## _wp15163

## Source details

**Canonical URL:** [_wp15163](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15163.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15163.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15163.pdf.json)

---

### I. Research question and main empirical findings
- Research question: whether private non-FDI capital flows amplify or dampen economic cycles in Low-Income Developing Countries (LIDCs) and whether the cyclicality differs from Emerging Markets (EMs).
- Dataset: new dataset constructed by Araujo et al. (2015).
- Main empirical findings:
  - Private capital inflows are procyclical in general.
  - Private capital inflows are less procyclical in LIDCs relative to EMs.
  - Flows to LIDCs are more persistent than flows to EMs.
  - Changes in risk aversion are a significant correlate of private capital flows in most specifications with the expected sign.
  - Flows to LIDCs tend to be less sensitive to changes in global risk aversion compared to the full sample.
  - Trade openness is positively associated with private flows.
  - Changes in the terms of trade are negatively related to flows.
  - Capital inflows are positively associated with trend growth, but these associations are weaker for LIDCs relative to EMs.
- Robustness and estimation:
  - Results robust to alternative estimation methods, including GMM techniques and Blundell-Bond system GMM estimator.
  - Conclusions hold across alternative samples, control variables, and alternative measures of flows (including adding FDI to gross private capital flows); net flows are less related to the cycle.

### II. Theoretical context and expectations
- Theoretical possibilities:
  - Capital flows could be procyclical, countercyclical, or acyclical depending on model set-up and frictions.
  - In frictionless endowment open-economy models, net flows are predicted to be countercyclical to supply shocks but can be procyclical to demand shocks (Obstfeld and Rogoff, 1996; Vegh, 2013).
  - Portfolio-growth and portfolio-reallocation effects can generate countercyclical gross inflows in some DSGE portfolio-choice models (Van Wincoop and Tille, 2010).
  - Financial frictions (e.g., pecuniary externalities) can generate procyclical net inflows and amplification of shocks (Bianchi, 2011; Brunnermeier, Eisenbach and Sannikov, 2012).
- Relevance for LIDCs:
  - LIDCs may be more credit constrained, suggesting potentially more procyclical flows, but smaller banking sectors and lower leverage could imply weaker financial-accelerator amplification and thus less procyclicality.
  - Existing models are not explicitly tailored to LIDCs; development of LIDC-specific models is an important avenue for further research.

### III. Data, measurement, and sample
- Coverage and period:
  - Dataset contains information for 58 LIDCs and 92 emerging markets for the period from 1990 to 2012.
- Key series and construction:
  - Non-FDI private inflows = portfolio investments liabilities + other investment liabilities + financial derivatives; excludes official liabilities.
  - Compilation follows IMF Balance of Payment Statistics (BOPS) and BPM5; gaps closed using IMF World Economic Outlook (WEO) database.
  - Focus on gross flows rather than net flows.
- Sample restrictions:
  - Fragile and small countries excluded from the core sample (sample restricted to non-fragile and non-small LIDCs).

### IV. Stylized facts on inflows and surge analysis
- Shares and levels:
  - Share of non-FDI private inflows in total flows is increasing for LIDCs and approaching EM levels.
  - Non-FDI inflows to LIDCs are on average considerably lower than inflows to EMs.
  - Inflows to the top quartile of LIDCs as a share of GDP are comparable to the median inflows in EMs.
  - After the crisis, inflows to the top quartile of LIDCs converged to the top quartile of EMs.
- Surge analysis:
  - Sample for surge analysis: 95 countries and 2034 observations.
  - Total surges detected: 296 surges.
  - Surges detected in LIDCs: 77 surges out of 296 total.
  - Waves of increased share of surges detected in: 1990-1994, 1996-1997, 2004-2008 and 2010-2012.
  - By 2007, surges spread and reached close to 40 percent of the developing countries in the sample.
  - After a decline following the global financial crisis, the number of surges in LIDCs rebounded in 2012.
- Surge definition:
  - A surge for a given country satisfies both criteria:
    - (i) gross non-FDI private inflows as percent of GDP are in the top quartile of its own country sample (time dimension);
    - (ii) gross non-FDI private inflows as percent of GDP are in the top quartile of the cross-country sample.

### V. Measurement of cyclical variables and first-pass evidence
- Cyclical decomposition:
  - Hodrick-Prescott (HP) filter applied to the log of Real GDP series at constant 2005 national prices from Penn World Table, Version 8.0.
  - HP smoothing parameter: 6.25 (Ravn and Uhlig (2002) suggestion for annual data); alternative λ = 1 and Christiano and Fitzgerald filter also tested.
  - Variations in smoothing parameters and use of Christiano and Fitzgerald filter do not substantially affect output gap estimates.
- First-pass reduced-form associations:
  - Positive and significant association between gross private capital inflows (excluding FDI) as a share of trend/potential GDP and the cyclical component of output in baseline specifications.
  - Including growth of potential output alongside the output gap still yields a positive and significant association.
  - Net flows are less related to the cycle than gross flows.

### VI. Estimation framework, controls, and inference
- Estimable specification:
  - c f_{i,t} = α_i + ρ c f_{i,t−1} + β \bar{y}_{i,t} + ∑_{m=1}^{M} δ_m X_{m,i,t} + λ_t + ε_{i,t}
    - c f_{i,t} = private capital flows measure as a share of potential GDP.
    - \bar{y}_{i,t} = output gap (cyclical component of GDP).
    - X_{m,i,t} = control variables.
- Estimators and inference:
  - Primary estimator: fixed-effects with Driscoll and Kraay (1998) corrected standard errors to account for cross-sectional dependence.
  - Additional estimator: system (Blundell-Bond) GMM estimator to handle endogeneity using lagged values as instruments (Roodman, 2009).
  - CCEMG estimator (Pesaran, 2006) used to model common risk factors and allow heterogeneity.
  - Driscoll-Kraay implemented in Stata 13 using code by Daniel Hoechle.
  - Open issue: finding valid external instruments (beyond lagged values) for the output gap remains unresolved.
- Control variables (selection and rationale):
  - Global (“Push”) factors: VIX/VXO implied volatility index (investor risk appetite), measures of global liquidity, changes in net interoffice assets of foreign banks in the US, commodity prices/terms of trade.
  - Country-specific (“Pull”) factors: lagged dependent variable; financial development/leverage (stock market capitalization; credit to private sector for LIDCs); de jure measures of capital controls (Chinn-Ito); institutional quality (ICRG); inflation; trade openness.
  - Endogeneity caveats noted for financial depth and de jure openness measures.

### VII. Baseline regression results (selected exact coefficients and diagnostics)
- Table 1 (Fixed Effects, Driscoll-Kraay SEs) — Full sample, column 1:
  - Lagged Dependent Variable: 0.229*** [0.073]
  - Output Gap: 0.296** [0.107]
  - ∆(VXO): -1.457*** [0.447]
  - Observations: 1,323
  - Number of countries: 75
  - R-2 (within): 0.180
- Table 2 (GMM) — Full sample, column 1:
  - Lagged Dependent Variable: 0.320*** [0.065]
  - Output Gap: 0.328*** [0.099]
  - ∆(VXO): -3.829*** [1.156]
  - Trade Openness: 2.883* [1.475]
  - Observations: 1,323
  - Number of countries: 75
  - Hansen test: 51.635
  - Sargan test: 161.6***
- Table 3 (Countries with Surges) — Full sample, column 1:
  - Lagged Dependent Variable: 0.224*** [0.073]
  - Output Gap: 0.297** [0.110]
  - ∆(VXO): -1.664*** [0.430]
  - Observations: 1,158
  - Number of countries: 67
  - R-2 (within): 0.183
- Table 4 (CCEMG) — Full sample, column 1:
  - Lagged Dependent Variable: 0.203*** [0.041]
  - Output Gap: 0.239*** [0.067]
  - Common factor linked to capital flows: 0.646*** [0.091]
  - Root Mean-Squared Error (sigma): 3.25
  - Observations: 1,223
  - Countries: 63
- Table 5 (Including Trend Growth) — Full sample, column 1:
  - Lagged Dependent Variable: 0.229*** [0.072]
  - Output Gap: 0.265** [0.108]
  - Trend Growth: 0.196** [0.069]
  - ∆(VXO): -1.328*** [0.411]
  - Observations: 1,323
  - Number of countries: 75
  - R-2 (within): 0.184
- Table 6 (Financial Reform Index) — Full sample:
  - Lagged Dependent Variable: 0.197** [0.077]
  - Output Gap: 0.261*** [0.066]
  - Financial Reform Index: 0.441*** [0.102]
  - Observations: 706
  - Number of countries: 53
  - R-2 (within): 0.218
- Table 7 (Alternative Dependent Variables) — examples:
  - Lagged Dependent Variable: 0.644*** (LIDCs col 1) [0.050]; 0.669*** (Full col 2) [0.045]
  - ∆(VXO): 0.697*** (LIDCs col 1) [0.087]; 0.429** (Full col 2) [0.173]
  - Observations vary by specification (e.g., 367; 1,321)
  - R-2 (within) examples: 0.512 (LIDCs col 1), 0.495 (Full col 2)
- Additional summary:
  - Median amplitude (difference inflows as share of potential GDP in good vs bad times): LIDCs: 0.4; EMs: 1.2.
  - Unconditional correlations (1990-2012): flows more procyclical in EMs relative to LIDCs.

### VIII. Robustness and extensions (selected outcomes)
- Robustness exercises include: alternative estimators (fixed-effects, Blundell-Bond system GMM), alternative samples (countries with surges), risk-on/risk-off regimes (CCEMG), disentangling permanent vs transitory shocks (including trend growth), alternative controls (institutional measures, leverage), alternative global variables (US interest rates, commodity price indices), alternative dependent variables (FDI, gross+FDI, net flows).
- Key robustness outcomes:
  - Positive association between gross private non-FDI inflows and the output gap persists across methods; coefficient smaller for LIDCs (weaker procyclicality).
  - Lagged dependent variable coefficients higher for LIDCs (greater persistence).
  - VXO (or other global risk measures) significant for full sample and EMs; association weaker in LIDCs.
  - Trade openness positive and significant across many specifications.
  - Financial openness matters for EMs, not for LIDCs in most specifications.
  - Net flows are less related to the cycle; adding FDI increases persistence and can affect cyclical coefficients.

### IX. Mechanisms, interpretation, and policy implications
- Mechanisms suggested:
  - Smaller banking sectors and lower leverage in LIDCs reduce amplification via financial accelerator, leading to less procyclicality.
  - LIDCs’ reliance on bank flows and trade finance (versus EMs’ tradable securities and portfolio flows) may explain lower procyclicality.
- Policy implications:
  - As LIDCs develop, private capital inflows are likely to become more procyclical.
  - Policies to manage capital inflows and mitigate destabilizing effects linked to procyclicality may become more relevant for LIDCs.
  - Prudential toolkit and financial-stability policies informed by EM experience (Ostry et al., 2012) would be increasingly useful.
  - Procyclical and volatile capital flows challenge traditional monetary policy frameworks in LIDCs based on growth-rate targets for monetary aggregates; reinforces need to accelerate adoption of modern forward-looking monetary policy frameworks (IMF, 2014b).

### X. Conclusions and avenues for future research
- Empirical conclusions (1990-2012):
  - Market-driven capital inflows to LIDCs are less procyclical and more persistent than inflows to EMs.
  - LIDCs are less sensitive to changes in global risk aversion relative to EMs.
  - Findings robust across estimation methods, samples, and controls.
- Suggested future research:
  - Investigate the role of banking sector size and economy-wide leverage in shaping cyclicality.
  - Examine the influence of types of financial instruments available to international investors.
  - Identify precise mechanisms driving lower procyclicality and greater persistence in LIDCs.

*Source: _wp15163 (IMF working paper content provided in the supplied PDF excerpt).*

### References _______________________________________________________________________________________  24

### _wp15163 - References _______________________________________________________________________________________  24

### I. Introduction — research question and main findings
- Research question: whether private non-FDI capital flows amplify or dampen economic cycles in low-income developing countries (LIDCs) and whether the cyclicality differs from emerging markets (EMs).
- Dataset used: new dataset constructed by Araujo et al. (2015).
- Main empirical findings:
  - Private capital inflows are procyclical in general, in line with the existing empirical literature.
  - Private capital inflows are less procyclical in LIDCs relative to EMs.
  - Flows to LIDCs are more persistent than flows to EMs.
  - Changes in risk aversion are a significant correlate of private capital flows in most specifications with the expected sign.
  - Flows to LIDCs tend to be less sensitive to changes in global risk aversion compared to the full sample.
  - Trade openness is positively associated with private flows.
  - Changes in the terms of trade are negatively related to flows.
  - Capital inflows are positively associated with trend growth, but these associations are weaker for LIDCs relative to EMs.
- Robustness and estimation methods:
  - Results are robust to alternative estimation methods, including GMM techniques and Blundell-Bond system GMM estimator.
  - Conclusions hold across alternative samples, control variables, and alternative measures of flows (including adding FDI to gross private capital flows), although net flows are less related to the cycle.

### II. Theoretical context and expectations
- Theoretical possibilities:
  - Capital flows could be procyclical, countercyclical, or acyclical depending on model set-up and frictions.
  - In frictionless endowment open-economy models, net flows are predicted to be countercyclical to supply shocks but can be procyclical to demand shocks (Obstfeld and Rogoff, 1996; Vegh, 2013).
  - Portfolio-growth and portfolio-reallocation effects can generate countercyclical gross inflows in some DSGE portfolio-choice models (Van Wincoop and Tille, 2010).
  - Financial frictions (e.g., pecuniary externalities) can generate procyclical net inflows and amplification of shocks (Bianchi, 2011; Brunnermeier, Eisenbach and Sannikov, 2012).
- Relevance for LIDCs:
  - LIDCs may be more credit constrained, suggesting potentially more procyclical flows, but smaller banking sectors and lower leverage could imply weaker financial-accelerator amplification and thus less procyclicality.
  - Existing models are not explicitly tailored to LIDCs; development of LIDC-specific models is an important avenue for further research.

### III. Existing empirical literature
- Stylized empirical facts from prior work:
  - Net capital flows to EMs are widely documented to be procyclical (Kaminsky, Reinhart and Vegh, 2005).
  - Gross capital inflows expand during good times and decline during recessions (Broner et al., 2013).
  - International portfolio investments by institutional investors are highly procyclical relative to global macroeconomic and financial conditions (Puy, 2013), with stronger procyclicality in EMs.
- Gap: To the authors’ knowledge, no prior studies focus specifically on the cyclical properties of gross capital inflows to LIDCs.

### IV. Data, measurement, and sample construction
- Coverage and period:
  - Dataset contains information for 58 LIDCs and 92 emerging markets for the period from 1990 to 2012.
- Data sources and methodology:
  - Builds on IMF Balance of Payment Statistics (BOPS) and uses IMF World Economic Outlook (WEO) database to close gaps.
  - Compilation methodology follows Balance of Payments Manual version 5 (BPM5).
- Definition of key series:
  - Non-FDI private inflows = portfolio investments liabilities + other investment liabilities + financial derivatives; excludes official liabilities.
  - For official liabilities, inflows destined for official sectors (government and central banks) are proxied using available data.
  - Focus given to gross flows rather than net flows.
- Sample restrictions:
  - Fragile and small countries are excluded from the core sample (sample restricted to non-fragile and non-small LIDCs).

### V. Stylized facts on inflows and surge analysis
- Share and level observations:
  - The share of non-FDI private inflows in total flows is increasing for LIDCs and getting closer to figures observed in EMs.
  - Non-FDI inflows to LIDCs are on average considerably lower than inflows to EMs.
  - Inflows to the top quartile of LIDCs as a share of GDP are comparable to the median inflows in EMs.
  - After the crisis, inflows to the top quartile of LIDCs converged to the top quartile of EMs.
- Surge identification and counts:
  - Sample for surge analysis: 95 countries and 2034 observations.
  - Total surges detected: 296 surges.
  - Surges detected in LIDCs: 77 surges out of 296 total.
  - Detected waves of increased share of surges: 1990-1994, 1996-1997, 2004-2008 and 2010-2012.
  - By 2007, surges spread and reached close to 40 percent of the developing countries in the sample.
  - After a decline following the global financial crisis, the number of surges in LIDCs rebounded in 2012.
- Surge definition:
  - A surge for a given country is a period satisfying both criteria:
    - (i) gross non-FDI private inflows as percent of GDP are in the top quartile of its own country sample (time dimension);
    - (ii) gross non-FDI private inflows as percent of GDP are in the top quartile of the cross-country sample.

### VI. Preliminary assessment of cyclicality and measurement approach
- Cyclical variable construction:
  - Cyclical component of GDP constructed using univariate filtering techniques due to limited availability of unemployment and capacity utilization for LIDCs.
  - Hodrick-Prescott filter applied to the log of the GDP series at constant 2005 national prices from version 8.0 of the Penn World Table.
- Reduced-form associations (first-pass):
  - Positive and significant association between gross private capital inflows (excluding FDI) as a share of trend/potential GDP and the cyclical component of output is found in baseline specifications.
  - Including growth of potential output alongside the output gap still yields a positive and significant association between capital inflows and the cyclical component of output.
  - Net flows are less related to the cycle than gross flows.

### VII. Robustness, controls, and extensions
- Robustness checks implemented:
  - Alternative estimation methods including fixed-effects regressions and Blundell-Bond system GMM estimator produce consistent results.
  - Results robust to alternative measures of financial and trade openness.
  - Results robust to different measures of institutional quality.
  - Results robust to alternative variables capturing leverage.
  - Inclusion of an index measuring financial reform does not fundamentally alter conclusions.
  - Accounting for “risk-on/risk-off regimes” yields similar findings on procyclicality.
- Control variables with noteworthy effects:
  - Changes in risk aversion: statistically significant in most specifications with expected sign.
  - Trade openness: positive and significant.
  - Changes in the terms of trade: negative and significant.

*Source: _wp15163 - References _______________________________________________________________________________________  24*

### Appendix A for a description of the data and sources) with a smoothing parameter of 6.25, as

### _wp15163 - Appendix A for a description of the data and sources) with a smoothing parameter of 6.25, as

### Methodology: cyclical decomposition and filters
- The HP filter is applied with a smoothing parameter of 6.25, as suggested by Ravn and Uhlig (2002) for annual data.
- Alternative smoothing parameter experimented: a smoothing parameter of 1 for the HP filter (suggested by Dabla-Norris, Minoiu, and Zanna (2015) for LIDCs).
- Alternative filter experimented: Christiano and Fitzgerald filter.
- Outcome of robustness checks: variations in smoothing parameters and use of the Christiano and Fitzgerald filter do not substantially affect the estimates of the output gap.

### Unconditional correlations: private capital flows and output cycle
- Sample period for country-level simple correlations: 1990-2012.
- Comparison groups: several LIDCs (Low-Income Developing Countries) and EMs (Emerging Markets).
- Main pattern:
  - Unconditional association between private capital flows (as a share of potential GDP) and the cyclical component of output tends to be more positive for EMs (grey bars in the Figure) relative to LIDCs (black bars).
  - Interpretation: flows are more procyclical in EMs relative to LIDCs.
- Robustness: patterns remain broadly similar when using HP smoothing parameter of 1 or the Christiano and Fitzgerald filter.

### Amplitude of capital flows over the domestic business cycle
- Definition: amplitude = difference between capital inflows as a share of potential GDP in "good times" and "bad times" (good/bad defined as periods when GDP growth is above/below the median).
- Median amplitude results:
  - LIDCs: median amplitude is close to 0.4.
  - EMs: median amplitude is 1.2.
- Interpretation:
  - Gross capital flows are typically procyclical (higher in good times, lower in bad times) for both groups.
  - The decline of capital flows in bad times is markedly larger in EMs than in LIDCs.
- Note: Kaminsky, Reinhart and Vegh (2005) used net flows rather than gross capital inflows (footnote).

### Estimation framework
- Estimable equation (as presented):
  - c f_{i,t} = α_i + ρ c f_{i,t−1} + β \bar{y}_{i,t} + ∑_{m=1}^{M} δ_m X_{m,i,t} + λ_t + ε_{i,t}
  - Where:
    - c f_{i,t} = private capital flows measure as a share of potential GDP in country i at year t.
    - \bar{y}_{i,t} = output gap (cyclical component of GDP).
    - X_{m,i,t} = control variable m.
    - α_i = country fixed-effects; λ_t = time fixed-effects; ε_{i,t} = disturbance term.
- Rationale for using output gap: provides a more direct measure of cyclical movements compared with real GDP growth.
- Caveat: economic concept of “potential output” is not clear-cut for LIDCs and lower middle-income economies undergoing structural transformation; shocks to trend growth may be primary source of fluctuations (Aguiar and Gopinath, 2007).
- Robustness strategy: consider regressions that include both the output gap and the growth rate of potential output.

### Estimation methods and inference
- Primary estimator: standard fixed-effects methods with Driscoll and Kraay (1998) corrected standard errors to account for cross-sectional dependence.
- Reason for correction: cross-sectional dependence (spillovers, spatial effects, common unobserved factors) can render conventional panel estimators inconsistent (Kapetanios, Pesaran and Yamagata, 2011).
- Evidence on common factors: Puy (2013) documents importance of global and regional common factors in driving bond and portfolio flows to developing countries.
- Endogeneity concerns:
  - Output gap likely endogenous to capital flows.
  - Some controls may be correlated with country fixed effects or determined by capital flows.
- Additional estimator: system (Blundell-Bond) GMM estimator to handle potential endogeneity using lagged values as instruments (Roodman, 2009).
- Implementation note: Driscoll-Kraay procedure implemented in Stata 13 using code written by Daniel Hoechle from the University of Basel.
- Open issue: finding valid external instruments (beyond lagged values) for the output gap remains unresolved.

### Control variables (selection and rationale)
- Classification: global (“Push”) factors and country-specific (“Pull”) factors; further distinction for global cycles, domestic financial cycles, and country characteristics.
- Global (Push) factors considered:
  - VIX or VXO implied volatility index to capture global economic uncertainty / investor risk appetite (Forbes and Warnock, 2012).
  - Measures of global liquidity: interest rates in advanced economies; measures of changes in global money supply.
  - Changes in net interoffice assets of foreign banks in the US as proxy for availability of wholesale bank funding (Bruno and Shin, 2014).
  - Changes in global commodity prices and changes in the terms of trade (country-specific flavor).
- Country-specific (Pull) factors considered:
  - Lagged dependent variable (past capital flows) to capture herding, persistence, and omitted variable effects.
  - Financial development / leverage: stock market capitalization as a share of GDP (where available); alternative used for LIDCs: credit to the private sector as a share of GDP.
  - De jure measures of capital controls / financial account openness (Chinn and Ito (2006); Quinn and Toyoda (2008)) — noted likely endogenous.
  - Measures of institutional quality / country risk (e.g., ICRG country risk rating; higher values indicate lower risk / better institutional quality).
  - Macroeconomic stability proxies (e.g., inflation rate).
  - Trade openness proxies to capture demand for trade finance and sensitivity to global demand.
- Endogeneity caveats for some pull variables: financial depth indicators and de jure openness measures may be endogenous to international capital flows.
- Policy rationale note: capital controls may serve various roles (transaction cost, alter risks, macroprudential “speed limits”), and the comprehensive debate is beyond scope.

### Baseline results (fixed-effects with cross-sectional corrected standard errors)
- Specification overview:
  - Table 1 presents fixed-effects regressions with Driscoll-Kraay standard errors for several specifications of Equation (1).
  - Parsimonious control set includes: lagged dependent variable; changes in the VXO index; de facto trade openness; leverage (private credit to GDP ratio); Chinn-Ito index (de jure international financial openness); changes in the terms of trade; ICRG country risk rating.
  - Small and fragile states excluded from all specifications. Ethiopia excluded due to significant weaknesses in national accounts statistics.
- Main findings:
  - For full sample and EMs: private capital flows are procyclical with β > 0 and statistically significant at conventional levels.
  - For LIDCs (specifications 5 and 6):
    - Evidence continues to point to procyclical capital flows, but with a smaller magnitude for the output gap coefficient compared with EMs.
    - Statistical significance reduced but coefficient remains significant at the 10 percent level.
    - Capital flows are more persistent in LIDCs relative to EMs: coefficients for the lagged dependent variable are higher for LIDCs.

*Source: _wp15163 - Appendix A for a description of the data and sources) with a smoothing parameter of 6.25, as*

### 0.4 compared to 0.2 for EMs), as shown in specifications 3 to 6 of Table 1. The positive and

### _wp15163 - 0.4 compared to 0.2 for EMs), as shown in specifications 3 to 6 of Table 1. The positive and

### Main findings on cyclicality and persistence of capital flows
- Market-driven private capital inflows to Low-income Developing Countries (LIDCs) are typically less procyclical than capital inflows to emerging markets (EMs).
- Capital flows to LIDCs are more persistent:
  - Statistically significant coefficients obtained for the lagged dependent variable for LIDCs, consistent with Franken and van Wijnbergen (2010) for 1981-2006.
- Changes in global risk aversion are a significant correlate of private capital flows in all specifications, but:
  - The association between the VXO variable and capital flows is weaker in LIDCs (smaller magnitude of coefficient in specification 6), suggesting LIDCs are less sensitive to cycles in global risk aversion.
- Trade openness and changes in the terms of trade present statistically significant coefficients in the LIDC sample (specification 6).
  - The terms of trade variable has a negative sign, consistent with a crowding out mechanism where export revenues substitute for financial inflows (Lane, 2014).
- Coefficients for financial openness, leverage, and country risk are not statistically significant in the vast majority of specifications.
- Overall conclusion: private capital flows are procyclical, but less so in LIDCs; flows to LIDCs are more persistent; LIDCs are less sensitive to global risk aversion.

### Mechanisms and interpretation
- Possible explanations for lower procyclicality in LIDCs:
  - Smaller banking sector and lower leverage reduce amplification via the financial accelerator.
  - Types of financial instruments: LIDCs rely more on bank flows and trade finance; EMs receive more tradable securities with asset prices that are procyclical and induce portfolio rebalancing (Lane, 2014).
- Practical considerations:
  - LIDCs’ reliance on bank flows and trade finance could make their inflows less procyclical compared to EMs’ tradable-securities-dominated inflows.
- Note on VXO and time effects:
  - Since VXO does not vary across countries, baseline specifications excluding time fixed effects were considered; results are very similar. Output gap significance is somewhat higher, magnitude stronger for EMs, and VXO is no longer significant for LICs.

### Robustness checks (Section VI)
- General:
  - Extensive robustness exercises include alternative estimation methods (system GMM), alternative samples (countries with surges in capital inflows per Araujo et al., 2015), risk-on/risk-off regimes (Lane, 2014), disentangling permanent vs transitory shocks (Aguiar and Gopinath, 2007), alternative control variables (including an index of financial reform), and alternative measures of capital flows (adding FDI; net vs gross flows).
  - Additional robustness checks available upon request.

- A. GMM Regressions
  - Re-estimated using Blundell-Bond system GMM with forward orthogonal deviations and robust standard errors; instruments collapsed by variable and lag distance.
  - Results broadly in line with fixed-effects regressions:
    - Private capital flows positively associated with the output gap in all specifications with statistically significant coefficients (albeit only at the 10 percent level in the LIDC-focused specification that includes control variables).
    - Coefficient for the output gap is smaller for LIDCs (weaker procyclicality).
    - Coefficient for the lagged dependent variable is higher for LIDCs (greater persistence).
    - VXO remains important for full sample and EMs.
    - Trade openness presents positive and significant coefficients in all specifications under GMM.
    - Financial openness matters for EMs, but not for LIDCs.
  - Diagnostics: Hansen test suggests overidentifying restrictions valid for all specifications; Sargan test rejects validity. Sargan not robust to heteroskedasticity or serial correlation; Hansen preferred here.

- B. Countries with Surges in Capital Inflows
  - Sample restricted to countries that experienced surges in capital inflows (Araujo et al., 2015).
  - Confirmed private capital flows positively associated with the output gap, but less so in LIDCs.
  - Other results closely follow unrestricted sample, including role of risk aversion (Table 3).

- C. Risk-on, Risk-off Regimes
  - Modeled common risk factors via a common dynamic factor and estimated using CCEMG estimator (Pesaran, 2006), allowing heterogeneity across countries.
  - Findings:
    - Capital flows are procyclical in EM-only specifications and full sample.
    - Evidence of procyclicality is mixed when only LIDCs are included:
      - Output gap coefficient significant at the 10 percent level in specification 5 (no control variables).
      - Not statistically significant in specification 6 (controls included).
    - Common dynamic factor linked to dependent variable significant in all specifications except specification 6; coefficient larger for EMs.
  - Time-period splits (global boom 2003-2007; 2008-2012 recession/recovery) yield:
    - 2003-2007: association between private capital flows and output gap generally positive, not robust for LIDCs.
    - 2008-2012: similar results, larger coefficients for output gap; not significant for LIDC-only specification with controls.
  - Conclusion: market-driven capital flows to LIDCs are less procyclical is robust to risk-on/risk-off regime approaches.

- D. Capital Inflows, Cyclical Fluctuations, and Trend Shocks
  - Added growth in trend-GDP to regressions (Table 5) to disentangle permanent vs transitory shocks (Aguiar and Gopinath, 2007).
  - Results:
    - Output gap coefficient remains positive and statistically significant in all specifications; marginally smaller in LIDC-only specification.
    - Gross private non-FDI capital inflows positively and significantly associated with trend growth in full sample and EM sample, suggesting inflows link to permanent shocks as well as temporary ones.
    - Trend growth not statistically significant (and presents negative sign) in LIDC-only specifications.
    - Control variable results align with baseline regressions.

- E. Alternative Control Variables
  - Alternative institutional quality measures:
    - Executive Constraints (Polity IV, scale 1 to 7) and quality of bureaucracy (ICRG).
    - Quality of bureaucracy results similar to baseline.
    - Using Executive Constraints, the cyclical output coefficient is no longer statistically significant in LIDC-only specification.
  - Alternative leverage measure: loan to deposit ratio — inclusion does not affect main results.
  - Trade openness alternative:
    - De jure measure used: “Trends in average most favored nation applied tariff rates in developing and industrial countries” (World Bank) — results similar to baseline.
  - Alternative global variables to VXO:
    - US interest rates (10-year bond rate and 3-month T-bill): cyclicality results for LIDCs still hold.
    - Commodity price index (industrial materials index from PCPS per IMF (2013a)): results unchanged.
    - ISM manufacturing index included; coefficients not statistically significant.
  - Financial reform index (Abiad, Detragiache, and Tressel, 2010):
    - Data limitations: reduces LIDC sample to 13 countries; time coverage up to 2005.
    - Table 6: financial reform index coefficient significant and positive for overall sample and EM-only sample, but not for LIDC sample.
    - Output gap coefficient remains but is no longer statistically significant in LIDC-only specification when financial reform index is included.
  - Note: VXO excluded from some specifications due to limited time dimension.

- F. Alternative Measures of the Dependent Variable
  - Table 7 explores other dependent variables:
    - Gross FDI flows only (specifications 1 and 2):
      - Output gap coefficient not statistically significant in full sample.
      - Output gap positive and significant at the 10 percent level for the LIDC sample.
      - Lagged dependent variable significant and positive with large coefficients relative to baseline — suggests FDI is more persistent and less related to the cycle than private non-FDI flows.
    - Adding FDI to private capital flows (specifications 3 and 4):
      - Both output gap and persistence higher in LIDCs relative to baseline.
    - Net capital flows (excluding FDI) as dependent variable (specifications 5 and 6), adding portfolio assets and other investment assets:
      - Net flows are less related to the cycle; coefficients positive but not significant.

### Policy implications
- As LIDCs develop, private capital inflows are likely to become more procyclical.
- Policies to manage capital inflows and mitigate destabilizing effects linked to procyclicality may become more relevant for LIDCs.
- Prudential toolkit and financial-stability policies informed by EM experience (Ostry et al., 2012) would be increasingly useful for policymakers in LIDCs.
- Procyclical and volatile capital flows challenge traditional monetary policy frameworks in LIDCs based on growth-rate targets for monetary aggregates because:
  - Instability and unpredictability of monetary aggregates largely reflect high volatility of reserve money linked to capital flows.
  - Reinforces need to accelerate adoption of modern forward-looking monetary policy frameworks in these countries (IMF, 2014b).

### Conclusions and avenues for future research (Section VII)
- By the second half of the 2000s, several LIDCs experienced increased non-FDI private inflows and surges in 2004-2008, indicating LIDCs were “catching the wave” of increased flows to developing countries.
- Main empirical conclusions:
  - Market-driven capital inflows to LIDCs are less procyclical and more persistent than inflows to EMs (period 1990-2012).
  - LIDCs are less sensitive to changes in global risk aversion relative to EMs.
  - Findings robust across estimation methods, samples, and control variables.
- Suggested mechanisms to investigate further:
  - Role of banking sector size and economy-wide leverage in shaping cyclicality.
  - Influence of types of financial instruments available to international investors (bank flows and trade finance in LIDCs vs tradable securities in EMs).
- Important avenue for future research is to identify the precise mechanisms driving these results.

*Source: IMF working paper content provided in the supplied PDF excerpt.*

### REFERENCES

### _wp15163 - REFERENCES

### Major cited works
- Abiad, A., Detragiache, E. and Tressel, T. (2010). “A New Database of Financial Reforms”. IMF Staff Papers, 57, 281-302.
- Aghion, P., Angeletos, G., Banerjee, A., Manova, K. (2010). “Volatility and Growth: Credit Constraints and the Composition of Investment”. Journal of Monetary Economics, 57, 246-265.
- Aguiar, M. and Gopinath, G. (2007). “Emerging Market Business Cycles: The Cycle Is the Trend”. Journal of Political Economy, 115, 69-102.
- Alesina, A., Campante, F., and Tabellini, G. (2008). “Why is Fiscal Policy Often Pro-cyclical”. Journal of the European Economic Association, 6(5): 1006-1036.
- Alfaro, L., Kalemli-Ozcan, S., and Volosovych, V. (2008). “Why Doesn’t Capital Flow from Rich to Poor Countries? An Empirical Investigation”. The Review of Economics and Statistics, 90: 347-368.
- (Additional referenced works include papers by Bakus et al. (1992), Bianchi (2011), Bluedorn et al. (2013), Broner et al. (2013), Brunnermeier et al. (2012), Bruno and Shin (2014), Cihak et al. (2012), Chinn and Ito (2006), Dabla‑Norris et al. (2015), Driscoll and Kraay (1998), Eberhardt (2012), Faria et al. (2007), Feenstra et al. (2013), Forbes and Warnock (2012), Frankel et al. (2013), Franken and van Wijnbergen (2010), Ghosh et al. (2014), Gopinath (2005), Guscina et al. (2014), multiple IMF reports (2013a, 2013b, 2014a, 2014b), Kaminsky, Reinhart and Vegh (2005), Kapetanios et al. (2011), Lane (2014), Lane and McQuade (2014), Milesi‑Ferretti and Tille (2011), Obstfeld and Rogoff (1996), Obstfeld (2012, 2014), Ostry et al. (2012), Pesaran (2006), Puy (2013), Quinn and Toyoda (2008), Rand and Tarp (2002), Ravn and Uhlig (2002), Reinhart and Reinhart (2009), Roodman (2009), Van Wincoop and Tille (2010), Vegh (2013).)

### Figures: visual comparisons and dynamics (1990–2012)
- Figure 1: Comparison between EMs and LIDCs by percentile (% GDP).
- Figure 2: Surges of private non-FDI capital flows (% of total); plotted series include Wave 1, Wave 2, Wave 3, Wave 4; panel shows "Freq. of surges: EM and LIDCs" and "Freq. of surges: LIDCs".
- Figure 3: Capital Flows and the Cyclical Component of Output (1990-2012); correlations computed using HP Filter, λ=6.25, series plotted separately for LIDCs and EMs.
- Figure 4: Amplitude of Private Inflows/GDP for periods of Growth above and below the Median; country-level labels shown (selected examples include Mauritania, Belarus, Brunei Darussalam, Ghana, Azerbaijan, Namibia, Jordan, Oman, Mali, Burkina Faso, Syria, Venezuela, Malaysia, Botswana, Mongolia, Lebanon, Nigeria, Gabon, Macedonia, Egypt, Gambia, Niger, Malawi, Uganda, Guatemala, Lao PDR, Rwanda, Senegal, China, Kyrgyz Republic, El Salvador, Mozambique, Benin, Bolivia, Morocco, Panama, Qatar, Moldova, Cameroon, South Africa, Lesotho, Albania, Saudi Arabia, Bangladesh, Sri Lanka, Philippines, Tanzania, Madagascar, Vietnam, Ecuador, Brazil, Equatorial Guinea, Paraguay, Cambodia, Chile, Costa Rica, Tunisia, Jamaica, Kenya, Zambia, Mexico, Bulgaria, Russian Federation, Iran, Uzbekistan, Uruguay, Dominican Republic, Pakistan, Hungary, Colombia, Romania, Indonesia, Honduras, Poland, India, Turkmenistan, Croatia, Argentina, Kazakhstan, Peru, Latvia, Thailand, Kuwait, Estonia, Lithuania, Turkey, Ukraine, Armenia, Serbia, Serbia Republic, Ukraine.)

### Key empirical results from regression tables (preserve reported coefficients and statistics exactly)
- Table 1. Fixed Effects Regressions for Private Non-FDI Capital Flows (selected coefficients and diagnostics)
  - Lagged Dependent Variable: 0.229*** (Full, column 1) [0.073]
  - Output Gap: 0.296** (Full, column 1) [0.107]
  - ∆(VXO): -1.457*** (Full, column 1) [0.447] (columns 3–5 report variant coefficients)
  - Observations: 1,323 (Full columns), 954 (EMs), 369 (LIDCs)
  - Number of countries: 75 (Full), 55 (EMs), 20 (LIDCs)
  - R-2 (within): 0.180 (Full, column 1)
  - Note: Driscoll-Kraay standard errors in brackets; *** p<0.01, ** p<0.05, * p<0.1. Time effects coefficients not reported.

- Table 2. GMM Regressions (selected coefficients and diagnostics)
  - Lagged Dependent Variable: 0.320*** (Full, column 1) [0.065]
  - Output Gap: 0.328*** (Full, column 1) [0.099]
  - ∆(VXO): -3.829*** (Full, column 1) [1.156]
  - Trade Openness: 2.883* (Full, column 1) [1.475]
  - Observations: 1,323 (Full); Number of countries: 75 (Full)
  - Hansen test: 51.635 (Full, column 1)
  - Sargan test: 161.6*** (Full, column 1)
  - Note: HAC robust standard errors clustered by country in brackets; Time effects coefficients not reported. "The Sargan statistic is not robust to heteroskedasticity or autocorrelation."

- Table 3. Regressions for Countries that Experienced Surges (selected coefficients)
  - Lagged Dependent Variable: 0.224*** (Full, column 1) [0.073]
  - Output Gap: 0.297** (Full, column 1) [0.110]
  - ∆(VXO): -1.664*** (Full, column 1) [0.430]
  - Observations: 1,158 (Full); Number of countries: 67 (Full)
  - R-2 (within): 0.183 (Full, column 1)

- Table 4. Regressions with Time-Variant Unobservable Factors (CCEMG Estimator)
  - Lagged Dependent Variable: 0.203*** (Full, column 1) [0.041]
  - Output Gap: 0.239*** (Full, column 1) [0.067]
  - Common factor linked to capital flows: 0.646*** (Full, column 1) [0.091]
  - Root Mean-Squared Error (sigma): 3.25 (Full, column 1)
  - Observations: 1,223 (Full); Countries: 63 (Full)
  - Note: Pesaran (2006) Common Correlated Effects Mean Group Estimator. Standard errors in brackets. Coefficients computed as outlier-robust means.

- Table 5. Regressions Including Trend Growth
  - Lagged Dependent Variable: 0.229*** (Full, column 1) [0.072]
  - Output Gap: 0.265** (Full, column 1) [0.108]
  - Trend Growth: 0.196** (Full, column 1) [0.069]
  - ∆(VXO): -1.328*** (Full, column 1) [0.411]
  - Observations: 1,323 (Full); Number of countries: 75 (Full)
  - R-2 (within): 0.184 (Full, column 1)

- Table 6. Regressions with the Financial Reform Index (selected coefficients)
  - Lagged Dependent Variable: 0.197** (Full) [0.077]
  - Output Gap: 0.261*** (Full) [0.066]
  - Financial Reform Index: 0.441*** (Full) [0.102]
  - Observations: 706 (Full); Number of countries: 53 (Full)
  - R-2 (within): 0.218 (Full)

- Table 7. Alternative Measures of Capital Flows (selected coefficients; columns label measures: Gross FDI, Gross (Private + FDI), Net Flows (excl. FDI))
  - Example: Lagged Dependent Variable: 0.644*** (LIDCs column 1) [0.050]; 0.669*** (Full column 2) [0.045]
  - Example: ∆(VXO): 0.697*** (LIDCs col 1) [0.087]; 0.429** (Full col 2) [0.173]
  - Observations vary by specification: e.g., 367 (LIDCs col 1), 1,321 (Full col 2)
  - R-2 (within) examples: 0.512 (LIDCs col 1), 0.495 (Full col 2)
  - Note: Driscoll-Kraay standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1.

### Appendix A: Description of selected variables (variable definitions and sources)
- Capital Flows: Private capital flows (excluding FDI) as a share of trend GDP. Both variables are denominated in US dollars. See main text for source on capital flows. GDP in US dollars from WEO dataset.
- VXO index: Implied volatility index on the S&P 500. Calculated by Chicago Board Options Exchange. Source: Haver Analytics.
- Trade openness: Sum of exports and imports divided by GDP. Source: WEO database.
- Terms of trade: Log of terms of trade index. Source: WEO database.
- Country Risk/Political Instability: Log of ICRG composite risk and political risk ratings. Source: ICRG.
- Financial Account Openness: Chinn-Ito kaopen index, based on data from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions. Source: Chinn & Ito (2006).
- Real GDP: Real GDP at constant 2005 national prices (rgdpna series). Source: Penn World Table, Version 8.0. Feenstra et al. (2013).
- Output Gap: Cyclical component of the log of the Real GDP series using the Hodrik-Prescott or Christiano and Fitzgerald filters. The filter was estimated over the entire period for which the GDP data is available in the PWT dataset. Authors’ calculations.
- Exchange rates: Nominal LCU/$ (average and end of period) and real exchange rate index. Source: WEO database.
- Leverage: Private credit to GDP ratio (missing values were interpolated). Source: Global Financial Development Database. Cihak et al. (2012).

### Appendix B: Descriptive statistics for selected variables (examples, preserve numeric entries)
- (Capital Flows)/(Potential Output): Mean 2.128, Std. Dev. 5.890, Min -46.246, Max 68.710, N = 2005; between 2.934, within 5.168, T-bar = 21.3298.
- Output Gap: Mean -0.039, Std. Dev. 3.363, Min -46.754, Max 22.375, N = 1951; between 0.216, within 3.356, T-bar = 21.9213.
- ∆Log of VXO index: Mean 2.994, Std. Dev. 0.322, Min 2.496, Max 3.545, N = 2152; between 0.004, within 0.322, T-bar = 22.8936.
- Log of Terms of Trade: Mean 4.631, Std. Dev. 0.333, Min 2.912, Max 6.360, N = 2048.
- Trade Openness: Mean 4.251, Std. Dev. 0.492, Min 2.581, Max 6.696, N = 2110.
- Capital Account Openness: Mean 0.030, Std. Dev. 1.464, Min -1.864, Max 2.439, N = 1906.
- Leverage: Mean 2.943, Std. Dev. 0.874, Min -2.161, Max 5.111, N = 1749.
- ICRG Country Risk: Mean 4.184, Std. Dev. 0.144, Min 3.212, Max 4.490, N = 1819.

### Appendix C: Descriptive statistics for selected non-small non-fragile LIDCs (selected country examples)
- Bangladesh: Capital Flows/ Potential GDP Obs. 23, Max 2.35, Min -0.1, Mean 0.25, Std. Dev. 0.49; Output Gap Obs. 22, Max 0.86, Min -0.62, Mean 0.04, Std. Dev. 0.34.
- Mauritania: Capital Flows/ Potential GDP Obs. 23, Max 16.44, Min -46.25, Mean -7.46, Std. Dev. 12.93; Output Gap Obs. 22, Max 8.37, Min -3.47, Mean -0.11, Std. Dev. 3.1.
- Mongolia: Capital Flows/ Potential GDP Obs. 23, Max 27.75, Min -4.62, Mean 2.87, Std. Dev. 6.55; Output Gap Obs. 25, Max 5.63, Min -6.14, Mean -0.21, Std. Dev. 3.19.
- (Additional country-level descriptive statistics are provided for Benin, Bolivia, Burkina Faso, Cambodia, Cameroon, Gambia, Ghana, Honduras, Kenya, Kyrgyz Republic, Lao PDR, Lesotho, Madagascar, Malawi, Mali, Moldova, Mozambique, Niger, Nigeria, Rwanda, Senegal, Tanzania, Uganda, Uzbekistan, Vietnam, Zambia.)

### Appendix D: List of LIDC countries and classification (fragile/small indicated exactly)
- Afghanistan, I.R. of — Fragile
- Bangladesh — (no tag in table)
- Benin — (no tag)
- Bhutan — Small
- Bolivia — (no tag)
- Burkina Faso — (no tag)
- Burundi — Fragile
- Cambodia — (no tag)
- Cameroon — (no tag)
- Central African Rep. — Fragile
- Chad — Fragile
- Comoros — Fragile
- Congo, Dem. Rep. of — Fragile
- Congo, Republic of — Fragile
- Côte d'Ivoire — Fragile
- Djibouti — Small
- Eritrea — Fragile
- Ethiopia — (no tag)
- Gambia, The — (listed as "Solomon IslandsFragileSmall" grouping in table; entries show Gambia, The and later Solomon Islands)
- Ghana — (no tag)
- Guinea — Fragile
- Guinea-Bissau — Fragile
- Haiti — Fragile
- Honduras — (no tag)
- Kenya — (no tag)
- Kiribati — FragileSmall
- Kyrgyz Republic — (no tag)
- Lao People's Dem.Rep — (no tag)
- Lesotho — (no tag)
- Liberia — Fragile
- Madagascar — (no tag)
- Malawi — (no tag)
- Mali — (no tag)
- Mauritania — (no tag)
- Moldova — (no tag)
- Mongolia — (no tag)
- Mozambique — (no tag)
- Myanmar — Fragile
- Nepal — Fragile
- Nicaragua — (no tag)
- Niger — (no tag)
- Nigeria — (no tag)
- Papua New Guinea — (no tag)
- Rwanda — (no tag)
- São Tomé & Príncipe — FragileSmall
- Senegal — (no tag)
- Sierra Leone — Fragile
- Solomon Islands — FragileSmall
- Sudan — Fragile
- Tajikistan — Fragile
- Tanzania — (no tag)
- Togo — Fragile
- Uganda — (no tag)
- Uzbekistan — (no tag)
- Vietnam — (no tag)
- Yemen, Republic of — Fragile
- Zambia — (no tag)
- Zimbabwe — Fragile

*Source: _wp15163 - REFERENCES (IMF document content provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15163.pdf_
