## _wp1234

## Source details

**Canonical URL:** [_wp1234](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1234.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1234.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1234.pdf.json)

---

### I. Introduction — purpose and framing
- Objective: reassess drivers of emerging market economies’ (EMEs) international reserves accumulation over 1980-2010.
- Central hypothesis: multiple motives operate simultaneously—exchange rate regime needs, precautionary demand (against current and capital account shocks), and mercantilist by-products of undervaluation—with varying importance across time and along the reserves distribution.
- Methodological innovations:
  - Nested model including exchange rate regime, insurance against current and capital account shocks, and mercantilist motives.
  - Undervaluation proxies based on three IMF assessment methodologies plus an alternate “excess” exports gravity-based proxy.
  - Use of quantile regressions to allow explanatory variable effects to vary systematically along the reserves distribution.

### II. Stylized facts on reserves accumulation
- Historical accumulation:
  - Reserves averaged about 5 percent of GDP in the 1980s and doubled every decade to reach some 25 percent of GDP by 2010.
  - Dispersion increased: difference between top and bottom quartiles widened from 3 percent of GDP in 1990 to 13 percent of GDP by 2010.
- Time-evolving explanatory power (bivariate R-squared patterns):
  - R-squared between imports and reserves declines only slightly across three periods (pre-Asian crisis; post-Asian crisis; emergence of global imbalances).
  - R-squared for capital account shock proxies rises from 0-10 percent in 1980-97 to 11-26 percent in 1998-2004, and 14-38 percent in 2005-10.
  - R-squared between reserves and the undervaluation measure is close to zero pre-Asian crisis, rising to around 4 percent afterwards.
- Cross-sectional notes:
  - About 40 percent of below-sample median reserves-to-GDP observations occur in the post-Asian crisis period.
  - 44 percent of above-median observations occur in the pre-Asian crisis sample.
- Interpretation:
  - Motives differ by time and by position in the reserves distribution; capital account shock variables and currency undervaluation are stronger determinants at higher reserves levels.

### III. Data, dependent variable, and sample
- Sample: panel of 43 emerging market economies over 1980-2010.
- Dependent variable: logarithm of gross international reserves, end-period stock, expressed in percent of GDP.
- Explanatory variable sets:
  - (i) Precautionary demand against current account shocks.
  - (ii) Precautionary demand against capital account shocks.
  - (iii) Mercantilist by-product measures (undervaluation and excess exports).

### IV. Precautionary and mercantilist proxies (construction and expectations)
- Current-account-related regressors and expected signs:
  - Imports/GDP — larger normal imports imply higher precautionary demand.
  - σ(exports/GDP) — 3-year moving standard deviation; higher ⇒ higher precautionary demand.
  - σ(Δypartner) — 3-year sd of partners’ real GDP growth; higher ⇒ higher precautionary demand.
  - Peg (hard or soft) — expected positive association.
  - σ(ΔNEER) — NEER volatility; higher ⇒ expected lower precautionary demand.
- Capital-account-related regressors and expected signs:
  - KA — Chinn-Ito index (de jure openness); higher ⇒ higher precautionary demand.
  - ShortDebt/GDP — short-term debt (residual maturity); higher ⇒ higher precautionary demand.
  - Money/GDP — banking system liabilities / broad money; higher ⇒ higher precautionary demand.
- Mercantilist proxies:
  - Undervaluation dummy based on average of three IMF CGER methodologies (macroeconomic balance, ERER, external stability) mapped into an indicator of more than 10 percent undervaluation.
  - Alternate proxy: “Excess” exports from a gravity model (standardized index in [-1,1]; dummy = 1 if positive).

### V. Empirical strategy
- Full model (equation (1)) includes:
  - Res/GDP (log); per capita income (log); population (log); Peg dummy; NEER volatility; Imports/GDP; Exports/GDP; partner growth volatility; KA; Short Debt/GDP; Money/GDP; Broad Debt measures; UnderVal; IntRateDiffer; additional interactions.
- Estimation steps:
  1. OLS over full sample, 1980-2010.
  2. Sub-samples: 1980-97; 1998-2004; 2005-2010.
  3. Quantile regressions (percentiles 0.05 to 0.95; reported quartiles q=25,50,75,90).
  4. Robustness tests (alternative mercantilism proxies, fixed effects, instrumentation, duration measures).

### VI. Main empirical findings — full sample (1980-2010)
- Scale-only regression:
  - Per capita income coefficient positive and significant; explains about 15 percent of variation in log(reserves/GDP).
- Adding exchange rate regime:
  - NEER volatility statistically significant; Peg dummy not significant.
- Interest rate differential (carry cost):
  - Statistically insignificant; raises R-squared marginally to about 17 percent.
- Adding current account precautionary variables:
  - R-squared increases to 47 percent.
  - Imports strongly related: one percentage point of GDP higher imports elicits an almost proportionate increase in reserves-to-GDP.
  - A one-standard deviation increase in each current account variable would (individually) raise implied reserves by about 2 percent of GDP (sample average about 13 percent of GDP).
- Adding capital account precautionary variables:
  - KA and banking system liabilities positive and significant; short-term debt positive but not significant.
  - Overall explanatory power rises to about 55 percent.
  - One-standard deviation increases in capital account variables individually raise reserves by 1.5 percent of GDP.
- Mercantilism (undervaluation) added:
  - Undervaluation (≥10 percent) associated with holding some 3 percent of GDP more reserves (other factors equal).
  - Incremental R-squared about 2 percent.
- Full model:
  - Explains about 56 percent of variation without annual or country fixed effects.
  - Most coefficients expected and significant except exchange rate regime variables, short-term debt, and interest rate differential (insignificant).

### VII. Sub-period evolution of motives
- Sub-periods: 1980-97; 1998-2004; 2005-10. Model’s explanatory power similar across periods but relative importance shifts.
- 1980-97 (pre-Asian crisis):
  - Exchange rate regime and current account precautionary motives dominate.
  - Peg dummy positive and significant; imports coefficient = 0.68 (highly significant).
  - Volatility of exports and partner growth significant.
  - Capital account variables small or insignificant.
  - Undervaluation coefficient insignificant.
  - One-standard deviation effects: current account ⇒ 1.2 percent of GDP; capital account ⇒ 0.5 percent of GDP; undervaluation ⇒ ~0.5 percent of GDP.
- 1998-2004 (post-Asian crisis):
  - Current account precautionary importance declines; capital account variables grow in significance and magnitude.
  - Currency undervaluation becomes positive and statistically significant.
- 2005-10 (global imbalances):
  - Exchange rate regime and most current account variables not significant (imports coefficient diminishes).
  - Broad money (banking liabilities) coefficient increases; short-term debt smaller but significant.
  - Financial openness becomes insignificant.
  - Interest rate differential larger in absolute value but statistically insignificant.
  - Undervaluation coefficient positive, larger, and highly significant.
  - One-standard deviation effects: current account ⇒ 1.6 percent of GDP; capital account ⇒ 2.0 percent of GDP; undervaluation ⇒ 7 percent of GDP.
- Chinn-Ito index averages (sample of EMEs):
  - -0.7 during the 1980s;
  - -0.1 in the 1990s;
  - +0.7 in the 2000s.
- Decomposition post-1998:
  - Current account variables explain about 30 percent of adjusted-overall R-squared (vs 39 percent pre-1998).
  - Capital account variables explain about 55 percent (vs 40 percent pre-1998).

### VIII. Quantile regressions — heterogeneity across the reserves distribution
- Purpose: identify determinants for low vs high reserve holders without selection bias.
- Key quantile findings:
  - Many regressors insignificant in OLS become significant at specific quantiles.
  - Exchange rate regime and current account precautionary motives more important for low reserve holders than high reserve holders.
  - Example: coefficient on imports declines from 0.85 for the lowest quartile to lower values at higher quantiles (full numeric progression in source tables).
  - Volatility of exports: significant for below-median reserve holders; not significant for above-median holders.
  - Volatility of partner growth: coefficient decreases with rising reserves; statistically insignificant throughout distribution.
  - Capital account variables:
    - Lower end: financial openness and short-term debt large and significant.
    - Above-median: these coefficients smaller and/or insignificant.
  - Broad money (banking liabilities): coefficient increases along reserves distribution; for high-reserve holders banking liabilities are more important.
    - Average shares for high-reserve group: banking system liabilities average 70 percent of GDP; short-term debt averages 17 percent of GDP.
  - Carry cost (interest rate differential):
    - Insignificant in OLS; negative and significant for above-median reserve holders, consistent with absolute carry cost proportional to amount held.
  - Exchange rate undervaluation:
    - Coefficient always positive and highly significant; declines along the reserves distribution.
    - Low-reserve undervaluation often reflects collapsed real exchange rates after crises; low-reserve undervaluation associated with reserves declining on average by about 2 percent.
    - High-reserve undervaluation associated with increasing reserves of about 15 percent.

### IX. Robustness checks and specification tests
- Alternative undervaluation proxies:
  - PPP-based misalignment and excess exports yield positive significant coefficients in OLS.
  - In quantile regressions PPP-based measure gives counterintuitive strong negative results for high-reserve holders, highlighting fragility of PPP measures.
- Sample endpoints:
  - Stopping sample in 2007 or 2008 makes virtually no difference.
- Fixed effects:
  - Adding annual fixed effects: little difference; scale variables lose significance; R-squared increases by 6 percent.
  - Adding country fixed effects: R-squared rises from 56 percent to 80 percent; most coefficients largely unaffected except partner growth losing significance.
  - Quantile regressions with time and country fixed effects do not appreciably alter main results.
- Endogeneity / instrumentation:
  - Instrumenting with second and third lags yields coefficients very similar to baseline; exogeneity tests of instruments not rejected (Hansen test p-value reported as 0.133; Endogeneity test p-value 0.531 in Table 3).
  - Instrumenting quantile regressions preserves main findings.
- Interaction terms:
  - Interaction of undervaluation with peg dummy, capital account openness, and imports produced no statistically significant effects in OLS and period regressions.

### X. Stock versus change in reserves and undervaluation duration
- Distinction:
  - Precautionary motives imply demand for a stock of reserves; mercantilist/export-led accumulation implies stock relates to duration of undervaluation.
- Duration measures (Table 5):
  - Cumulative number of years undervalued.
  - Same but reset to zero in any year not undervalued.
- Findings:
  - Duration of undervaluation matters for average stock, in period analysis post-Asian crisis, and across all quantiles.
  - Results in Tables 1 and 2 broadly preserved when using duration measures.
  - Partner growth volatility becomes statistically significant across quantiles with greater effect on low-reserve holders.
- Descriptive stats on undervaluation duration:
  - About 35 percent of observations are undervalued.
  - About half of countries undervalued for at least 5 years.
  - About 15 percent undervalued for at least 10 years.

### XI. Model fit and diagnostics
- Time-series fit (descriptive):
  - Large negative residuals after 1982 debt crisis until early 1990s: EMEs held fewer reserves than precautionary model predicts.
  - From early 1990s onwards, reserves increase faster than model predicts; insurance against capital account shocks dominates after the Asian crisis.
  - Undervaluation becomes progressively important in the latter half of the 2000s.
  - Average residual becomes positive around 2000, except in 2008 when EMEs ran down reserves.
- Cross-sectional fit:
  - Quantile regressions improve fit relative to OLS, especially for low-reserve observations.
  - Mean absolute deviation smaller using quantile regressions; largest improvements for low-reserve holders.
  - Country-level: bottom reserve holders fit best with 25th percentile betas; top reserve holders fit best with 90th percentile betas (examples in source: Mexico, Ecuador, Dominican Republic at bottom; China, Russia, Malaysia, Lebanon at top).
- Select figures and reported bivariate relationships (examples preserved exactly as in source):
  - 1980-1997, Log(Reserves to GDP) vs. Log(Imports to GDP): y = 0.71***x - 1.82***, R² = 0.26, Corr =  0.51.
  - 2005-2010, Log(Reserves to GDP) vs. Log(Broad Money to GDP): y = 0.67***x - 1.31***, R² = 0.38, Corr =  0.62.
  - 1980-1997, Log(Reserves to GDP) vs. Undervaluation: y =  0.002***x - 2.75***, R² =  0.02, Corr =  0.13.
  - 2005-2010, Log(Reserves to GDP) vs. Undervaluation: y = 0.01*x - 1.77***, R² =  0.04, Corr =  0.20.

### XII. Summary conclusions and implications
- Multiple motives are required to account for EME reserve accumulation:
  - Precautionary demand against current account shocks.
  - Precautionary demand against capital account shocks.
  - Intentional or unintentional exchange rate undervaluation (mercantilist/export-led factors).
- Temporal and distributional patterns:
  - Current account insurance relatively more important in early sample years and for low-reserve holders.
  - Capital account insurance gained importance after the 1997-98 Asian crises and for high-reserve holders.
  - Undervaluation important both as post-crisis collapsed real exchange rates for low-reserve holders and as deliberate mercantilist intervention for some high-reserve holders in the 2000s.
- Residual interpretation:
  - A positive residual remains in recent years (except during the global financial crisis), suggesting EMEs may be more risk averse or expect larger potential shocks than before.
- Limitations and next steps:
  - The paper does not assess adequacy of EME reserve holdings for the identified purposes; left for future research.

*Source: _wp1234 (excerpts from the PDF chapter covering introduction, stylized facts, data, specification, empirical results, robustness checks, and references).*

### References .............................................................................................................

### _wp1234 - References

### I. Introduction — purpose and framing
- Objective: reassess drivers of emerging market economies’ (EMEs) international reserves accumulation over 1980-2010.
- Central hypothesis: multiple motives—exchange rate regime needs, precautionary demand (against current and capital account shocks), and mercantilist by-products of undervaluation—operate simultaneously and with varying importance across time and along the reserves distribution.
- Key methodological innovations used in the paper:
  - Use of a nested model that includes exchange rate regime, insurance against current and capital account shocks, and mercantilist motives.
  - Improved undervaluation proxies based on IMF assessment methodologies (three approaches) and an alternate proxy of “excess” exports from a gravity model.
  - Use of quantile regressions to allow explanatory variable effects to vary systematically along the reserves distribution.

### II. Stylized facts on reserves accumulation
- Historical pattern:
  - Reserves averaged about 5 percent of GDP in the 1980s and doubled every decade to reach some 25 percent of GDP by 2010.
  - Dispersion increased: difference between top and bottom quartiles widened from 3 percent of GDP in 1990 to 13 percent of GDP by 2010.
- Time evolution of determinants (bivariate R-squared patterns):
  - R-squared between imports and reserves declines only slightly across three periods (pre-Asian crisis; post-Asian crisis; emergence of global imbalances).
  - R-squared for capital account shock proxies rises from 0-10 percent in 1980-97 to 11-26 percent in 1998-2004, and 14-38 percent in 2005-10.
  - R-squared between reserves and the undervaluation measure is close to zero pre-Asian crisis, rising to around 4 percent afterwards.
- Cross-sectional dispersion notes:
  - About 40 percent of below-sample median reserves-to-GDP observations occur in the post-Asian crisis period.
  - 44 percent of above-median observations occur in the pre-Asian crisis sample.
- Interpretation:
  - Low reserves-to-GDP observations are not confined to early sample years; motives differ by time and by position in the reserves distribution.
  - Variables related to capital account shocks (broad money and short-term debt) and currency undervaluation are stronger determinants at higher reserves levels.

### III. Data, dependent variable, and sample
- Sample: panel of 43 emerging market economies over 1980-2010.
- Dependent variable: logarithm of gross international reserves, end-period stock, expressed in percent of GDP.
- Three sets of explanatory variables included:
  - (i) Precautionary demand against current account shocks.
  - (ii) Precautionary demand against capital account shocks.
  - (iii) Mercantilist by-product measures (undervaluation and excess exports).

### IV. Precautionary motive proxies (current account)
- Current-account-related regressors included and their expected sign:
  - Imports/GDP — larger normal imports imply higher precautionary demand.
  - σ(exports/GDP) — three-year moving standard deviation of exports-to-GDP; higher volatility implies higher precautionary demand.
  - σ(Δypartner) — three-year standard deviation of trading partners’ real GDP growth; higher volatility implies higher precautionary demand.
  - Peg (dummy for hard or soft pegs) — expected to be associated with greater demand for precautionary reserves.
  - σ(ΔNEER) — nominal effective exchange rate volatility; higher volatility expected to be associated with lower precautionary demand.

### V. Precautionary motive proxies (capital account)
- Capital-account-related regressors included and their expected sign:
  - KA — Chinn-Ito index (de jure openness of the capital account from IMF Annual Report on Exchange Arrangements and Restrictions); higher openness implies higher precautionary demand.
  - ShortDebt/GDP — short-term debt on a residual maturity basis; higher short-term debt implies higher precautionary demand.
  - Money/GDP — banking system liabilities/broad money; larger banking liabilities imply higher precautionary demand.

### VI. Mercantilist motive proxies and construction
- Undervaluation measure construction:
  - Based on three methodologies underlying IMF exchange rate assessments (Lee et al., 2008):
    - Macroeconomic balance approach: undervaluation if a real appreciation is required to close the gap between actual and norm current account.
    - Equilibrium real exchange rate approach: compare real exchange rate to equilibrium implied by fundamentals (productivity differentials, terms of trade, government consumption, net foreign asset position).
    - External stability approach: compare the real exchange rate to the level that would stabilize the net foreign asset position at a benchmark.
  - The three implied misalignments are averaged and mapped into an indicator of more than 10 percent undervaluation (Undervaluation).
- Alternate mercantilist proxy:
  - “Excess” exports — the difference between actual exports and those predicted by a gravity model of trade.

### VII. Main empirical insights (summary of results described)
- Joint explanatory power:
  - Precautionary and mercantilist motives together can explain a large portion of the sample variation in EME reserve holdings.
- Heterogeneity across the reserves distribution:
  - Low reserves-to-GDP countries: precautionary motives mainly centered on current account shocks.
  - High reserves-to-GDP countries: precautionary motives center more on capital account shocks (e.g., banking system liabilities).
  - Carry costs of reserves (proportional to amount held) are pertinent primarily for countries with relatively high reserves-to-GDP ratios.
  - Currency undervaluation is relevant across the reserves distribution, though in some low-reserve cases it may reflect collapsed real exchange rates after crises rather than deliberate mercantilism.
- Time variation:
  - Pre-1997 (early sample): insurance against current account shocks is the most important determinant of reserves.
  - Post-Asian crisis: insurance against capital account shocks gains importance; mercantilist motives through undervaluation become relevant.

### VIII. Methodological implications
- Need for richer undervaluation proxies:
  - Simple PPP-based misalignment measures ignore other fundamentals and can be unreliable, especially among countries with similar income levels.
- Importance of allowing heterogeneity:
  - Standard OLS (average effects) can yield biased results if motives differ across the reserves distribution.
  - Quantile regression techniques capture both which variables explain high vs. low reserves and how variables differentiate within those groups.

*Source: _wp1234 - References (excerpts from the PDF chapter covering introduction, stylized facts, data, and specification).*

### Appendix A. In our robustness tests, we use an alternative proxy of mercantilism, namely

### Appendix A. In our robustness tests, we use an alternative proxy of mercantilism, namely whether the country exports more than would be expected on the basis of a trade gravity model (Qureshi and Tsangarides, 2010): excess exports

### Robustness proxies and conceptual issues
- Alternative proxy of mercantilism used: whether the country exports more than would be expected on the basis of a trade gravity model (Qureshi and Tsangarides, 2010): excess exports.
- Limitation: these proxies cannot distinguish between undervalued exchange rates resulting from crises and those that correspond to deliberate mercantilism.
- Distinction noted: undervaluation of the exchange rate is associated with the accumulation of reserves, not with the holding of a stock per se (in contrast, precautionary motives give rise to the demand for a stock of reserves).
- If reserves are accumulated as a by-product of deliberate mercantilism, over time this will be reflected in a rising stock of reserves, which is captured in the baseline regression.
- Robustness check considered: the effect of the accumulation of periods of undervalued exchange rate on the stock of reserves.

### Model specification and added controls
- Costs of holding reserves captured by: the interest rate differential between the Treasury bill rate in the country and the US short-term T-bill rate (the typical return on reserve assets).
- Scale controls included because dependent variable is percent of GDP: per capita income and population to allow for non-homogenous scale effects as income rises.
- Baseline regression excludes time and country fixed effects to explain rise in average reserve holdings and dispersion across countries; robustness tests add both fixed effects.
- Full model (equation (1)) includes:
  - Res/GDP (log)
  - per capita income (log), population (log)
  - Peg (exchange rate regime dummy), NEER volatility (log or measure), Imports/GDP, Exports/GDP
  - y_partner volatility measures (ln/Δ)
  - K A Short Debt/GDP, Money/GDP, Broad Debt measures
  - UnderVal (undervaluation dummy or measure)
  - IntRateDiffer (interest rate differential)
  - Additional terms and interactions as specified in the source equation (1).

### Empirical analysis: overview
- Estimation strategy proceeds in four steps:
  1. Estimate equation (1) using OLS over full sample, 1980-2010.
  2. Consider sub-samples: 1980-97 (pre-Asian crisis); 1998-2004 (post-Asian crisis); 2005-2010 (global imbalances).
  3. Quantile regressions to allow regressors' effects to vary with the amount of reserves held.
  4. Report robustness tests, including alternative mercantilism proxies and fixed effects.

### A. Full sample (1980-2010): stepwise findings
- Scale variables only (per capita income and population) (Table 1 [1]):
  - Coefficient on per capita income is positive and statistically significant.
  - Regression explains about 15 percent of the variation of the (logarithm) of the reserves-to-GDP ratio.
- Adding exchange rate regime variables (Table 1 [2]):
  - Nominal exchange rate volatility is statistically significant.
  - Pegged exchange rate regime dummy is not statistically significant.
- Interest rate differential (carry cost proxy):
  - Statistically insignificant.
  - Adding it raises R-squared only marginally to about 17 percent.
- Adding current account precautionary variables (Table 1 [3]):
  - R-squared increases to 47 percent (more than doubles).
  - Reserve holdings strongly related to imports: one percentage point of GDP higher imports elicit an almost proportionate increase in reserves-to-GDP.
  - Volatility of trading partner growth is statistically significant; volatility of the country’s own exports is not (but becomes significant when controlling for other variables).
  - A one-standard deviation increase in each current account variable would (individually) raise implied reserves by about 2 percent of GDP (sample average about 13 percent of GDP).
- Adding capital account precautionary variables (Table 1 [4]):
  - De jure financial openness and banking system liabilities are positive and statistically significant.
  - Short-term debt coefficient positive but not statistically significant.
  - Overall explanatory power rises to about 55 percent.
  - One-standard deviation increases in capital account variables individually raise reserves by 1.5 percent of GDP.
- Mercantilism (undervaluation) added to exchange rate variables (Table 1 [5]):
  - Proxy for undervaluation is strongly related to reserve holdings.
  - A country whose currency is undervalued by (at least) 10 percent would, other factors equal, hold some 3 percent of GDP more foreign exchange reserves than a country without an undervalued currency.
  - Contribution to explanatory power is modest (incremental R-squared of about 2 percent), possibly due to limited variation of the undervaluation dummy.
- Full model with all variables (Table 1 [6]):
  - Explains about 56 percent of the variation in reserve holdings percent without annual or country fixed effects.
  - Most coefficients are of expected sign and statistically significant, except:
    - Exchange rate regime variables (pegs/volatility),
    - Short-term debt,
    - Implied carry cost / interest rate differential (all statistically insignificant).

### B. Sub-period analysis: evolving motives (Tables 1 [7]-[9])
- Sub-periods: 1980-97; 1998-2004; 2005-10.
- General: model’s explanatory power roughly the same across periods, but relative importance of motives changes over time.

- 1980-97 (pre-Asian crisis; Table 1 [7]):
  - Dominance of exchange rate regime and current account precautionary motives.
  - Pegged exchange rate regime dummy positive and significant; nominal exchange rate volatility not significant.
  - Imports coefficient = 0.68 (highly significant), larger than in later periods (0.47).
  - Volatility of exports and volatility of partner country growth statistically significant in this period but not later.
  - Capital account variables either insignificant (short-term debt) or smaller coefficients (broad money) than subsequent periods.
  - Undervaluation coefficient economically and statistically insignificant.
  - One-standard deviation increases: current account variables => 1.2 percent of GDP; capital account variables => 0.5 percent of GDP; undervaluation => ~0.5 percent of GDP.

- 1998-2004 (post-Asian crisis; Table 1 [8]):
  - Precautionary reserves against current account shocks become less important (imports coefficient falls; volatility of exports and partner growth turn insignificant).
  - Capital account variables (financial openness, broad money, short-term debt) become statistically significant or larger.
  - Currency undervaluation becomes positive and statistically significant.

- 2005-10 (global imbalances; Table 1 [9]):
  - Exchange rate regime and most current account precautionary variables not statistically significant (except imports, whose coefficient diminishes further).
  - Capital account: broad money coefficient increases; short-term debt slightly smaller; both statistically significant.
  - Financial openness becomes insignificant (possibly because most EMEs largely open capital accounts by this period).
  - Interest rate differential coefficient larger in absolute value but remains statistically insignificant.
  - Undervaluation coefficient positive, larger than previous periods, and highly statistically significant.
  - One-standard deviation increases: current account variables => 1.6 percent of GDP; capital account variables => 2.0 percent of GDP; undervaluation => 7 percent of GDP.

- Interpretation:
  - Motives shift over time: 1980s–1990s reserves to defend pegs / buffer current account shocks; post-Asian crisis rise in insurance against capital account shocks; mercantilism (undervaluation) contributes to higher reserves, particularly in 2005-10.
  - Different reserve levels across countries at a point in time may correspond to different reasons for holding them.

- Additional numeric context:
  - The Chinn-Ito capital account openness index (ranges from -1 to +1) averages:
    - -0.7 during the 1980s;
    - -0.1 in the 1990s;
    - +0.7 in the 2000s for the sample of EMEs.

- Decomposition note:
  - Post-1998, current account variables explain about 30 percent of the “adjusted-overall” R-squared (compared to 39 percent pre-1998), while capital account variables explain about 55 percent compared to 40 percent in the pre-1998 period.

### C. Quantile regressions: heterogeneity across the reserves distribution
- Objective: test whether determinants differ for low vs high reserve holders using quantile regressions (Koenker and Bassett, 1978).
- Quantile regressions estimate conditional functions at any point of reserves’ distribution and avoid sample selection bias from OLS on subsamples.
- Estimated quartiles (and broader percentiles from 0.05 to 0.95) via equation (2) analogous to (1) but indexed by quantile q (25, 50, 75, 95 noted).
- Key findings:
  - Some variables show common effects, but several show important differences across quantiles.
  - Many regressors that were insignificant in OLS for the average reserve holder become highly significant at various points in the reserve distribution.
  - Exchange rate regime and precautionary demand against current account shocks are more important for low reserve holders than for high reserve holders.
  - Example: coefficient on imports declines from 0.85 for observations in the lowest quartile (Table 2 [2]) to lower values at higher quantiles (trend described but full numeric progression in source tables).
- Visualization:
  - Figure 4 traces coefficient changes across percentiles (0.05 to 0.95) with confidence intervals; dashed lines show OLS averages for comparison.
  - Many graphs show little overlap between quantile estimates and average OLS estimates, indicating differences in magnitude and significance across the reserves distribution.

*Source: Appendix A and IV. Empirical Analysis from the provided IMF content unit.*

### 0.37 for observations above the 75th percentile (Table 2 [5]), with the differences in coefficients

### _wp1234 - 0.37 for observations above the 75th percentile (Table 2 [5]), with the differences in coefficients

### Distributional determinants of reserve holdings
- Volatility of exports:
  - Significant determinant for below-median reserve holders (Table 2 [2]-[3]).
  - Not significant for above-median reserve holders (Table 2 [4]-[5]).
- Volatility of partner-country growth:
  - Coefficient decreases with rising reserve holdings, but is statistically insignificant throughout the distribution.
- Capital account variables:
  - At the lower end (Table 2 [2]-[3]) financial openness and short-term debt coefficients are large and statistically significant.
  - For above-median reserve holders (Table 2 [4]-[5]) these coefficients are smaller and/or insignificant.
- Broad money (banking system liabilities):
  - Coefficient increases along the reserves distribution.
  - For high-reserve holders, banking system liabilities are a more important determinant of reserves accumulation than for low-reserve holders.
  - Average shares (for high-reserve group): banking system liabilities average 70 percent of GDP; short-term debt averages 17 percent of GDP.
- Cost of holding reserves (interest rate differential):
  - Insignificant in OLS regressions.
  - Becomes negative and statistically significant for above-median reserve holders (Table 2 [4]-[5]).
  - Interpretation: absolute carry cost proportional to reserve holdings; more relevant as reserves increase.
- Exchange rate undervaluation:
  - Coefficient always positive and highly significant (Table 2 [1]-[4]).
  - Coefficient declines along the reserves distribution.
  - For low-reserve holders, undervaluation often reflects collapsed real exchange rates after the 1980s debt crisis and other EME crises rather than deliberate sterilized intervention.
  - Empirical associations:
    - Low-reserve holders with undervalued exchange rates: reserves declined on average by about 2 percent.
    - High-reserve holders with undervaluation: associated with increasing reserves of about 15 percent.

### Robustness tests and specification checks
- Alternative undervaluation proxies:
  - Replacing the undervaluation variable with a PPP-based measure (Table 3 [2]) or “excess” exports (Table 3 [3]) yields a positive and statistically significant coefficient in OLS.
  - In quantile regressions, PPP-based measure produces counterintuitive results for high-reserve holders (strongly negative and significant), highlighting fragility of PPP-based measures (Table 4 [5]-[8]).
- Sample period sensitivity:
  - Stopping the sample in 2007 or 2008 (Table 3 [4]-[5]) instead of 2010 makes virtually no difference.
- Fixed effects:
  - Baseline excludes time and country fixed effects to test explanatory power without trends or country dummies.
  - Adding annual fixed effects (Table 3 [6]) makes little difference; scale variables (per capita income and population) lose significance; regression R-squared increases by 6 percent.
  - Adding country fixed effects (Table 3 [7]) raises regression R-squared from 56 percent to 80 percent; individual coefficients largely unaffected except partner country growth losing significance. Volatility of nominal exchange rate, imports, volatility of exports, broad money, and exchange rate undervaluation remain statistically significant.
  - In quantile regressions, introducing time (Table 4 [9]-[12]) and country (Table 4 [13]-[16]) fixed effects does not appreciably alter results.
- Endogeneity and instrumentation:
  - Instrumenting potentially endogenous regressors with second and third lags (Table 3 [8]) yields coefficients very similar to baseline; volatility of nominal exchange rate flips sign and volatility of partner country growth loses significance.
  - Exogeneity tests of instruments not rejected.
  - Instrumenting in quantile regressions similarly does not change the main findings: for below-median reserve holders (Table 4 [17]-[18]), current account precautionary variables remain statistically significant and larger than for above-median holders (Table 4 [19]-[20]).
- Interaction terms:
  - Interaction of undervaluation with peg dummy, capital account openness, and imports produced no statistically significant effects in OLS and period regressions.

### Stock versus change in reserves; undervaluation duration
- Conceptual point:
  - Baseline specification assumes explanatory variables influence the stock of reserves (consistent with precautionary motives).
  - Mercantilist/export-led accumulation implies stock should relate to duration of undervaluation.
- Two duration measures introduced (Table 5):
  - Cumulative number of years a country is undervalued.
  - Same measure but reset to zero in any year not undervalued.
- Empirical findings:
  - Duration of undervaluation matters for the average stock of reserves, in period analysis post-Asian crisis, and across all quantiles.
  - Results in Tables 1 and 2 broadly preserved with duration measures.
  - Partner growth volatility becomes statistically significant across all quantiles, with greater effect on low-reserve holders.
  - Volatility of exchange rate not important for reserve holders beyond the 25th percentile.
- Descriptive stats on undervaluation duration:
  - About 35 percent of observations are undervalued (45, aligned).
  - About half of countries are undervalued for at least 5 years.
  - About 15 percent are undervalued for at least 10 years.

### Model fit and accounting for EME reserve holdings
- Time-series fit (Figure 5, descriptive):
  - Cumulative change in reserves (percent of GDP, averaged across countries) and fitted quantile regression values show:
    - Large negative residuals following the 1982 onset of the debt crisis until the early 1990s: EMEs involuntarily held fewer reserves than precautionary model predicts.
    - During 1980s–early 1990s, undervaluation contributes positively to predicted reserve increases (likely reflecting post-crisis real exchange rate declines).
    - Precautionary demand against current account shocks contributes negatively in that period.
    - From early 1990s onwards, reserve holdings increase faster than model predicts; insurance against capital account shocks is the most important contributor, especially after the Asian financial crisis.
    - Undervaluation becomes progressively important in the latter half of the 2000s, likely reflecting intentional undervaluation in the context of rising reserves.
    - Average residual becomes positive starting around 2000, except in 2008 when EMEs ran down reserves.
- Cross-sectional fit (Figure 6 and Figure 7a/b, descriptive):
  - Quantile regressions improve fit relative to OLS, especially for observations in the bottom and/or lowest quartile.
  - Mean absolute deviation consistently smaller using quantile regressions; largest improvements for low-reserve holders.
  - Country-level comparisons (circa 2007):
    - Bottom reserve holders (e.g., Mexico, Ecuador, Dominican Republic) fit best using 25th percentile betas.
    - Top reserve holders (e.g., China, Russia, Malaysia, Lebanon) fit best using 90th percentile betas.
    - OLS can misrepresent country-specific fits: e.g., China appears to have significantly more reserves than predicted by OLS, but aligns well when compared using the 90th percentile estimates.

### Conclusions and interpretation
- Multiple motives required to account for EME reserve accumulation:
  - Precautionary demand against current account shocks.
  - Precautionary demand against capital account shocks.
  - Intentional or unintentional exchange rate undervaluation (mercantilist/export-led factors).
- Temporal and distributional patterns:
  - Current account insurance relatively more important in early sample years and for low-reserve holders.
  - Undervaluation important for low-reserve holders, often reflecting post-crisis collapsed real exchange rates rather than deliberate mercantilism.
  - After 1990s capital account crises (especially 1997-98 Asian crises), insurance against capital account shocks became more important.
  - From early 2000s, undervaluation again important in the context of rising reserves, suggesting deliberate sterilized intervention in some cases.
- Residual patterns and implications:
  - Even allowing for different motives across the reserves distribution, a positive residual remains in recent years (except during the global financial crisis).
  - Possible interpretations: EMEs are becoming more risk averse, or potential shocks are larger than previously believed.
- Limitations and next steps:
  - Paper does not assess adequacy of EME reserve holdings for the identified purposes; this is left to future research.

*Source: _wp1234 - 0.37 for observations above the 75th percentile (Table 2 [5]), with the differences in coefficients*

### REFERENCES

### _wp1234 - REFERENCES

### References
- Aizenman, Joshua, and Jaewoo Lee, 2008, “Financial Versus Monetary Mercantilism Long-Run View of Large International Reserves Hoarding,” The World Economy, Vol. 31, pp. 593–611.
- Aizenman, Joshua, and Jaewoo Lee, 2007, “International Reserves: Precautionary Versus Mercantilist Views, Theory and Evidence,” Open Economies Review, Vol. 18, pp. 191–214.
- Aizenman, Joshua, and Nancy P. Marion, 2003, “The High Demand for International Reserves in the Far East: What Is Going On?” Journal of the Japanese and International Economies, Vol. 17, pp. 370–400.
- Bar-Ilan, Avner, and Nancy P. Marion, 2009, “A Macroeconomic Perspective on Reserve Accumulation,” Review of International Economics, Vol. 17, pp. 802–23.
- Bastourre, Diego, Jorge Carrera, and Javier Ibarlucia, 2009, “What is Driving Reserves Accumulation,” Review of International Economics, Vol. 17, pp. 861–77.
- Ben-Bassat, Avraham, and Daniel Gottlieb, 1992, “Optimal International Reserves and Sovereign Risk,” Journal of International Economics, Vol. 33, pp. 345–62.
- De Beaufort Wijnholds, J. Onno, and Arend Kapteyn, 2001, “Reserve Adequacy in Emerging Market Economies,” IMF Working Paper 01/143 (Washington: International Monetary Fund).
- Caballero, Ricardo J., and Stavros Panageas, 2008, “Hedging Sudden Stops and Precautionary Contractions,” Journal of Development Economics, Vol. 85, pp. 28–57.
- Cheung Yin-Wong, and Hiro Ito, 2009, “Cross-Country Empirical Analysis of International Reserves,” CESifo Working Paper No. 2654 (Munich: Center for Economic Studies and Ifo Institute for Economic Research).
- Cheung Yin-Wong, and Xingwang Qian, 2009, “Hoarding of International Reserves: Mrs. Machlup’s Wardrobe and the Joneses,” Review of International Economics, Vol. 17, pp. 824–43.
- Chinn, Menzie D., and Hiro Ito, 2006, “What Matters for Financial Development? Capital Controls, Institutions, and Interactions,” Journal of Development Economics, Vol. 81, pp. 163–92.
- Courchene, Thomas J., and G.M. Youssef, 1967, “The Demand for International Reserves,” Journal of Political Economy, Vol. 75, pp. 404–13.
- Delatte, Anne-Laure, and Julien Fouquau, 2011, “What Drove the Massive Hoarding of International Reserves in Emerging Countries? A Time-Varying Approach,” Review of International Economics, forthcoming.
- Dooley, Michael P., David Folkerts-Landau, and Peter Garber, 2003, “An Essay on the Revived Bretton Woods System,” NBER Working Paper No. 9971 (Cambridge, Massachusetts: National Bureau of Economic Research).
- Dooley, Michael P., David Folkerts-Landau, and Peter Garber, 2005, International Financial Stability (New York: Deutsche Bank).
- Durdu, Ceyhun Bora, Enrique G. Mendoza, and Marco E. Terrones, 2009, “Precautionary Demand for Foreign Assets in Sudden Stop Economies: an Assessment of the New Mercantilism,” Journal of Development Economics, Vol. 89, pp. 194–209.
- Frenkel, Jacob, 1983, “International Liquidity and Monetary Control” in International Money and Credit: The Policy Roles, ed. by George von Furstenberg (Washington: International Monetary Fund).
- Frenkel, J. and B. Jovanovic, 1981, “Optimal International Reserves: a Stochastic Framework,” Economic Journal, Vol. 91, pp. 507–14.
- Edwards, Sebastian, 1983, “The Demand for International Reserves and Exchange Rate Adjustments: the Case of LDCs, 1964–1972,” Economica, Vol. 50, pp. 269–80.
- Ghosh, Atish R., Jonathan D. Ostry, and Charalambos Tsangarides, 2010, “Exchange Rate Regimes and the Stability of the International Monetary System,” IMF Occasional Paper No. 270 (Washington: International Monetary Fund).
- Hamada Koichi, and Kazuo Ueda, 1977, “Random Walks and the Theory of Optimal International Reserves,” Economic Journal, Vol. 87, pp. 722–42.
- Heller, Heinz Robert, 1966, “Optimal International Reserves,” Economic Journal, Vol. 76, pp. 296–311.
- International Monetary Fund, 2000, “Debt and Reserve-Related Indicators of External Vulnerability,” Available via the Internet: http://www.imf.org/external/np/pdr/debtres/index.htm.
- International Monetary Fund, 2001, “Issues in Reserves Adequacy and Management,” Available via the Internet: http://www.imf.org/external/np/pdr/resad/2001/101501.pdf.
- International Monetary Fund, 2011, “Assessing Reserves Adequacy,” available via the Internet: http://www.imf.org/external/np/pp/eng/2011/021411b.pdf.
- Jeanne, Olivier, 2007. “International Reserves in Emerging Market Countries: Too Much of a Good Thing?” Brookings Papers on Economic Activity: 1, Brookings Institution.
- Jeanne, Olivier, and Romain Rancière, 2006, “The Optimal Level of International Reserves for Emerging Market Countries: Formulas and Applications,” IMF Working Paper 06/229 (Washington: International Monetary Fund).
- Koenker, Roger, and Gilbert Bassett Jr., 1978, “Regression Quantiles,” Econometrica, Vol. 46, pp. 33–50.
- Lee, Jaewoo, Gian Maria Milesi-Ferretti, Jonathan D. Ostry, Alessandro Prati, and Luca A. Ricci, 2008, “Exchange Rate Assessments: CGER Methodologies,” IMF Occasional Paper No. 261 (Washington: International Monetary Fund).
- Rodrik, Dani, 2008, “The Real Exchange Rate and Economic Growth,” Brookings Papers on Economic Activity: 2, pp. 365–412.
- Koenker Roger, and Kevin F. Hallock, 2001, “Quantile Regression,” Journal of Economic Perspectives, Vol. 15, pp. 143–56.
- Obstfeld Maurice, Jay Shambaugh, and Alan M. Taylor, 2010, “Financial Stability, the Trilemma, and International Reserves,” American Economic Journal: Macroeconomics, Vol. 2, pp. 57–94.
- Sula, Ozan, 2008, “Demand for International Reserves in Developing Nations: A Quantile Regression Approach,” Journal of International Money and Finance, Vol. 30, pp. 764–77.
- Tsangarides, Charalambos, Pierre Ewenczyk, Michal Hulej, and Mahvash S. Qureshi, 2009, “Are Africa’s Currency Unions Good for Trade?” IMF Staff Papers, Vol. 56, pp. 876–918.
- Zalduendo, Juan, 2008, “Bivariate Assessments of Real Exchange Rates Using PPP Data” IMF Working Paper 08/153 (Washington: International Monetary Fund).

### Tables and Figures (titles and select reported summary statistics)
- Figure 1. International Reserves held by EME’s, 1980-2010 (in percent of GDP)
  - Depicts Reserves' inter-quartile range, Cross-country average, Cross-country median (Source: WEO and authors' calculations).

- Figure 2. Bivariate Relationships by Period
  - Reports regression lines, R² and Corr for multiple periods and bivariate pairs (examples preserved exactly as in source):
    - 1980-1997, Log(Reserves to GDP) vs. Log(Imports to GDP): y = 0.71***x - 1.82***, R² = 0.26, Corr =  0.51
    - 1998-2004, Log(Reserves to GDP) vs. Log(Broad Money to GDP): y = 0.53***x - 1.65***, R² = 0.26, Corr =   0.51
    - 2005-2010, Log(Reserves to GDP) vs. Log(Broad Money to GDP): y = 0.67***x - 1.31***, R² = 0.38, Corr =  0.62
    - 1980-1997, Log(Reserves to GDP) vs. Undervaluation: y =  0.002***x - 2.75***, R² =  0.02, Corr =  0.13
    - 2005-2010, Log(Reserves to GDP) vs. Undervaluation: y = 0.01*x - 1.77***, R² =  0.04, Corr =  0.20

- Figure 3. Bivariate Relationships by Quantile
  - Reports quantile-specific regression lines, R² and Corr (examples preserved exactly as in source):
    - up to 25th percentile, Log(Reserves to GDP) vs. Log(Imports to GDP): y = 0.18*** x - 3.23***, R² = 0.05, Corr = 0.21
    - 75th percentile onwards, Log(Reserves to GDP) vs. Log(Broad Money to GDP): y = 0.39*** x - 1.17***, R² = 0.41, Corr = 0.64
    - 75th percentile onwards, Log(Reserves to GDP) vs. Undervaluation: y = 0.005* x - 1.37**, R² = 0.08, Corr= 0.29

- Figure 4. Comparison of Quantile Regression and OLS Coefficient Estimates
  - Series of coefficient-by-quantile plots for variables including Log(percapita income), Log(population), Log(imports to GDP), Volatility of neer, Interest rate differential w/US, Volatility of exports/GDP (3-year sd), Financial openness, Log(broad money to GDP), Hard and Soft peg dummy, Short term debt to GDP, Volatility of partner growth (3-year sd), Undervaluation dummy2 (Source: Authors' calculations).

- Figure 5. Actual and Fitted Cumulative Change in Reserves
  - Time series of Actual and quantile regression fitted change in log(reserves/GDP) excluding effects of scale variables (population and per capita income) and exchange rate regime variables; components shown: Capital account, Current account, Mercantilist, Residual, Reserves-scale.

- Figures 6, 7a, 7b. Model fit diagnostics
  - Figure 6: Mean Absolute Deviations, Quantile Regressions and OLS by country (list of country codes and MAD comparisons).
  - Figures 7a and 7b: Actual vs. Fitted Values, Quantile Regressions (25th, 50th, 75th, 90th percentiles) and OLS for 2007; plotted by country (ARG, ARM, BGR, ... ZAF).

- Table 1. Current Account, Capital Account and Mercantilist Determinants of Reserve Demand
  - Sample columns (1) through (9) with reported coefficients and standard errors for variables including Log(per capita income), Log(population), Hard and Soft peg dummy, Volatility of neer, Interest rate differential w/ US, Log(imports to GDP), Volatility of exports/GDP (3-year sd), Volatility of partner growth (3-year sd), Financial openness, Log(broad money to GDP), Short term debt to GDP, Exchange rate undervaluation, Constant.
  - Observations: 996 (columns with varying samples down to 258).
  - R-squared values: 0.189, 0.281, 0.484, 0.546, 0.296, 0.562, 0.428, 0.519, 0.532.
  - Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

- Table 2. Reserves Demand Across Quantiles
  - Columns: qreg25, qreg50, qreg75, qreg90 (quantile regression estimates 1980-2010).
  - Coefficients and standard errors for Log(per capita income), Log(population), Hard and Soft peg dummy, Volatility of neer, Interest rate differential w/ US, Log(imports to GDP), Volatility of exports/GDP (3-year sd), Volatility of partner growth (3-year sd), Financial openness, Log(broad money to GDP), Short term debt to GDP, Exchange rate undervaluation, Constant.
  - Observations: 996 for each column.
  - Pseudo R2: 0.364, 0.370, 0.367, 0.382.
  - Notes: Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

- Table 3. Robustness of OLS Estimates
  - Columns (1) through (8) labeled sample variations (full, excess exports, pppu, upto 2007, upto 2008, FE & year FE, country variables, instrument variables).
  - Reported coefficients and standard errors for the same set of explanatory variables as Table 1 plus additional robustness variables (Excess Exports, Exchange rate undervaluation (ppp)).
  - Observations: 996, 950, 996, 867, 910, 996, 996, 873.
  - R-squared values: 0.562, 0.565, 0.553, 0.542, 0.547, 0.623, 0.800, 0.567.
  - Hansen test (p-value): 0.133.
  - Endogeneity test (p-value): 0.531.

*Source: _wp1234 - REFERENCES (PDF chapter/section).*

### 1. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

### _wp1234 - 1. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

### Robustness of Quantile Regressions (Table 4) — key coefficient patterns and statistics
- Sample size and structure:
  - Observations: 996 (columns 1–16), 891 (columns 17–20).
  - Pseudo R2: 0.364, 0.370, 0.367, 0.382, 0.357, 0.362, 0.355, 0.372, 0.194, 0.234, 0.292, 0.320, 0.105, 0.0971, 0.101, 0.114, 0.355, 0.347, 0.347, 0.374.
- Scale variables (selected coefficients with robust standard errors in parentheses):
  - Log(per capita income):
    - Column 1 (qreg25): 0.339*** (0.0486)
    - Column 2 (qreg50): 0.392*** (0.0252)
    - Column 3 (qreg75): 0.356*** (0.0313)
    - Column 4 (qreg90): 0.295*** (0.0386)
    - Additional reported values include: 0.288*** (0.0439), 0.410*** (0.0251), 0.350*** (0.0236), 0.317*** (0.0369), 0.0869 (0.0535), 0.158*** (0.0316), 0.165*** (0.0344), 0.218*** (0.0368), -0.199* (0.106), -0.1 (0.106), 57 (0.113), 0.0920 (0.146), 0.106 (0.0467), 0.330*** (0.0367), 0.369*** (0.0428), 0.318*** (0.0358), 0.250***.
  - Log(population):
    - Column 1: 0.187*** (0.0305)
    - Column 2: 0.148*** (0.0150)
    - Column 3: 0.0801*** (0.0177)
    - Column 4: 0.0330* (0.0194)
    - Additional reported values include: 0.169*** (0.0280), 0.145*** (0.0150), 0.0833*** (0.0139), 0.0430** (0.0194), 0.0824*** (0.0288), 0.0888*** (0.0163), 0.0419** (0.0174), -0.0129 (0.0162), -0.683** (0.284), -0.519* (0.279), -0.723*** (0.273), -1.632*** (0.312), 0.176*** (0.0287), 0.135*** (0.0215), 0.0609** (0.0250), 0.0244 (0.0179).
- Regime (Hard and Soft peg dummy):
  - Coefficients vary across specifications; examples include:
    - 0.0657 (0.0689), -0.0129 (0.0349), 0.0115 (0.0431), -0.0190 (0.0529)
    - 0.0319 (0.0625), -0.00802 (0.0347), -0.0289 (0.0327), -0.0707 (0.0517)
    - 0.140** (0.0633), 0.0934*** (0.0357), 0.137*** (0.0374), 0.144*** (0.0367)
    - 0.0715* (0.0424), 0.0812** (0.0375), 0.0 978*** (0.0350), 0.0610 (0.0475)
    - 0.0750 (0.0825), -0.0461 (0.0621), 0.0232 (0.0756), -0.0496 (0.0605).
- Volatility of neer (nominal effective exchange rate):
  - Examples across quantiles:
    - -0.00787* (0.00430), -0.00380* (0.00218), -0.00216 (0.00268), -0.00272 (0.00296)
    - -0.0117*** (0.00391), -0.00356 (0.00219), -0.000494 (0.00204), -0.00244 (0.00308)
    - -0.00807** (0.00387), -0.00148 (0.00222), -0.000113 (0.00245), 0.00360 (0.00240)
    - -0.0141*** (0.00238), -0.00421** (0.00210), -0.00430** (0.00219), -0.00606** (0.00300)
    - -0.00299 (0.00824), -0.00932 (0.00637), 0.00358 (0.00785), -0.00226 (0.00620).
- Opportunity cost (Interest rate differential w/ US):
  - Samples of coefficients and standard errors:
    - 0.0608 (0.0867), -0.138*** (0.0424), -0.112** (0.0565), -0.176*** (0.0513)
    - 0.0387 (0.0779), -0.135*** (0.0423), -0.149*** (0.0439), -0.232*** (0.0513)
    - 0.110 (0.0911), -0.0136 (0.0446), -0.111** (0.0498), -0.224*** (0.0546)
    - 0.0197 (0.0475), -0.0383 (0.0413), -0.0621 (0.0429), -0.0366 (0.0525)
    - 0.131 (0.0839), -0.0506 (0.0690), -0.124 (0.0884), -0.255*** (0.0664).
- Current account (Log(imports to GDP)):
  - Strong positive coefficients across quantiles and specifications:
    - 0.853*** (0.0765), 0.614*** (0.0390), 0.557*** (0.0461), 0.371*** (0.0493)
    - 0.773*** (0.0687), 0.599*** (0.0390), 0.487*** (0.0355), 0.374*** (0.0454)
    - 0.610*** (0.0706), 0.519*** (0.0415), 0.440*** (0.0424), 0.239*** (0.0389)
    - 0.730*** (0.0867), 0.622*** (0.0769), 0.459*** (0.0809), 0.428*** (0.102)
    - 0.948*** (0.0721), 0.621*** (0.0569), 0.569*** (0.0658), 0.380*** (0.0475).
- Volatility of exports/GDP (3-year sd):
  - Coefficients include:
    - 0.400*** (0.0725), 0.200*** (0.0416), 0.0700 (0.0434), -0.0819 (0.0513)
    - 0.347*** (0.0641), 0.183*** (0.0416), 0.0654 (0.100), 0.355*** (0.0999)
    - 0.337*** (0.0941), 0.195*** (0.0575), 0.0638 (0.0634), -0.104 (0.0814)
    - 0.324*** (0.0615), 0.241*** (0.0553), 0.0981* (0.0592), 0.0403 (0.0867)
    - 0.349*** (0.0754), 0.123* (0.0682), -0.0687 (0.0846), -0.215*** (0.0726).
- Capital account and financial variables:
  - Financial openness:
    - 0.154*** (0.0235), 0.0828*** (0.0118), 0.0381** (0.0149), 0.0502*** (0.0185)
    - Additional values include 0.164*** (0.0222), 0.0743*** (0.0119), 0.0524*** (0.0114), 0.0391** (0.0169)
    - 0.0827*** (0.0201), 0.0557*** (0.0122), 0.0614*** (0.0143), 0.0469*** (0.0146)
    - 0.0194 (0.0192), 0.00914 (0.0166), 0.000528 (0.0156), -0.0549*** (0.0180)
    - 0.153*** (0.0227), 0.0756*** (0.0169), 0.0301 (0.0208), 0.0449** (0.0177).
  - Log(broad money to GDP:
    - Coefficients consistently positive and significant across columns:
      - Examples: 0.335*** (0.0615), 0.320*** (0.0317), 0.358*** (0.0401), 0.377*** (0.0461)
      - Many repeated positive significant coefficients across specifications with standard errors reported.
  - Short term debt to GDP:
    - Examples: 0.504*** (0.177), 0.348*** (0.0968), 0.105 (0.140), -0.0261 (0.193)
    - Other reported patterns include sometimes positive and significant, and sometimes negative and significant (e.g., -0.251** (0.107), -0.340*** (0.130)).
- Mercantilist measures (Exchange rate undervaluation):
  - Exchange rate undervaluation:
    - 0.385*** (0.0676), 0.203*** (0.0355), 0.200*** (0.0456), 0.177*** (0.0526)
    - 0.371*** (0.0611), 0.197*** (0.0362), 0.195*** (0.0403), 0.0411 (0.0395)
    - 0.177*** (0.0371), 0.0773** (0.0354), 0.0782** (0.0373), 0.123** (0.0498)
    - 0.588** (0.0963), 0.296*** (0.0757), 0.280*** (0.0944), 0.274*** (0.0711).
  - Exchange rate undervaluation (ppp): 0.339*** (0.0659), 0.114*** (0.0377), 0.0293 (0.0361), -0.102** (0.0507).
- Constant terms:
  - Examples: -7.647*** (0.775), -7.167*** (0.383), -5.431*** (0.470), -3.951*** (0.566),
    -6.968*** (0.700), -7.285*** (0.381), -5.429*** (0.361), -4.124*** (0.580),
    -0.334*** (0.0257), 0.0608*** (0.0152), 0.352*** (0.0166), 0.644*** (0.0166),
    -0.191*** (0.0138), 0.0294** (0.0132), 0.242*** (0.0133), 0.445*** (0.0177),
    -7.425*** (0.754), -6.689*** (0.563), -4.816*** (0.669), -3.396*** (0.549).

Notes: 1. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Table 2 results; With ppp-based misalignments; With year fixed effects added; With year and country fixed effects added; With instrumental variables.

### Robustness of Undervaluation Measure (Table 5) — key coefficient patterns and statistics
- Sample sizes and fit:
  - Observations reported: 996 (many columns), 449 (columns 3–4), 289 (columns 5–6), 258 (columns 7–8).
  - R-squared: 0.567, 0.569, 0.430, 0.443, 0.495, 0.479, 0.544, 0.527.
  - Pseudo R2 (selected): 0.369, 0.366, 0.369, 0.372, 0.367, 0.366, 0.380, 0.387.
- Scale variables (selected coefficients):
  - Log(per capita income):
    - Column 1 (Full): 0.373*** (0.0888)
    - Column 2 (Full): 0.295*** (0.105)
    - Column 3 (80-97): 0.393** (0.156)
    - Column 4 (80-97): 0.308* (0.170)
    - Column 9 (qreg25): 0.322*** (0.0420)
    - Column 10 (qreg25): 0.260*** (0.0471)
    - Column 11 (qreg50): 0.374*** (0.0272)
    - Column 12 (qreg50): 0.347*** (0.0364)
    - Column 13 (qreg75): 0.324*** (0.0261), Column 14 (qreg75): 0.301*** (0.0325), Column 15 (qreg90): 0.276** (0.0433), Column 16 (qreg90): 0.269*** (0.0435).
  - Log(population):
    - Examples: 0.127** (0.0537), 0.131** (0.0501), 0.113 (0.0721), 0.111 (0.0694), 0.0746 (0.0849), 0.114 (0.0732), -0.0127 (0.0705), -0.00177 (0.0711)
    - Quantile examples: 0.157*** (0.0269), 0.195*** (0.0266), 0.135*** (0.0163), 0.152*** (0.0193), 0.0519*** (0.0147), 0.0724*** (0.0160), 0.00400 (0.0218), 0.00507 (0.0185).
- Regime (Hard and Soft peg dummy):
  - Coefficients include: 0.0297 (0.118), 0.0596 (0.115), 0.309* (0.180), 0.334* (0.166), -0.0160 (0.131), -0.00868 (0.132), -0.127 (0.0903), -0.126 (0.0910)
  - Quantile columns show: 0.0820 (0.0584), 0.0903 (0.0609), -0.00808 (0.0372), 0.00969 (0.0442), -0.0601* (0.0358), 0.0116 (0.0387), -0.0678 (0.0607), -0.0397 (0.0524).
- Volatility of neer:
  - Examples across subsamples: -0.00607 (0.00463), -0.00615 (0.00488), -8.79e-05 (0.00639), 0.00113 (0.00617)
  - Additional: -0.0170*** (0.00604), -0.0170** (0.00640), -0.00343 (0.00609), -0.00684 (0.00680)
  - Quantile columns: -0.00752** (0.00368), -0.00815** (0.00377), -0.00333 (0.00233), -0.00300 (0.00275), -0.00165 (0.00230), -0.000504 (0.00251), -0.00343 (0.00394), -0.00471 (0.00353).
- Opportunity cost (Interest rate differential w/ US):
  - Examples: -0.0359 (0.0795), -0.0317 (0.0780), 0.0136 (0.101), 0.00365 (0.101), -0.0795 (0.875), -0.238 (0.849), -0.381 (0.954), -0.901 (0.992)
  - Quantile columns: 0.0531 (0.0743), 0.00854 (0.0699), -0.142*** (0.0454), -0.112** (0.0538), -0.115*** (0.0441), -0.119** (0.0472), -0.205** (0.0862), -0.245*** (0.0733).
- Current account (Log(imports to GDP)):
  - Coefficients robustly positive:
    - 0.675*** (0.123), 0.694*** (0.119), 0.691*** (0.147), 0.693*** (0.139)
    - 0.441*** (0.138), 0.462*** (0.142), 0.428*** (0.145), 0.447*** (0.128)
    - Quantile columns: 0.839*** (0.0648), 0.838*** (0.0646), 0.626*** (0.0417), 0.675*** (0.0496), 0.506*** (0.0392), 0.561*** (0.0410), 0.352*** (0.0622), 0.354*** (0.0527).
- Volatility of exports/GDP (3-year sd):
  - Examples: 0.215*** (0.0676), 0.268*** (0.0891), 0.200** (0.0779), 0.226*** (0.0771)
  - Larger (and sometimes imprecise) estimates reported for earlier periods: 2.285* (1.286), 2.201* (1.300), 0.376 (2.057), 2.206 (2.093)
  - Quantile columns: 0.355*** (0.0623), 0.389*** (0.0769), 0.170*** (0.0448), 0.213*** (0.0495), 0.0377 (0.0450), 0.0522 (0.0493), -0.118* (0.0701), -0.110* (0.0625).
- Volatility of partner growth (3-year sd):
  - Examples: 0.316** (0.122), 0.355** (0.136), 0.354** (0.170), 0.359* (0.182)
  - Other subsample values: 0.320 (0.265), 0.404 (0.271), 0.196 (0.557), -0.193 (0.562)
  - Quantile columns: 0.324* (0.165), 0.431** (0.200), 0.254** (0.104), 0.306*** (0.111), 0.173** (0.0795), 0.297*** (0.0896), 0.105 (0.141), 0.280** (0.120).
- Capital account and financial variables:
  - Financial openness:
    - Examples: 0.108** (0.0464), 0.0849* (0.0426), 0.110* (0.0572), 0.109* (0.0560)
    - Other columns: 0.112* (0.0618), 0.118* (0.0647), -0.0257 (0.0484), -0.0551 (0.0540)
    - Quantile columns: 0.138*** (0.0200), 0.127*** (0.0206), 0.0722*** (0.0125), 0.0647*** (0.0150), 0.0419*** (0.0124), 0.0360*** (0.0134), 0.0450** (0.0215), 0.0251 (0.0185).
  - Log(broad money to GDP):
    - Examples: 0.272*** (0.0957), 0.289*** (0.0899), 0.197* (0.106), 0.216** (0.101)
    - Quantile columns: 0.311*** (0.0531), 0.332*** (0.0531), 0.301*** (0.0339), 0.316*** (0.0400), 0.356*** (0.0335), 0.343*** (0.0357), 0.359*** (0.0548), 0.362*** (0.0455).
  - Short term debt to GDP:
    - Examples: 0.276 (0.196), 0.321* (0.165), -0.397 (0.631), -0.345 (0.631)
    - Quantile columns: 0.485*** (0.151), 0.520*** (0.160), 0.363*** (0.104), 0.342*** (0.124), 0.118 (0.109), 0.124 (0.117), -0.00462 (0.229), 0.000934 (0.197).
- Mercantilist measures — exchange rate undervaluation duration:
  - Exchange rate undervaluation duration 1:
    - 0.0545*** (0.0155), 0.0285 (0.0218), 0.0540* (0.0286), 0.0496*** (0.0167), 0.0642*** (0.0115), 0.0348*** (0.00697), 0.0335*** (0.00673), 0.0369*** (0.0109).
  - Exchange rate undervaluation duration 2:
    - 0.0395*** (0.0136), 0.0402* (0.0201), 0.00581 (0.0172), 0.0387** (0.0186), 0.0484*** (0.00744), 0.0285*** (0.00592), 0.0269*** (0.00533), 0.0289*** (0.00644).
- Constants (examples):
  - -6.730*** (1.501), -6.234*** (1.522), -6.996*** (2.282), -6.384*** (2.286)
  - -3.634* (2.065), -4.192** (1.931), -1.582 (1.769), -1.705 (2.001)
  - Quantile constants: -7.031*** (0.685), -7.250*** (0.711), -6.801*** (0.415), -6.916*** (0.515), -4.676*** (0.387), -4.940*** (0.443), -3.277*** (0.627), -3.351*** (0.564).

Notes: 1. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).

### Appendix A — Construction of Mercantilist Variables (methodologies)
- Overview:
  - Misalignment estimates constructed using three equilibrium exchange rate methodologies: macroeconomic balance (MB), reduced-form equilibrium real exchange rate (ERER), and external sustainability (ES), plus a PPP methodology.
  - Assessments based on IMF Occasional Paper 261, “Exchange Rate Assessments: CGER Methodologies”.
  - Procedure: construct misalignment estimates from each method, average the estimates, then map the average misalignment into a dummy for greater than 10 percent overvalued.
- Macroeconomic Balance (MB):
  - MB calculates difference between “underlying” medium-term current account balance (based on WEO forecasts) and estimated equilibrium current account balance (“current account norm”).
  - The exchange rate adjustment is obtained using country-specific elasticity of the current account with respect to the real exchange rate.
  - Estimated MB equation (as printed):
    - ca_it = β0 + β1 oldage_it + β2 pop_it + β3 income_it + β4 output_it + β5 oil_it + β6 fiscal_it + β7 NFA_it + ε_it
  - Variable definitions:
    - ca: Ratio of the current account balance to output for economy
    - oldage: Old age dependency ratio relative to trading partners
    - pop: Population growth relative to trading partners
    - income: GDP per capita relative to trading partners
    - output: Real output growth relative to trading partners
    - oil: Ratio of oil trade balance to nominal GDP
    - fiscal: Fiscal balance to nominal GDP ratio, relative to trading partners
    - NFA: Ratio of initial net foreign assets to nominal GDP
  - Note 25: semi-elasticity constructed based on medium run elasticity of volume of exports or imports w.r.t. real effective exchange rate, and ratios of exports or imports to output.
- Reduced-form Equilibrium Real Exchange Rate (ERER):
  - Two-step estimation per country: (i) panel regression of RER on fundamentals; (ii) compute equilibrium RER from medium-term projections of fundamentals.
  - ERER estimated equation (as printed):
    - ln Q_it = β0_i + β1 to_t + β2 prod_it + β3 govconsum_it + β4 NFA_it + ε_it
  - Variables:
    - Q: Real effective exchange rate
    - β0: Economy specific fixed effect
    - to_t: Terms of trade
    - prod: Relative productivity
    - NFA: Initial net foreign assets to nominal GDP ratio
- External Sustainability (ES):
  - ES computes difference between actual current account and CA that would stabilize NFA at benchmark level s_nfa.
  - Uses elasticities from MB to translate CA gap into RER adjustment.
  - Stabilizing condition uses growth rate of real GDP (g) and inflation (π); s_nfa set as NFA/GDP at end of prior year.
- Purchasing Power Parity (PPP):
  - RER = ExchangeRate / PPP; RER > 1 suggests currency value lower than PPP.
  - Adjusted PPP accounts for Balassa-Samuelson effect via regression:
    - ln(RER_it) = α + β ln(PerCapitaGDP_it) + country and time fixed effects + ε_it.
  - PPP estimates used for comparison; absolute PPP does not hold short-run, relative PPP holds in long-run after adjustment.

### Appendix A — Construction of Excess Exports (gravity-based measure)
- Approach:
  - Excess exports derived from deviations of actual bilateral exports from predicted bilateral exports from a gravity model.
  - Gravity regression: log X_ijtk = a_0 + W'_ijt b + u_ijt, where W_ijt includes distance, land border, regional trade agreement membership, bilateral free trade agreements, historical ties, common currency, etc.
  - Estimation methods considered: pooled OLS, pooled OLS with country effects (Anderson and van Wincoop), panel country-pair fixed effects, and Hausman-Taylor.
  - Focus on country-pair fixed effects and Hausman-Taylor as more robust; Hausman-Taylor can identify time-invariant determinants and control for endogeneity but is instrument-sensitive.
- Export potential index construction:
  - PotX_it = sum of fitted bilateral export flows from exporter to all partners.
  - Export potential index = ActualX_it / PotX_it.
  - Standardized index (range [-1,1]): (ActualX_it - PotX_it) / (ActualX_it + PotX_it). Positive index in (0,1] indicates higher effective exports than predicted. Negative in [-1,0) indicates lower.
  - Dummy variables for “excess exports”: 1 if standardized index positive, 0 otherwise.

### Appendix B — Data, sample, and summary statistics
- Countries in the sample (list as provided):
  - Argentina, Armenia, Bosnia and Herzegovina, Brazil, Bulgaria, Chile, China, Colombia, Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, El Salvador, Georgia, Guatemala, Hungary, India, Indonesia, Jamaica, Jordan, Kazakhstan, Latvia, Lebanon, Lithuania, Malaysia, Mexico, Morocco, Pakistan, Panama, Peru, Philippines, Poland, Romania, Russia, South Africa, Thailand, Tunisia, Turkey, Ukraine, Uruguay, Venezuela, Vietnam.
- Variable definitions and sources (selection):
  - Log(reserves to GDP): Natural log of ratio of foreign exchange reserves to USD GDP — IMF, World Economic Outlook.
  - Log(per capita income): Natural log of per capita income (at PPP) — IMF, World Economic Outlook.
  - Log(population): Natural log of population — IMF, World Economic Outlook.
  - Log(imports to GDP): Natural log of ratio of imports to GDP — IMF, World Economic Outlook.
  - Volatility of neer: Twelve month standard deviation of end of period nominal effective exchange rate — IMF, International Financial Statistics and authors' calculations.
  - Interest rate differential w/ US: ln[(1 + i)/(1 + i_US)], with i_US the US interest rate corresponding to the national rate definition — IMF, IFS and authors' calculations.
  - Volatility of exports/GDP (3-year sd): 3 year standard deviation of export to GDP (goods) — IMF, WEO and authors' calculations.
  - Financial openness: Chinn-Ito index measuring capital account openness — Chinn and Ito (2009).
  - Log(broad money to GDP): Natural log of ratio of M2 to GDP — IMF, WEO.
  - Peg dummy: Dummy equal to 1 if currency pegged, 0 otherwise — Ghosh, Qureshi, and Tsangarides (2011).
  - Short term debt to GDP: Ratio of total short-term debt outstanding to GDP — IMF, WEO.
  - Volatility of partner growth (3-year sd): 3 year standard deviation of growth of trading partners' real GDP — IMF, WEO and authors' calculations.
  - Undervaluation dummy: Dummy equal to 1 if currency is undervalued and 0 otherwise — Authors' calculations.
- Summary statistics (selected):
  - Log(reserves to GDP): Mean -2.33, Std. Dev. 0.87, Min -5.42, Max 0.07.
  - Log(per capita income): Mean 8.47, Std. Dev. 0.74, Min 6.03, Max 9.88.
  - Log(population): Mean 17.01, Std. Dev. 1.60, Min 14.56, Max 21.01.
  - Log(imports to GDP): Mean -1.20, Std. Dev. 0.60, Min -3.07, Max 0.01.
  - Volatility of neer: Mean 13.04, Std. Dev. 8.15, Min 0.81, Max 45.45.
  - Interest rate differential w/ US: Mean 0.13, Std. Dev. 0.40, Min -0.12, Max 5.08.
  - Volatility of exports/GDP (3-year sd): Mean 0.02, Std. Dev. 0.08, Min 0.00, Max 2.54.
  - Financial openness: Mean 0.16, Std. Dev. 1.53, Min -1.84, Max 2.48.
  - Log(broad money to GDP): Mean -0.84, Std. Dev. 0.62, Min -2.71, Max 1.05.
  - Peg dummy: Mean 0.61, Std. Dev. 0.49, Min 0, Max 1.
  - Short term debt to GDP: Mean 0.14, Std. Dev. 0.17, Min 0.00, Max 1.73.
  - Volatility of partner growth (3-year sd): Mean 1.19, Std. Dev. 0.93, Min 0.02, Max 6.91.
  - Undervaluation dummy: Mean 0.32, Std. Dev. 0.47, Min 0, Max 1.
  - Excess exports: Mean 0.27, Std. Dev. 0.24, Min -0.58, Max 0.94.

*Source: _wp1234 - 1. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2012/_wp1234.pdf_