## Capital Flows to Latin America in the Aftermath of the Commodities Super-Cycle (capitalflow)

## Source details

**Canonical URL:** [Capital Flows to Latin America in the Aftermath of the Commodities Super-Cycle (capitalflow)](https://www.imf.org/-/media/files/publications/reo/whd/2019/october/english/capitalflow.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/reo/whd/2019/october/english/capitalflow.pdf.md)
- [Structured JSON version](/-/media/files/publications/reo/whd/2019/october/english/capitalflow.pdf.json)

---

### Role, focus, and context
- Capital flows enable growth via the capital accumulation channel and the productivity channel; these channels matter for LAC because: (i) domestic savings are low relative to investment, and (ii) productivity growth has been persistently weak.
- Chapter focus:
  - Reviews recent trends in capital flows to LAC and explores policy options to mitigate boom-bust patterns.
  - Presents evidence on insufficient domestic savings to finance investment; analyzes behavior and drivers of different categories of capital flows over the last decade; examines determinants and risks posed by sudden stops; and explores policy options to deal with sudden stops.
- Key contextual reference: persistent weak growth in LAC and the end of the commodity super-cycle are central to recent changes in the nature and determinants of flows.

### Low savings and investment in LAC
- Key statistics:
  - Investment and savings in LAC averaged 22 and 18 percent of GDP during 2000–17, respectively.
  - These averages are about 4 and 6.5 percentage points of GDP lower than the average of other EMEs.
- Empirical observations:
  - Domestic savings in LAC are low even after controlling for structural factors (income levels and share of working-age population); residuals from the savings regression are on average negative for LAC, especially for Brazil, Colombia, and Uruguay.
  - Access to foreign financing has partly covered the savings deficit, notably during the commodity boom when investment-to-GDP rose while domestic savings lagged and the current account deficit increased.
- Correlation between inflows and investment:
  - For the average EME, investment growth rises by 1.7 percentage points after a capital flow surge (effect not statistically significant; coefficient: 1.698; standard error: (1.828)).
  - For the LA7 countries, investment growth increases by 11 percentage points after a capital flow surge (coefficient: 11.03**; standard error: (4.236); ** p<0.05).

### Portfolio flows, composition, and volatility
- Aggregate flows and trends:
  - Net capital flows to the region were around 5 percent of GDP in 2010–14.
  - Net capital flows declined to around 3 percent of GDP since 2014.
- Composition shifts (LA7, since 2013):
  - Gross portfolio inflows fell by 2 percentage points of GDP in five years.
  - FDI declined by 1 percentage point of GDP.
- Regional contrasts:
  - FDI declines in metal exporters (Chile and Peru) since 2014; FDI increased in Brazil.
  - Gross portfolio inflows declined in most countries, except Peru and Argentina (Argentina’s rise in 2016–17 was largely reversed in 2018).
- Volatility:
  - Portfolio and other investment inflows to LAC have been more volatile than FDI inflows.
  - Volatility appears to have risen in recent years, potentially due to spikes in global financial market volatility (2015 and 2018), higher global interest rates, lower commodity prices, and greater sensitivity of flows to external conditions.

### Have the drivers of capital flows changed?
- Empirical approach:
  - Dynamic panel for EMEs controlling for push factors (VIX and US long-term interest rates) and pull factor (growth differential with advanced economies); separate regressions for total, portfolio, and FDI inflows; system-GMM (Arellano-Bond) estimator.
- Main findings on determinants:
  - Spikes in global risk aversion (VIX) and higher 10-year US government bond yields tend to reduce gross portfolio inflows to LAC; they have no bearing on gross FDI inflows.
  - Growth differentials vis-à-vis advanced economies have large and statistically significant effects on both portfolio and FDI inflows.
- Rising sensitivity over time (LA7, 2012–18):
  - Coefficients on VIX and 10-year US bond yield almost double relative to longer samples.
  - A 10-point rise in the VIX reduces portfolio inflows to the LA7 economies by 1.7 percentage points of GDP.
  - A 100 basis points increase in the 10-year US bond yield lowers portfolio inflows by 0.8 percentage points of GDP.
  - A 1 percentage points rise in the growth differential increases portfolio flows to the LA7 countries by 0.7 percentage points of GDP, and FDI flows by 0.4 percentage points of GDP.
- Implication:
  - Given modest regional growth prospects and significant downside risks (heightened US-China trade tensions, increased global uncertainty, slower global growth, lower commodity prices), LAC could experience episodes of sudden stops as global financial conditions tighten and domestic growth slows.

### Sudden stops: definitions, statistics, and costs
- Definition and identification:
  - Sudden stops defined using the Forbes and Warnock (2012) algorithm: year-over-year change in four-quarter moving average of capital inflows more than two standard deviations below the historical average for at least a quarter.
  - Using quarterly gross private capital inflows data from 1970 for 165 countries, 278 sudden stop episodes identified, of which 43 are from LAC.
  - Average duration of a sudden stop episode: 3.89 quarters.
- Growth-deceleration episodes linked to sudden stops:
  - 130 episodes of growth decelerations linked to sudden stops identified, of which 21 cases are in LAC.
  - Average duration of a growth-deceleration episode linked to a sudden stop: 10.6 quarters.
- Output costs:
  - The loss of output in LAC after a sudden stop averages 7.4 percent of the pre-sudden stop trend GDP, with a maximum of 29.3 percent.
  - Output loss computed as (Y* - Y)/Y* where Y* is counterfactual real GDP continuing at the pre-sudden stop trend (eight-quarter average growth rate before the sudden stop).

### Liability dollarization and foreign participation in local markets
- Liability dollarization:
  - Defined as bank foreign borrowing as a share of GDP in line with Cavallo et al. (2017).
  - Latest data timing: Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Mexico, Paraguay = 2019:Q2; Peru = 2019:Q1; Uruguay = 2018:Q4.
  - From a longer historical perspective, liability dollarization, fiscal deficits, and public debt are all higher today than at the end of 2007 in most countries in the region.
  - The external position has improved relative to 2007: most countries have larger reserve buffers relative to their gross external financing needs today than in 2007.
- Foreign participation in local debt markets:
  - Since 2012 foreign participation has increased significantly in Colombia (from low levels), marginally in Mexico and Peru (from high levels), and remained stable at low levels in Brazil.
  - Compared with other EMs outside the region, Peru stands out as having a relatively high foreign participation rate in local currency government debt markets.

### Empirical analysis of policy responses to sudden stops
- Policy responses considered in duration analysis: (i) monetary policy; (ii) sales of international reserves; (iii) changes in capital account restrictions; and (iv) fiscal policy (measured as change over four quarters following onset).
- Duration-model findings:
  - Exchange rate flexibility increases likelihood of exiting a sudden stop: more flexible regimes increase the probability of exit by at least 60 percent (the report also frames magnitude as "60 percentage points" in discussing effects).
  - A rise in the terms of trade of 2.5 percent (1 standard deviation) increases probability of exit by about 25 percent.
  - Monetary policy tightening increases probability of exiting a sudden stop: a rise of 3 percentage points in ex-post real interest rates increases that probability by 15 percent.
  - Capital controls, tighter fiscal policy, or sales of international reserves do not have statistically significant effects on the duration of sudden stops (stand of fiscal policy measured as change in the structural primary balance).
- Cross-sectional regressions on output costs:
  - More flexible exchange rate regimes and increases in the terms of trade reduce the costs of sudden stops; coefficients on ∆ Terms of trade are statistically significant and negative across specifications (example coefficient: -1.780*** in one specification).
  - Other policy variables (∆ Monetary policy, ∆ Reserves, ∆ Capital Controls, ∆ Fiscal policy) generally do not show robust, statistically significant effects on cross-sectional variation in output costs.

### Policy implications and recommended instruments
- Measures that reduce vulnerability to sudden stops:
  - Higher reserve buffers.
  - Lower liability dollarization.
  - Fiscal and current account consolidation.
  - Exchange rate flexibility.
  - Low inflation.
  - Strong institutions.
  - Higher share of FDI in inflows.
- Policy guidance based on empirical results:
  - Exchange rate flexibility materially reduces the duration and output losses of sudden stops.
  - Monetary policy tightening can be effective in shortening sudden stops when it raises real interest rates sufficiently (example magnitude: 3 percentage points in ex-post real rates raises exit probability by 15 percent and reduces duration by close to 15 percent).
  - Relying solely on capital account restrictions or reserve sales to shorten sudden stops is not supported by the empirical evidence in this analysis.
  - Policy responses to external adverse shocks could include coordination of monetary policy instruments, exchange rate flexibility, foreign exchange market interventions (FXI), macroprudential measures, and capital flow management measures.
  - When central banks use FXI, clear communication is important to preserve the credibility of inflation targeting (IT) frameworks and avoid the appearance of conflicting objectives between FXI and IT.

*Source: Introduction and Section 2, "Capital Flows to Latin America in the Aftermath of the Commodities Super-Cycle," Regional Economic Outlook: Western Hemisphere, International Monetary Fund, October 2019.*

### Introduction

### Introduction

### Role and risks of capital flows
- Capital flows enable growth via the capital accumulation channel and the productivity channel.
- These channels are particularly relevant for countries in LAC because: (i) domestic savings are low relative to investment, and (ii) productivity growth has been persistently weak.
- Volatile capital flows can disrupt economies during difficult times or fuel overheating and asset price bubbles in boom periods.
- Policymakers should have a framework in place to manage effects of capital flow surges and reversals, especially in the current context of high global uncertainty.

### Chapter focus
- Reviews recent trends in capital flows to LAC and explores policy options to mitigate boom-bust patterns.
- Presents evidence on insufficient domestic savings to finance investment; analyzes behavior and drivers of different categories of capital flows over the last decade; examines determinants and risks posed by sudden stops; and explores policy options to deal with sudden stops.

### Authors and acknowledgements
- Chapter prepared by Carlos Gonçalves, Antonio David, and Samuel Pienknagura, with research assistance by Genevieve Lindow and Pablo Bejar and literature review by Jaime Leyva.

### Key contextual reference
- Persistent weak growth in LAC and the end of the commodity super-cycle are central to recent changes in the nature and determinants of flows.

---

### The Low Savings, Low Investment Problem in LAC
- Investment and savings in LAC averaged 22 and 18 percent of GDP during 2000–17, respectively.
- These averages are about 4 and 6.5 percentage points of GDP lower than the average of other EMEs.
- Domestic savings in LAC are low even after controlling for structural factors (income levels and share of working-age population); residuals from the savings regression are on average negative for LAC, especially for Brazil, Colombia, and Uruguay.
- Access to foreign financing has partly covered the savings deficit, notably during the commodity boom when investment-to-GDP rose while domestic savings lagged and the current account deficit increased.
- Correlation of capital inflows and investment:
  - For the average EME, investment growth rises by 1.7 percentage points after a capital flow surge (effect not statistically significant).
  - For LAC countries, investment growth increases by 11 percentage points after a capital flow surge.

---

### Portfolio Flows Take Center Stage
- Aggregate net and gross flows:
  - Net capital flows to the region were around 5 percent of GDP in 2010–14.
  - Net capital flows declined to around 3 percent of GDP since 2014.
- Gross inflows drove most of the fall in net inflows; gross capital inflows are much larger than gross capital outflows.
- Composition shifts since 2013 (LA7 countries):
  - Gross portfolio inflows fell by 2 percentage points of GDP in five years.
  - FDI declined by 1 percentage point of GDP.
- Regional contrasts and drivers:
  - FDI declines in metal exporters (Chile and Peru) since 2014, but FDI increased in Brazil.
  - Gross portfolio inflows declined in most countries, except Peru and Argentina (Argentina’s rise in 2016–17 was largely reversed in 2018).
- Volatility:
  - Portfolio and other investment inflows to LAC have been more volatile than FDI inflows.
  - Volatility appears to have risen in recent years, potentially due to spikes in global financial market volatility (2015 and 2018), higher global interest rates, lower commodity prices, and greater sensitivity of flows to external conditions.

---

### Have the Drivers of Capital Flows Changed?
- Empirical approach:
  - Dynamic panel for a group of EMEs controlling for push factors (VIX and US long-term interest rates) and pull factor (growth differential with advanced economies).
  - Separate regressions for total, portfolio, and FDI inflows.
- Main findings on determinants:
  - Spikes in global risk aversion (VIX) and higher 10-year US government bond yields tend to reduce gross portfolio inflows to LAC, but they have no bearing on gross FDI inflows.
  - Growth differentials vis-à-vis advanced economies have large and statistically significant effects on both portfolio and FDI inflows.
- Rising sensitivity over time (LA7, 2012–18):
  - Coefficients on VIX and 10-year US bond yield almost double relative to longer samples.
  - A 10-point rise in the VIX reduces portfolio inflows to the LA7 economies by 1.7 percentage points of GDP.
  - A 100 basis points increase in the 10-year US bond yield lowers portfolio inflows by 0.8 percentage points of GDP.
  - A 1 percentage points rise in the growth differential increases portfolio flows to the LA7 countries by 0.7 percentage points of GDP, and FDI flows by 0.4 percentage points of GDP.
- Implication:
  - Given modest regional growth prospects and significant downside risks (heightened US-China trade tensions, increased global uncertainty, slower global growth, lower commodity prices), LAC could experience episodes of sudden stops as global financial conditions tighten and domestic growth slows.

---

### Sudden Stops and How to Deal with Them
- Definition and context:
  - “Sudden stops” are extreme dry-ups in foreign finance that sharply tighten credit constraints with potentially substantial economic costs.
  - Sudden stops frequently follow periods of strong capital inflows and growth and are part of the boom-bust pattern in many EMEs.
- Determinants of sudden stops:
  - External factors, particularly swings in global risk aversion, are important triggers.
  - Initial composition of capital flows matters: larger share of FDI inflows reduces likelihood of post-surge reversals ending in crises.
  - Domestic vulnerabilities (large fiscal and current account deficits, high liability dollarization) increase likelihood and costs of sudden stops.
  - High international reserve buffers reduce vulnerabilities.
  - Factors that help prevent repatriation-driven sudden stops include low liability dollarization, low inflation, flexible exchange rates, and high-quality institutions.
- Empirical indicators and country assessments:
  - External financing requirements (sum of the current account deficit and short-term external debt) relative to the stock of international reserves tend to be high in periods preceding sudden stops and external crises.
  - Recently, external funding pressures have been high in some LAC countries (including Argentina and Ecuador), but most countries seem to have adequate reserve buffers relative to projected external financing requirements.
  - Liability dollarization has remained broadly stable over the past year in most countries and declined significantly in Costa Rica.
- Policy effectiveness:
  - Policymakers cannot control global risk sentiment but can influence the likelihood and consequences of sudden stops through policy choices.
  - Both the severity and duration of sudden stops can be attenuated by an appropriate policy mix.
  - Policy tools and features that reduce vulnerability include: higher reserve buffers, lower liability dollarization, fiscal and current account consolidation, exchange rate flexibility, low inflation, strong institutions, and a higher share of FDI in inflows.

---

*Source: Introduction, "Capital Flows to Latin America in the Aftermath of the Commodities Super-Cycle," Regional Economic Outlook: Western Hemisphere, International Monetary Fund, October 2019.*

### 2.  Liability Dollarization

### 2.  Liability Dollarization

### Liability dollarization and recent trends
- Liability dollarization is defined as bank foreign borrowing as a share of GDP in line with Cavallo et al. (2017). Latest data for Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Mexico, and Paraguay are 2019:Q2; Peru is 2019:Q1; and Uruguay is 2018:Q4.
- From a longer historical perspective, liability dollarization, fiscal deficits, and public debt are all higher today than at the end of 2007 in most countries in the region (Figure 8, panels 1, 2, and 3).
- The external position has improved relative to 2007: most countries have larger reserve buffers relative to their gross external financing needs today than in 2007 (Figure 8, panel 4).

### Foreign participation in local debt markets
- Since 2012 foreign participation has increased significantly in Colombia (from low levels), marginally in Mexico and Peru (from high levels), and remained stable at low levels in Brazil.
- Compared with other EMs outside the region, Peru stands out as having a relatively high foreign participation rate in local currency government debt markets.

### Identifying sudden stops: scope and basic statistics
- Sudden stops are defined using the Forbes and Warnock (2012) algorithm: an event in which the year-over-year change in four-quarter moving average of capital inflows is more than two standard deviations below the historical average for at least a quarter.
- Using quarterly gross private capital inflows data from 1970 for 165 countries, a total of 278 sudden stops episodes are identified, of which 43 are from LAC.
- The average duration of a sudden stop episode is 3.89 quarters.
- For episodes associated with output declines, the analysis compares the eight-quarter moving average GDP growth rate before the sudden stop with growth at the onset and subsequent quarters. A total of 130 episodes of growth decelerations linked to sudden stops are identified, of which 21 cases are in LAC.
- The average duration of a growth-deceleration episode linked to a sudden stop is 10.6 quarters.

### Empirical analysis of policy responses: duration models and correlates
- Four policy responses are considered in duration analysis: (i) monetary policy; (ii) sales of international reserves; (iii) changes in capital account restrictions; and (iv) fiscal policy (policy variables usually measured as the change over four quarters following the onset of a sudden stop episode).
- Key empirical findings (duration analysis):
  - Flexibility of the exchange rate regime increases the likelihood of “exiting” a sudden stop episode: more flexible exchange rate regimes increase the probability of exit by at least 60 percent.
  - A rise in the terms of trade of 2.5 percent (1 standard deviation) increases the probability of exit by about 25 percent.
  - Monetary policy tightening increases the probability of exiting a sudden stop: a rise of 3 percentage points in ex-post real interest rates increases that probability by 15 percent.
  - Capital controls, tighter fiscal policy, or sales of international reserves do not have statistically significant effects on the duration of sudden stops (the stance of fiscal policy is measured as the change in the structural primary balance).
- Cross-sectional regressions on the output costs of sudden stops:
  - More flexible exchange rate regimes and increases in the terms of trade reduce the costs of sudden stops. Coefficients on ∆ Terms of trade are statistically significant and negative across specifications (e.g., -1.780*** in one specification).
  - Other policy variables (∆ Monetary policy, ∆ Reserves, ∆ Capital Controls, ∆ Fiscal policy) generally do not show robust, statistically significant effects on the cross-sectional variation in output costs across specifications.

### The output costs of sudden stops
- The loss of output in LAC after a sudden stop averages 7.4 percent of the pre-sudden stop trend GDP, with a maximum of 29.3 percent.
- Output loss is computed as (Y* - Y)/Y* where Y* is the counterfactual level of real GDP had the country continued to grow at the pre-sudden stop trend growth rate (the average growth rate in the eight quarters before the sudden stop). The cumulative differences are calculated over the quarters after the onset for which growth remains below the pre-sudden stop trend.

### Synthesis of empirical implications
- Exchange rate flexibility:
  - Floating exchange rate regimes reduce the duration of sudden stops and the output losses associated with them.
  - The effects are quantitatively large: a flexible exchange rate regime increases the probability of exiting a sudden stop by at least 60 percentage points relative to a fixed exchange rate regime (the report uses both “60 percent” and “60 percentage points” in discussing magnitude).
- Monetary policy:
  - Monetary policy tightening shortens the duration of sudden stops: a rise in real interest rates of 3 percentage points reduces the duration of a sudden stop by close to 15 percent.
  - The findings do not corroborate the blanket view that tightening monetary policy during crises is always counterproductive.
- Fiscal policy, FX reserve sales, and capital controls:
  - Fiscal policy, sales of international reserves, and capital controls do not have statistically significant effects on the duration of sudden stops in the primary empirical specifications.
  - Fiscal responses may present offsetting forces (demand boost versus higher sovereign risk premia), which can mute net effects on exit probability and duration.
- Terms of trade:
  - Improvements in the terms of trade materially raise the probability of exiting sudden stops and reduce associated output losses.

### Policy implications and recommended instruments
- Policy frameworks that enhance exchange rate flexibility help mitigate the duration and costs of sudden stops.
- Monetary policy tightening can be effective in shortening sudden stops when it raises real interest rates sufficiently.
- Relying solely on capital account restrictions or reserve sales to shorten sudden stops is not supported by the empirical evidence in this analysis.
- Policy responses to external adverse shocks could include:
  - Coordination of monetary policy instruments,
  - Exchange rate flexibility,
  - Foreign exchange market interventions (FXI),
  - Macroprudential measures,
  - Capital flow management measures.
- When central banks in the region use FXI, clear communication is important to preserve the credibility of inflation targeting (IT) frameworks and avoid the appearance of conflicting objectives between FXI and IT.

*International Monetary Fund | October 2019*

### References

### capitalflow - References

### Key empirical findings on capital inflow surges and investment
- Evidence indicates large inflows and capital formation are highly positively correlated in LAC but not in other EMEs; pattern is particularly evident during the commodity super-cycle.
- For the average EME, investment growth increases by 1.7 percentage points after a capital flow surge (coefficient: 1.698; Panel A, Column (1); standard error: (1.828); not statistically significant).
- For the LA7 countries, investment growth accelerates sharply by 11 percentage points in the years following surges of gross capital inflows (coefficient: 11.03**; Panel A, Column (2); standard error: (4.236); ** p<0.05).
- The post-2000 period shows a larger difference:
  - Surges in the 1990s were not associated with higher investment growth in the LAC-7 countries.
  - In the 2000–18 period, surges were associated with large increases in investment growth.
- Shorter windows show stronger immediate investment responses:
  - Panel B (2 years prior vs. 2 years post): EMEs coefficient 2.868 (standard error (1.967)); LA7 coefficient 17.36*** (standard error (5.684); *** p<0.01).
  - The estimated differences are larger for the 2-year window than the 4-year window, and some estimates not significant at 4-year become significant at 2-year.
- Annex Table 1 (selected coefficients and statistics):
  - Panel A (4 years prior vs. 4 years post, 1990–2018): Post Capital Flows Surge: EMEs 1.698 (1.828); LAC 11.03** (4.236). Constant: EMEs 6.047*** (1.365); LAC 1.515 (3.157). Observations: EMEs 5047; LAC 2173. R-squared: EMEs 0.002; LAC 0.088.
  - Panel B (2 years prior vs. 2 years post, 1990–2018): Post Capital Flows Surge: EMEs 2.868 (1.967); LAC 17.36*** (5.684). Constant: EMEs 5.127*** (1.513); LAC -0.432 (4.403). Observations: EMEs 2674; LAC 0740. R-squared: EMEs 0.008; LAC 0.197.
  - (Table also reports sample splits for 1990–2000 and 2000–18 with coefficients and standard errors for columns (3)-(6) and respective observations and R-squared values.)
- LA7 countries typically have lower investment rates in pre-surge years compared to other EMEs, implying investment growth in LA7 is closely tied to the availability of external financing.

### Identification of surges and empirical specification
- Surge identification (modified Hausmann, Pritchett and Rodrik (2005) algorithm):
  - For every 4-year window, average growth rate of yearly gross capital inflows (GKF) is computed as:
    - ḡ_i^GKF = (1/4) ∑_{j=i}^{i+4} (GKF_j − GKF_{j−1}) / GKF_{j−1}
  - For every date t, growth acceleration is defined as dda g_i^GKF = ḡ_i^GKF − ḡ_{i−4}^GKF
  - A surge is identified when:
    - ḡ_i^GKF ≥ ḡ_{75}^GKF (growth rate higher than 75th percentile),
    - dda g_i^GKF ≥ dda g_{75}^GKF (growth acceleration higher than 75th percentile),
    - GKF_i ≥ 0 (gross capital inflows positive at episode start).
- Investment response estimation: simple OLS regression estimated:
  - IGR_{i} = α + β * POST_SURGE_{i} + ε_{i}
  - 훽 captures difference in investment growth between pre-surge and post-surge; relation is not causal and may reflect mutual dependence and omitted factors.
- Robustness: equation estimated for several time periods, country samples, and windows around the capital flow surge.

### Determinants of capital flows: data and econometric methodology
- Sample periods: 2000–18 and 2012–18.
- Data: Financial Flows Analytics (FFA) database (Bluedorn et al., 2013) for 165 countries at quarterly frequency starting in 1990, complemented with IMF's Balance of Payments Statistics, Haver Analytics, CEIC and EMED databases.
- FFA data focus: private capital flows within "other investment" category excluding flows to general government and monetary authorities, IMF lending and reserve asset accumulation.
- Push factors included:
  - (i) growth in advanced economies;
  - (ii) US 10-year government bond yields;
  - (iii) global risk aversion measured by the CBOE VIX index.
- Pull factor included: domestic growth. Domestic interest rates excluded due to endogeneity.
- Lagged dependent variable included to account for irreversibility of investment decisions and to attenuate omitted variable bias.
- Estimation approach: system-GMM (Arellano-Bond estimator) used because inclusion of lagged dependent variable biases standard fixed-effects estimators.

### Duration analysis of sudden stops: modeling choices
- Duration models used because “time to an event” distributions are non-symmetric and OLS is inappropriate.
- Survival function S(t) = P(T > t) = ∫_t^∞ f(z) dz; hazard function h(t) = −(dS(t)/dt)/S(t).
- Models:
  - Non-parametric: universal survival distribution, no controls.
  - Semi-parametric: non-parametric common baseline distribution shifted multiplicatively by controls.
  - Fully parametric: test and estimate functional forms for baseline distribution.
- Chapter uses parametric models and proportional hazard (PH) specification:
  - h(t_j | x_j) = h_0(t) * exp(x_j β)
  - Weibull baseline often used when data exhibits duration dependence with h_0(t) = θ t^{θ−1}, where θ captures duration dependence.

### Data sources and definitions used in sudden stops analysis
- Capital flows: Total Gross Non-Official Flows (ICAPFLP) in U. S. dollars from IMF's Financial Flows Analytics (FFA) database; deflated using the US GDP deflator.
- Real GDP in national currency units: primarily quarterly data from IMF's International Financial Statistics Database (IFS); Haver Analytics used when IFS missing or limited. Countries where Haver was used: AZE; BHR; BLR; BLZ; BRA; CHN; CMR; COL; DEU; DNK; FIN; GHA; GTM; HND; IDN; IND; ITA; JOR; JPN; KAZ; KWT; LKA; LSO; MEX; MNE; MNG; MOZ; NAM; NGA; NIC; PAN; SLV; UGA; URY; VNM; ZAF; ZMB.
- Real GDP per capita: Maddison Project database (Bolt et al., 2018).
- International reserves: Official reserve assets in millions of US dollars from IMF’s IFS database.
- Monetary policy rates: policy rates from Thomson Reuters Datastream; IMF IFS used for money market rates and discount rates when policy rates are not available. Real (ex-post) interest rate calculated using CPI inflation from IMF's IFS database.
- Fiscal balance: General government primary net lending/borrowing as a share of GDP from IMF's World Economic Outlook (WEO) database; regressions use residuals from OLS of a constant on real GDP growth and commodity terms of trade growth to control for automatic stabilizers.
- Capital account restrictions: index of de jure capital account openness by Chinn and Ito (2006) based on IMF’s AREAER; higher index = more open capital account.
- Terms of trade: log of the commodity net export price index constructed by Gruss (2014).
- Exchange rate regimes: coarse de facto classification from Ilzetzki, Reinhart, and Rogoff (2017); categories 1 and 2 = fixed-exchange rate regimes; categories 3 and 4 = floating exchange rate regimes.

*Source: capitalflow - References (capitalflows chapter annexes and bibliography), capital flow analysis and annexes as provided.*

---


_Source: https://www.imf.org/-/media/files/publications/reo/whd/2019/october/english/capitalflow.pdf_