ambiguity. - Median correlation between capital flows and country spreads: -0.25. - IMF quarterly Balance of Payments data (BoP Q): - Capital flows: net capital inflows from Financial Account; country spread: quarterly EMBI average. - Median correlation between capital flows and country spreads: -0.28. - Monthly capital flows series from Koepke and Paetzold (2020) with EMBI (KS Total): - Median correlation between capital flows and country spreads: -0.27. - Quarterly net capital flows from Balance of Payments used for variance decomposition (balanced panel restricted from 1999Q1:2022Q2; Poland excluded): - Main conclusions with quarterly data: - Common credit supply shocks are the predominant driver of country spreads. - Idiosyncratic credit demand and supply shocks are the most important drivers of capital flows. - Contribution of common shocks to capital flows is lower under temporal aggregation. - One-year ahead variance decomposition of country spreads (Table 6; values by country and median preserved as presented): - Share of variance explained by common credit supply shocks εS,Gt: 58 15 26 29 37 28 34 43 34 33 44 40 29 (as presented across columns; median 34) - Share of variance explained by common credit demand shocks εD,Gt: 42 11 19 21 27 21 28 31 32 25 29 21 25 (median 25) - Combined common shocks εS,Gt + εD,Gt: 100 26 45 50 64 49 67 74 75 59 69 50 59 (median 59) - Country-specific credit supply shocks εS,It: 0 40 26 26 19 27 18 13 14 20 15 27 20 (median 20) - Country-specific credit demand shocks εD,It: 0 34 28 23 17 25 15 13 12 21 16 23 21 (median 21) - Combined idiosyncratic shocks εS,It + εD,It: 0 74 54 49 36 52 33 26 26 41 31 50 41 (median 41) - One-year ahead variance decomposition of net capital flows (Table 7; values by country and median preserved as presented): - Share of variance explained by common credit supply shocks εS,Gt: 58 7 20 82 95 18 14 21 41 41 2 8 (as presented across columns; median 8) - Share of variance explained by common credit demand shocks εD,Gt: 42 5 15 62 14 16 10 21 10 21 0 9 6 (median 6) - Combined common shocks εS,Gt + εD,Gt: 100 12 35 14 50 92 14 24 24 42 44 24 21 (median 14) - Country-specific credit supply shocks εS,It: 0 47 31 45 27 48 53 34 34 13 47 36 43 (median 43) - Country-specific credit demand shocks εD,It: 0 40 34 41 23 44 12 34 44 45 33 49 37 (median 37) - Combined idiosyncratic shocks εS,It + εD,It: 0 87 65 85 50 92 65 78 78 58 80 76 83 (median 40 reported in text but table shows idiosyncratic totals as listed) ### Gross Capital Flows - Rationale: literature emphasizes studying gross capital flows; evidence shows stronger comovement in gross inflows and outflows across regions while net flows show no systematic correlation patterns. - Sample: balance of payments quarterly data for subset of baseline countries with full data; sample period 1999:Q1-2022:Q2 (ten economies). - Note: gross capital flows data for China and Poland start later; China and Poland not included in balanced-panel DFM estimation. Data for Ecuador and Panama available from 1998:Q1 and 1999:Q1 respectively. Deflator used: US Producer Price Index by Commodity: All Commodities (PPIACO). - Correlations (Table 8) — contemporaneous correlation between real capital outflows/inflows/net flows and EMBI for sample 1999:Q1-2022:Q2 (significance markers preserved): - Outflows by country: ARG −0.15, BRZ −0.20*, CHN −0.27***, COL −0.14, ECU −0.28***, MEX −0.26**, MLY −0.42***, PAN −0.37***, PHL −0.24**, POL −0.06, SWF −0.25, TUR −0.25 (as presented; median not separately listed in table). - Inflows by country: ARG −0.42***, BRZ −0.44***, CHN −0.60***, COL −0.31***, ECU −0.42***, MEX −0.40***, MLY −0.55***, PAN −0.41***, PHL −0.54***, POL −0.61***, SWF −0.43***, TUR −0.43 (median indicated as −0.43). - Net Flows by country: ARG −0.37***, BRZ −0.43***, CHN −0.67***, COL −0.23**, ECU −0.23**, MEX −0.28***, MLY −0.28**, PAN −0.14, PHL −0.48***, POL −0.58***, SWF −0.33, TUR −0.33 (median indicated as −0.33). - Findings: - Correlation between country spread and net capital flows at quarterly frequency: −0.33 (compared with −0.14 at monthly frequency). - Gross capital inflows display the most negative correlations with country spread. - Importance of common factors for gross flows (Table 9 — share of variance of country-specific gross inflows/outflows explained by the common factor; Adjusted R2 reported): - Outflows: ARG 0.01, BRZ 0.19, CHN 0.11, COL 0.01, ECU 0.01, MEX 0.49, MLY 0.18, PAN 0.26, PHL 0.01, POL 0.01, SWF 0.01, TUR 0.06 (median 0.06). - Inflows: ARG 0.03, BRZ 0.79, CHN 0.34, COL 0.01, ECU 0.32, MEX 0.34, MLY 0.30, PAN 0.03, PHL 0.08, POL 0.45, SWF 0.31, TUR 0.31 (median 0.31). - Interpretation: - Common factor explains gross capital inflows more than outflows, but its importance is significantly lower than the common factor in country spreads. - Presence of non-trivial common component in gross inflows is consistent with Global Financial Cycle evidence of comovement in gross measures. ### Portfolio Investment and FDI - Baseline uses total net capital flows; robustness checked with disaggregated net foreign direct investment (FDI) and portfolio inflows. - Sample for this exercise: 1999:Q1-2022:Q2 (Poland excluded due to data availability). - Results (Tables A.10–A.12 in Appendix referenced): - Correlation between these disaggregated flows and spreads is low and negative. - Variance decomposition: both FDI and portfolio flows are driven mostly by supply and idiosyncratic shocks. - Conclusion: main baseline conclusions hold when focusing on FDI and portfolio flows. ### Economies with Open Financial Account - Subsample defined as economies coinciding with Fernández et al. (2018): Brazil, Colombia, Mexico, Malaysia, Philippines, South Africa, Turkey. - Median correlation between capital flows and spreads for this subset: −0.12 (full sample median −0.11). - Conclusion: low and negative correlation between spreads and capital flows is not driven by economies with closed financial accounts. ### Pre-Covid Sample - Robustness check: sample ending in 2019:12, excluding Covid period. - Dynamics of estimated common factors in capital flows and country spread with shorter sample: little change; correlation between the factors of sovereign spreads and capital flows using extended and reduced sample: 0.98 in both cases. - Variance decompositions (Tables A.16 and A.17 in Appendix referenced): comparable to baseline. - Conclusion: empirical facts are not affected materially by Covid. *Source: wpiea2026060-source-pdf - 3.6 Robustness* ### 4.1 The Model ### 4.1 The Model ### Model setup - Two-EME/RBC framework with country index omitted except when interacting common and idiosyncratic variables. Full equilibrium conditions in Appendix C. - Households: identical, GHH preferences with utility - E0 ∑_{t=0}^∞ β^{t} [(c_t − ω^{−1} h_t^{ω})^{1−γ} −1] / (1−γ) (equation (6)) - Parameters: β (discount factor), γ (CRRA), ω (inverse Frisch elasticity). - Budget constraint (equation (7)): c_t + i_t + φ/2 (k_{t+1} − k_t)^2 + R_{t−1} d_{t−1} = y_t + d_t − ψ/2 (d_t − d)^2 - i_t: investment; φ: capital adjustment cost parameter; d_t: net external debt; R_t: gross interest rate on debt; ψ: Portfolio Adjustment Costs (PAC). - No-Ponzi constraint (equation (8)): lim_{j→∞} E_t d_{t+j} / ∏_{s=0}^{j} (1 + r_s) ≤ 0. - Production (equation (9)): y_t = A_t k_t^{α} h_t^{1−α}, α ∈ (0,1); A_t: productivity. - Capital accumulation (equation (10)): k_{t+1} = (1−δ) k_t + i_t, δ ∈ [0,1]. - Capital flows defined as change in debt stock (equation (11)): kf_t = d_t − d_{t−1}. - Equilibrium variables: {c_t, h_t, y_t, i_t, k_{t+1}, d_t, kf_t} that maximize (6) subject to (7)–(10) given exogenous processes. ### Driving forces (shocks and processes) - Shocks: international gross interest rate R^*_t, domestic interest rate R_t, total factor productivity A_t. - Productivity shocks proxy demand-side drivers of capital flows; shocks to international and domestic rates proxy supply-side forces. - Correlated shocks in R_t and A_t allowed to capture cross-country comovement. - Productivity process: ln A_t = ρ_A ln A_{t−1} + ε_At - ε_At jointly Normal across countries with variance-covariance matrix Σ_A: - [ σ_{A11}^2 σ_{A12} σ_{A21} σ_{A22}^2 ] - σ_{A12} = σ_{A21} ≠ 0 captures correlated TFP shocks. - Country interest rate: R_t = R^*_t + ε_{r t} - Country interest rate shocks jointly Normal with Σ_R: - [ σ_{11}^2 σ_{12} σ_{21} σ_{22}^2 ] - σ_{12} = σ_{21} ≠ 0 captures correlation between sovereign spreads (estimated correlation reported as 0.66). - International interest rate process: ĤR^*_t = ζ ĤR^*_{t−1} + ε^*_t - ĤR^*_t: log-deviation of R^*_t from steady state, ζ: persistence, ε^*_t ~ Normal(0, σ_{R^*}^2). - Baseline: (gross) spread s_t = R_t / R^*_t is exogenous and equal to one in steady state. Extension will allow endogenous spread dynamics. ### Calibration (quarterly, Brazil and Mexico, 1997:Q1–2019:4) - Parameters from Schmitt-Grohe and Uribe (2003), adjusted to quarterly frequency: - γ = 2 - ω = 1.455 - δ (annual) = 0.10 → quarterly δ = 0.025 (table entry) - α = 0.32 - β calibrated so that β R^* = 1 with mean international interest rate of 4% annual → R^* = 1.04 (annual); β = 0.9901 (table). - Capital and portfolio adjustment costs calibrated to match investment and trade balance-to-output volatility. - TFP shock standard deviation and persistence calibrated to match output volatility and persistence; covariance of TFP shocks set to match Brazil–Mexico output correlation. - International interest rate process estimated using U.S. quarterly data; Real TBILL proxy used (3-month gross Treasury bill rate divided by average gross inflation based on GDP Deflator over past four quarters). - Country spread shock parameters σ_{R,ii} and σ_{R,ij} calibrated to match spread volatility and spread correlation. - Key calibrated parameter values (Table 10): - γ CRRA parameter: 2 - ω Inverse Frisch elasticity: 1.455 - δ Depreciation rate: 0.025 - α Capital share: 0.32 - R^* Annual Interest rate in SS: 1.04 - β Discount factor: 0.9901 - ζ Persistence R^*_t: 0.92 - σ_{R^*} Std.Dev. of R^* shock: 0.00183 - d Debt in steady state (Average TBY): Brazil 0.38, Mexico 3.9 - φ Capital adjustment cost (Investment volatility): Brazil 0.00386, Mexico 0.00767 - ψ Portfolio adjustment costs (TBY volatility): Brazil 0.000000147, Mexico 0.0000005 - ρ_A Persistence TFP (Output persistence): Brazil 0.6355, Mexico 0.765 - σ_{A,ii} Std.Dev. TFP Shock (Output volatility): Brazil 0.00653, Mexico 0.00505 - σ_{A,ij} Covariance TFP Shocks (Output correlation): 0.3, 0.3 - σ_{R,ii} Std.Dev. Spread Shocks (Spread Volatility): Brazil 0.0083, Mexico 0.0033 - σ_{R,ij} Covariance Spread Shocks (Spread Correlation): 0.707, 0.707 ### Model evaluation (second moments; Table 11) - Theoretical unconditional second moments compared to data for Brazil and Mexico. Selected moments (Data vs Model): - std(y_t): Brazil 1.73 vs 1.73; Mexico 1.68 vs 1.68 - std(i_t): Brazil 5.49 vs 5.46; Mexico 4.40 vs 4.40 - std(tby_t): Brazil 2.50 vs 2.50; Mexico 1.63 vs 1.25 - std(kf_t / y_t): Brazil 3.16 vs 0.73; Mexico 1.14 vs 0.60 - std(R^*_t): Brazil 0.45 vs 0.45; Mexico 0.45 vs 0.45 - std(s_t): Brazil 0.83 vs 0.83; Mexico 0.32 vs 0.32 - corr(s_t, kf_t / y_t): Brazil −0.21 vs −0.10; Mexico −0.03 vs −0.11 - corr(s_t, y_t): Brazil −0.18 vs 0.00; Mexico −0.11 vs 0.00 - corr(i_t, y_t): Brazil 0.86 vs 0.80; Mexico 0.77 vs 0.79 - corr(y_t, y_{t−1}): Brazil 0.72 vs 0.72; Mexico 0.83 vs 0.83 - Cross-country moments (Data vs Model): - corr(y_BR, y_MEX): 0.30 vs 0.30 - corr(s_BR, s_MEX): 0.65 vs 0.65 - corr((kf_t / y_t)_BR, (kf_t / y_t)_MEX): 0.35 vs 0.34 - Assessment: - The model matches business cycle dynamics and comovement: high comovement of country spreads and lower comovement of net capital flows across countries. - Limitations: model delivers net capital flows that are less volatile and more persistent than in the data; likely due to absence of richer portfolio decision mechanisms. ### Impulse responses (summary; Figure 9 discussion) - Responses for Brazil to one S.D. shocks: - TFP (domestic) shock: immediate increase in capital flows to Brazil, followed by a protracted decrease; initial rise driven by investment (higher marginal productivity of capital), subsequent rise in savings to smooth consumption; spread falls protractedly associated with increased savings. - Domestic spread shock (increase one S.D.): raises country interest rate → current account improves via capital outflows; investment falls due to higher marginal cost of capital. - International interest rate shock (one S.D. increase): shifts credit supply curve upward → country spread gradually increases via weaker business cycle conditions; capital outflows as higher country interest rate reduces investment. - Cross-country transmission: - An increase in Mexico’s spread by one S.D. raises Brazilian spread considerably and induces capital outflows from Brazil due to matched spread correlation. - Correlated TFP shocks imply a transitory improvement in Mexico raises Brazilian TFP (smaller magnitude) with similar real effects. ### Assessing determinants of synchronization (counterfactual experiments; Table 12) - Three counterfactuals simulated; compute cross-country correlation of income, spreads, and capital flows. Columns: Data | Baseline | Exp.1 | Exp.2 | Exp.3 - Baseline reproduces calibration results. - Exp. 1: turn off correlation of TFP shocks (σ_{A,ij} = 0) → gauges common credit demand drivers. - Exp. 2: set correlation of country spread shocks to zero (σ_{R,ij} = 0) → removes main source of common credit supply shocks. - Exp. 3: set correlation of country spread shocks equal to that of TFPs (σ_{R,ij} = σ_{A,ij}). - Results (Table 12): - corr(y_BR, y_MEX): Data 0.30 | Baseline 0.30 | Exp.1 0.00 | Exp.2 0.30 | Exp.3 0.30 - corr(s_BR, s_MEX): Data 0.65 | Baseline 0.65 | Exp.1 0.65 | Exp.2 0.00 | Exp.3 0.30 - corr(kf_BR, kf_MEX): Data 0.35 | Baseline 0.34 | Exp.1 0.08 | Exp.2 0.28 | Exp.3 0.33 - Key findings: - Correlated TFP shocks (business cycle synchronization) are key for capital flows comovement: turning off TFP correlation reduces corr(y_BR, y_MEX) from 0.30 to 0.00 and corr(kf_BR, kf_MEX) from 0.34 to 0.08. - Correlated spread shocks (Global Financial Cycle) drive spread synchronization: setting σ_{R,ij} = 0 reduces corr(s_BR, s_MEX) from 0.65 to 0.00 with only limited effects on capital flows and income correlations. - When σ_{R,ij} is set equal to σ_{A,ij} (Exp.3), income correlation unchanged (0.30) and capital flow correlation declines marginally from 0.34 to 0.33. - Conclusion: high correlation in country spreads across EMEs primarily reflects common credit supply shocks (Global Financial Cycle), whereas synchronization of net capital flows across EMEs is driven mainly by common TFP (credit demand) shocks. Common supply shocks affect spreads strongly but have limited quantitative impact on capital flow synchronization. *Source: wpiea2026060-source-pdf - 4.1 The Model (IMF).* ### 4.7 Extensions of the Theoretical Model ### 4.7 Extensions of the Theoretical Model ### Estimated Interest Rate Processes - Extension: estimate an interest rate process for each economy following Uribe and Yue (2006), allowing for correlated sovereign spread shocks across the two economies. - Country interest rate process (variables with ^ denote log-deviations; tby i,t is trade balance-to-output ratio; εr_t is country interest rate shock): - ˆR_t = ρ_R ˆR_{t−1} + ρ_{R*} ˆR*_{t} + ρ_{R1*} ˆR*_{t−1} + ρ_y ˆy_{t} + ρ_{y1} ˆy_{t−1} + ρ_i ˆı_{t} + ρ_{i1} ˆı_{t−1} + ρ_{tby} tby_{t} + ρ_{tby1} tby_{t−1} + ε_{r_t} - Spread definition: s_t = R_t / R*_t. - Calibration: remains unchanged relative to Table 10; estimated interest rates consistent with Uribe and Yue (2006). Theoretical moments remain almost unchanged; detailed results in Appendix C.3.1. - Decomposition: redid the decomposition of frictions in explaining synchronization of capital flows (see Section 4.6); Table 13 reports unconditional moments from simulated data of the two-EME theoretical model augmented with the estimated interest rate process. - Unconditional moments (Data, Baseline, Exp. 1, Exp. 2, Exp. 3): - corr(y_BR, y_Mex): 0.30, 0.30, 0.00, 0.30, 0.30 - corr(s_BR, s_MEX): 0.65, 0.65, 0.64, 0.24, 0.30 - corr(kf_BR, kf_MEX): 0.35, 0.39, 0.18, 0.37, 0.37 - Notes on experiments: - Baseline: baseline calibration described in Section 4.3. - Exp. 1: assumes σ_A_{i,j} = 0 (no correlation between productivity shocks). - Exp. 2: assumes σ_R_{i,j} = 0 (no correlation between country interest rate shocks). - Exp. 3: assumes σ_R_{i,j} = σ_A_{i,j} (correlation between sovereign spreads equals that of output). - Main findings under this extension: - Business cycle synchronization accounts for more than half of capital flows comovement. - Correlated spread shocks explain little of capital flows comovement across the two economies. - Spreads comove even without correlated spread shocks because the estimated interest rate processes make interest rates respond to business cycle conditions. - Capital flows are mostly explained by credit demand shocks. ### Credit Risk Shocks - Motivation: baseline model underpredicts the volatility of capital flows; three modifications introduced to improve performance on this dimension. - Modifications: 1. Replace Portfolio Adjustment Costs with the Internal Debt Elastic Interest rate closing device (Schmitt-Grohe and Uribe, 2003): the interest rate depends on the stock of external debt. 2. Households internalize that their debt decisions affect the interest rate when choosing consumption and saving, affecting capital flows. 3. Allow the elasticity of the interest rate with respect to external debt to be stochastic ("credit risk shocks"). - Country interest rate process in this version: - R_t = R*_t + ψ_t ((d_t − ̄d)^{-1}) + ε_{r_t} - ψ_t = ψ + ε_{ψ_t} - ψ is the steady state value of ψ_t; ε_{ψ_t} is a credit risk shock to the elasticity of the interest rate with respect to the level of debt. - Mechanism: - Credit risk shocks shift the slope of the credit supply curve (as opposed to parallel shifts from international interest rate and spread shocks), amplifying impacts on flows. - Households internalizing debt effects adds additional variability in capital flows. - Calibration strategy: - Calibrate correlation of ε_{ψ_t} across countries to match comovement of capital flows. - Calibrate ψ in steady state to match trade balance-to-output ratio in steady state. - Calibrate variance of ε_{ψ_t} to match volatility of capital flows to output. - Other parameters remain as in baseline calibration. - Outcomes and trade-offs: - The model can match the volatility of capital flows (Appendix Table A.25 displays moments). - Introducing credit risk shocks induces excess volatility in investment associated with capital flows. - Correlation between capital flows and country spreads at the country level remains similar to baseline. - Comovement between capital flows across countries drops from 0.35 to 0.03. - Overall, adding these features improves matching of capital flow volatility but entails a considerable trade-off in model performance along other dimensions. ### Solving the Model with Global Methods - Motivation: baseline solved with first-order perturbation methods; literature (de Groot et al., 2019) shows perturbation vs fixed-point global solutions can yield different moments for small open economy models with incomplete markets, particularly for Net Foreign Asset position. - Approach: - Solve baseline model using fixed-point-iteration global solutions (FiPIt algorithm from Mendoza and Villalvazo, 2020), which uses fixed-point iteration on Euler equations. - Two cases solved: 1. Baseline including Portfolio Adjustment Costs to avoid unit root in NFA. 2. Set β(1 + r*) < 1 and drop Portfolio Adjustment Costs (not necessary in this case). - Parametrization kept as close as possible to original; some parameters recalibrated following Section 4.3 strategy. - Appendix C.3.1 displays calibrated parameters, simulation details, and theoretical moments. - Main takeaway: - Performance improves in some moments, notably the volatility of capital flows. - Capital flow volatility in Brazil doubles compared to baseline results, reducing the gap with data. - For Mexico, implied volatility of capital flows also doubles and now matches the data. - Explanation: baseline linearized model yields Portfolio Adjustment Costs values close to zero, imposing a limit on capital flow volatility without a unit root in debt; the global solution does not suffer this constraint and can better account for volatility. - Cost: cross-country correlation in capital flows increases to 0.63, above that in the data (0.35). ### Conclusions (from section 5 as it relates to model extensions) - Access to world capital markets by EMEs is historically volatile. - Findings summarized: - Country spreads faced by EMEs relative to world interest rates are predominantly determined by global supply forces, which can be traced back to US monetary policy. - At those prices, idiosyncratic factors play an important role in explaining fluctuations in capital flows. - A calibrated structural model with correlated productivity and country interest rate shocks matches several dynamics: - The correlation between productivity shocks explains around half of capital flows comovement while it does not affect the correlation of sovereign spreads. - The correlation between interest rate shocks explains around two thirds of fluctuations in spreads while they do not affect significantly the observed correlation of net capital flows. - Research implication: exploring policy implications and how the decomposition of drivers of capital flows and prices can inform policy responses to capital flow volatility is a promising area for further work. *Source: 4.7 Extensions of the Theoretical Model (wpiea2026060-source-pdf)* ### References ### References and Appendix (wpiea2026060-source-pdf - References) ### Major bibliographic themes - Extensive literature on global financial conditions, capital flows, sovereign spreads, and the global financial cycle. - Key recurring topics and methods: - Country spreads, sovereign risk, and determinants of sovereign bond spreads. - Gross and net capital flows dynamics, surges, stops, flight, and retrenchment. - Role of U.S. monetary policy and the Global Financial Cycle in emerging markets. - Model solutions and open-economy macroeconomics with incomplete markets and occasionally binding constraints. - Empirical tools: factor models, structural vector autoregressions, Hodrick-Prescott filtering, cohesion/comovement measures. ### Data sources and empirical setup (Appendix A.1) - Capital flows series computed using equation (3) (as specified in the main text). - Variables and sources: - X: exports in current USD FOB at monthly frequency. Source: International Financial Statistics, IMF. - M: imports in current USD CIF at monthly frequency. Source: International Financial Statistics, IMF. - R: stock of foreign reserves in current usd at monthly frequency. Source: International Financial Statistics, IMF. - Real expression: series of capital flows are expressed in real terms using the U.S. Producer Price Index by Commodity: All Commodities (PPIACO) available at FRED. - Country spreads proxy: monthly average of the JP Morgan Emerging Market Bond Index Global (Stripped Spread). - Filtering and detrending: - Hodrick-Prescott filter with λ = 1600 applied to real output, real investment, and trade balance-to-output (logs) to extract cyclical components. - For interest rate processes, a log quadratic trend is extracted from real output and real investment following Uribe and Yue (2006) before estimating the process. - Calibration data for the two-EME model: moments for Brazil and Mexico from International Financial Statistics (IFS) and national accounts where IFS lacked data. Output and investment downloaded in domestic currency nominal terms; trade balance downloaded in domestic currency real terms. GDP deflator used to deflate output and investment. Trade balance-to-output ratio computed as trade balance (real) divided by output (real). ### Identification of Sudden Stops (Appendix B.1 and Table A.1) - Sudden Stops defined following Calvo et al. (2008), with three conditions: - i) at least one observation where year-on-year decline in capital flows is at least two standard deviations below its sample mean (the ‘unpredicted’ prerequisite); - ii) the sudden stop phase ends when annual change in capital flows surmounts one standard deviation below its sample mean; - iii) onset ascertained by the first time annual change drops one standard deviation below the mean. - First and second moments of capital flow series calculated each period using an expanding window with a minimum of 24 (months of) observations and a start date fixed at January 1997. - Table A.1: List of Systemic Sudden Stop Episodes for the baseline sample (1997m1 to 2022m7). Examples (selected): - Argentina: 2000m11 to 2002m10; 2018m9 to 2020m8 - Brazil: 1999m1 to 1999m8; 2008m7 to 2009m7 - China: 2005m12 to 2007m1; 2008m10 to 2009m8; 2011m12 to 2013m3; 2014m9 to 2016m5 - Mexico: 2009m3 to 2009m12; 2015m6 to 2016m3; 2019m11 to 2021m7 - Panama: 1998m2 to 1998m5; 2015m9 to 2016m8; 2021m4 to 2022m6 - Philippines: 1997m6 to 1999m7; 2000m3 to 2001m4; 2012m3 to 2013m2 - Poland: 1997m1 to 1997m7; 1999m3 to 2000m9; 2008m11 to 2009m9; 2012m2 to 2012m8 - South Africa: 1997m1 to 1997m2; 2008m8 to 2009m9; 2010m8 to 2011m9; 2020m9 to 2021m10 - Turkey: 1998m10 to 1999m9; 2001m6 to 2002m3; 2008m10 to 2009m12; 2012m4 to 2012m9; 2014m3 to 2016m4; 2018m10 to 2019m8 ### Cross-country comovement and cohesion findings (Appendix B.2) - Low correlation of net capital flows across EMEs documented (net capital flows defined as sum of change in foreign reserves plus the difference between imports and exports). - Component analysis: - The change in foreign reserves is more correlated across EMEs than the trade-balance-related component. - However, the maximum correlation is around 0.4, which is significantly lower than the corresponding number for country spreads. - Implication: Both components (change in foreign reserves and trade balance) are important to explain the relatively lower cross-EME correlation of capital flows. ### Intertemporal correlations between capital flows and sovereign spreads (Appendix B.1 and Table A.2) - Intertemporal correlation between capital flows () and EMBI() reported for Argentina, Brazil, China, Colombia, Ecuador, Malaysia, Mexico, Panama, Philippines, Poland, South Africa and Turkey for the period 1997:1-2022:7. - Capital flows defined as cumulative trade deficit plus change in international reserves at monthly frequency. - Significance notation used consistently: - ∗∗∗ p < 0.01 - ∗∗ p < 0.05 - ∗ p < 0.1 - Selected examples from Table A.2 (intertemporal correlations ρ(s,f) at various lags/leads; values preserved as presented): - ρ(s,f)L6: Argentina 0.12 ∗∗; Brazil 0.01; China −0.08; Colombia −0.09; Ecuador 0.07; Mexico 0.12 ∗∗; Poland 0.00; Turkey −0.13 ∗∗ - ρ(s,f)L1: Argentina 0.03; Brazil −0.08; China −0.18 ∗∗∗; Colombia −0.10 ∗; Ecuador −0.10 ∗; Mexico −0.03; Philippines −0.02; Poland −0.08; South Africa −0.10 ∗; Turkey −0.14 ∗∗ - ρ(s,f)F0: Argentina 0; Brazil −0.12 ∗∗; China −0.20 ∗∗∗; Colombia −0.13 ∗∗; Ecuador −0.11 ∗; Mexico −0.07; Panama −0.03; Philippines −0.10 ∗; Poland −0.15 ∗∗; South Africa −0.15 ∗∗∗; Turkey −0.16 ∗∗∗ - Overall pattern: negative and low contemporaneous correlation between country spreads and capital flows, which persists using leads and lags. ### Dynamic factors and variance decomposition (Appendix B.3) - Dynamic factors: - Figure A.3 shows the common dynamic factors for country spreads and capital flows in levels (non-cumulated) for Argentina, Brazil, China, Colombia, Ecuador, Malaysia, Mexico, Panama, Philippines, Poland, South Africa and Turkey for the period 1997:2-2022:7. - Figure 4 in main text uses cumulated factors; Appendix shows non-cumulated factors for comparison. - Variance decomposition: - Baseline DFM (Eq.5) identified with sign restrictions used to compute one-year ahead variance decomposition to capture dynamic contributions of each shock. - Alternative measure: variance decomposition on impact (closer to R2 measures used in previous literature). - Tables A.3 and A.4 display variance decomposition of country spreads and net capital flows driven by the four credit market shocks; results described as very similar to baseline ones. ### Key technical and sample details - HP filter parameter: λ = 1600. - Expanding-window minimum for moments: 24 (months). - Baseline sample period referenced: 1997:1–2022:7 (and 1997:2–2022:7 for some figures). - Countries analyzed (selected lists repeated across appendices and figures): Argentina, Brazil, China, Colombia, Ecuador, Malaysia, Mexico, Panama, Philippines, Poland, South Africa, Turkey. *Italic: Content derived from "References" and appendices A–B of wpiea2026060-source-pdf.* ### conclusions regarding the importance of external vs idiosyncratic shocks remains unchanged. ### wpiea2026060-source-pdf - conclusions regarding the importance of external vs idiosyncratic shocks remains unchanged. ### Major empirical findings on drivers of spreads and capital flows - While Global and common EME shocks are equally important in accounting for fluctuation in country spreads, Global credit shocks are relatively more important than common EME credit shocks in explaining capital flows fluctuations. - Country spread dynamics are largely explained by credit supply shocks; capital flow dynamics are largely explained by credit demand shocks. ### Variance decomposition — country spreads (selected on-impact and one-year ahead results) - On impact (baseline DFM, median shares for country spreads): - ε_S,G_t (common credit supply shock): 64 (median) - ε_D,G_t (common credit demand shock): 36 (median) - ε_S,G_t + ε_D,G_t (total common shocks): 100 (median) - ε_S,I_t (country-specific credit supply shock): 18 (median) - ε_D,I_t (country-specific credit demand shock): 18 (median) - ε_S,I_t + ε_D,I_t (total idiosyncratic shocks): 36 (median) - Extended sample including advanced economies (one-year ahead, Table A.5, median shares): - ε_S,G_t: 20 (median) - ε_D,G_t: 17 (median) - ε_S,EME_t: 18 (median) - ε_D,EME_t: 15 (median) - Total External (ε_S,G_t + ε_D,G_t + ε_S,EME_t + ε_D,EME_t): 70 (median) - ε_S,I_t: 15 (median) - ε_D,I_t: 15 (median) - ε_S,I_t + ε_D,I_t: 30 (median) - Alternative DFM with persistence in idiosyncratic innovations (one-year ahead, Table A.7, median shares): - ε_S,G_t: 36 (median) - ε_D,G_t: 22 (median) - ε_S,G_t + ε_D,G_t: 58 (median) - ε_S,I_t: 21 (median) - ε_D,I_t: 21 (median) - ε_S,I_t + ε_D,I_t: 42 (median) - Excluding Covid (one-year ahead, Table A.16, median shares): - ε_S,G_t: 39 (median) - ε_D,G_t: 22 (median) - ε_S,G_t + ε_D,G_t: 61 (median) - ε_S,I_t: 20 (median) - ε_D,I_t: 18 (median) - ε_S,I_t + ε_D,I_t: 38 (median) ### Variance decomposition — capital flows (selected on-impact and one-year ahead results) - On impact (baseline DFM, median shares for net capital flows, Table A.4): - ε_S,G_t: 64 (median) - ε_D,G_t: 3 (median) - ε_S,G_t + ε_D,G_t: 8 (median) - ε_S,I_t: 46 (median) - ε_D,I_t: 44 (median) - ε_S,I_t + ε_D,I_t: 90 (median) - Extended sample including advanced economies (one-year ahead, Table A.6, median shares): - ε_S,G_t: 29 (median) - ε_D,G_t: 9 (median) - ε_S,EME_t: 1 (median) - ε_D,EME_t: 1 (median) - Total External (sum of common global and EME shocks): 20 (median) — (Table reports "Total External 10010012759154291801129251920" formatting for columns; median interpreted as 20) - ε_S,I_t + ε_D,I_t: 80 (median) - Alternative DFM with persistent idiosyncratic innovations (one-year ahead, Table A.8, median shares): - ε_S,G_t: 7 (median) - ε_D,G_t: 5 (median) - ε_S,G_t + ε_D,G_t: 12 (median) - ε_S,I_t: 44 (median) - ε_D,I_t: 44 (median) - ε_S,I_t + ε_D,I_t: 88 (median) - Excluding Covid (one-year ahead, Table A.17, median shares): - ε_S,G_t: 8 (median) - ε_D,G_t: 3 (median) - ε_S,G_t + ε_D,G_t: 8 (median) - ε_S,I_t: 48 (median) - ε_D,I_t: 44 (median) - ε_S,I_t + ε_D,I_t: 92 (median) ### Extensions and robustness checks of the empirical DFM - Extended DFM (including Australia, Canada, Denmark, Finland, Germany, Italy, Japan, Norway, UK, US, Switzerland) enables identification of: - Global credit shocks (affect all economies), - Common EME credit shocks (affect EMEs), - Idiosyncratic credit shocks. - Regional factors (DFM with regional factor FR,t, equation (13)): - Latin American regional factor includes Argentina, Brazil, Colombia, Ecuador, Mexico, Panama. - Asian regional factor includes China, Malaysia, Philippines, Turkey. - Regional cumulated factors capture known events (e.g., Asian spreads spike in 1997 related to the Asian crisis). - Alternative datasets and measures: - Correlation between capital flows and country spreads is consistently negative and low across robustness datasets (Table A.9). - Pre-Covid sample results are almost unchanged; correlation between factors including and excluding Covid is 0.98. - Net FDI vs portfolio flows: - FDI capital flows are almost entirely explained by idiosyncratic shocks (Table A.12: ε_S,I_t + ε_D,I_t = 99 median for FDI inflows). - Portfolio flows produce contributions similar to net capital flows (Table A.14). ### Theoretical model — key analytical results - Two-period model (Section C.1) — default probability and equilibrium interest rate: - Probability of default π_i = d_i1 / [(1 + r_i) φ y_i2] (equations (14)-(15)). - Credit supply has positive slope; approximation used in Section 2: r_i = r* + φ ( ̃d_i1 )^2 + ε_i, φ > 0, where ε_i captures credit supply shocks. - Two-EME DSGE model (Sections C.2–C.3) — equilibrium moments and decompositions: - Model calibration targets and moments shown for Brazil and Mexico (Tables A.18, A.23, A.25, A.27, A.29 depending on specification). - Baseline theoretical one-year ahead variance decomposition of spreads (Table A.19, model): - ε_S,G_t: 41 (Brazil model), 34 (Mexico model) - ε_D,G_t: 23 (Brazil model), 03 (Mexico model) - ε_S,G_t + ε_D,G_t: 64 (Brazil model), 37 (Mexico model) - ε_S,I_t + ε_D,I_t: 36 (Brazil model), 61 (Mexico model) - Baseline theoretical one-year ahead variance decomposition of capital flows (Table A.20, model): - Brazil model: ε_S,G_t + ε_D,G_t = 51; ε_S,I_t + ε_D,I_t = 49 - Mexico model: ε_S,G_t + ε_D,G_t = 30; ε_S,I_t + ε_D,I_t = 70 - Credit supply shocks are the main driver of sovereign spreads in the model; credit demand shocks (notably idiosyncratic TFP-driven credit demand) are the main driver of capital flows: - Credit demand shocks explain up to 78% (Brazil) and 83% (Mexico) of capital flows variability in one formulation; idiosyncratic credit demand shocks explain 63% (Brazil) and 59% (Mexico) of capital flows fluctuations in that decomposition. - Model variations: - Adding an estimated interest-rate process (Table A.21) and calibrating accordingly (Table A.22) produces comparable moments (Table A.23). - Including credit risk shocks (affecting slope of credit supply) improves match to credit market moments (Tables A.24–A.25). - Solving with global methods (FiPit) and two alternative formulations (portfolio adjustment costs; β(1 + r) < 1) produces broadly consistent moments; global methods increase correlation of capital flows between BRZ and MEX relative to baseline but main qualitative result (spreads driven by credit supply, capital flows by credit demand) remains. ### Interpretation and synthesis - Empirical and theoretical results consistently indicate a disconnect in sources of comovement: - High comovement of country spreads across EMEs driven by common/global credit supply shocks (Global Financial Cycle). - Low comovement of capital flows across EMEs driven by country-specific (idiosyncratic) credit demand shocks. - Robustness: results hold across alternative DFM specifications, inclusion of advanced economies, persistence in idiosyncratic innovations, regional factors, alternative capital-flow datasets, and pre-Covid samples. *Source: Content from wpiea2026060-source-pdf - conclusions regarding the importance of external vs idiosyncratic shocks remains unchanged.* --- _Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026060-source-pdf.pdf_