## _wp0813

## Source details

**Canonical URL:** [_wp0813](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0813.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0813.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0813.pdf.json)

---

### I. Introduction and objective
- World trade to world GDP increased more than 40 percent since the early 1990s; international financial cross-holdings to world GDP more than doubled.
- Emerging markets’ share of world trade rose from 27 percent in 1990 to 37 percent in 2004.
- Paper objective: estimate a reduced-form panel cointegrating relationship for the CPI-based real effective exchange rate using a novel dataset of 48 countries (covering 1980–2004) and a comprehensive set of fundamentals including productivity (tradables vs. nontradables), net foreign assets, and commodity-based terms of trade.

### II. Data, sample, and econometric methodology
- Sample and period:
  - 48 countries; period 1980–2004. Alternative sample treats the Euro area as one country.
- Key variables (constructed as described in Appendix):
  - Net foreign assets (scaled by trade; also normalized to exports, imports, GDP in robustness checks).
  - Productivity differential: output per worker in tradables minus nontradables (relative to trading partners).
  - Commodity terms of trade: weighted main commodity export prices divided by weighted main commodity import prices, relative to manufacturing exports price.
  - Government consumption: ratio to GDP.
  - Trade restriction index: dummy = 1 before liberalization, 0 after.
  - Price controls: share of administered prices in CPI basket (transition economies only).
- Unit root and cointegration tests:
  - Levin, Lin, and Chu (2002) and Im, Pesaran, and Shin (2003) panel unit root tests indicate unit roots.
  - Kao (1999) panel cointegration test widely rejects null of no cointegration at the 1 percent level.
- Estimation:
  - Dynamic OLS (DOLS) with fixed country effects:
    - log(RER_it) = α_i + β' X_it + Σ_{s=-p}^{p} γ_s ΔX_{i,t+s} + u_it
  - Baseline includes one lead and one lag of first differences; Newey-West correction applied.
  - Robustness checks include FMOLS, alternative productivity measures (excluding agriculture), different net foreign asset normalizations, different leads/lags, alternative government consumption sources, crisis-slope checks, OLS on 3-year averages.

### III. Theoretical rationale for key variables
- Net foreign assets:
  - Debtor countries require more depreciated RER to service liabilities; creditor countries can sustain more appreciated RER.
  - Net foreign assets scaled by trade captures servicing capacity relative to trade flows.
- Productivity differential (Balassa-Samuelson):
  - Higher productivity growth in tradables vs. nontradables raises relative price of nontradables → real appreciation.
  - Measured as difference in output per worker in tradables and nontradables relative to trading partners.
- Commodity terms of trade:
  - Higher commodity terms of trade appreciate RER via income/wealth effects.
- Government consumption:
  - Higher government consumption to GDP likely appreciates RER if consumption falls on nontradables.
- Trade liberalization and price controls:
  - Liberalization associated with real depreciation episodes (trade restriction index: pre-liberalization = 1, post = 0).
  - Reduction in share of administered prices associated with real appreciation.

### IV. Long-run estimation results and magnitudes (large sample, DOLS with fixed effects)
- Net foreign assets (scaled by trade):
  - Estimated coefficient ≈ 0.04.
  - Example interpretation: a deterioration of the ratio of net foreign assets to trade of about 50 percentage points (as experienced by the United States between 2001 and 2005) would imply a depreciation of the equilibrium exchange rate by about 2 percent.
- Productivity differential (tradables vs. nontradables):
  - Estimated coefficient ≈ 0.2.
  - A ten percent increase in domestic productivity of tradables relative to nontradables (relative to trading partners) tends to appreciate equilibrium exchange rate by about 2 percent.
- Commodity terms of trade:
  - A ten percent increase implies an equilibrium appreciation of 5½ percent.
  - Long-run coefficient (log commodity terms of trade) ≈ 0.56 (p-value (0.00)*** in main regression).
- Government consumption:
  - An increase in government consumption to GDP ratio of 1 percentage point associated with an appreciation of the equilibrium RER of close to 3 percent.
  - Long-run coefficient in main regression: 2.84 (p-value (0.00)***).
- Trade liberalization dummy:
  - Move to a liberalized trade regime depreciates equilibrium RER by about 12 percent.
  - Long-run coefficient (trade restriction index): 0.13 (p-value (0.01)***).
- Price controls (administered prices share):
  - Elimination of administered prices accounting for 7 percent of the price basket associated with an appreciation of about 4 percent.
  - Long-run coefficient (price controls): -0.04 (p-value (0.02)**).
  - Example: Slovak Republic liberalization of prices accounting for 20 percent of price basket between 1997 and 2004 would imply equilibrium appreciation ≈ 12 percent.
- Cointegration and speed of adjustment:
  - Kao (1999) cointegration test rejects no cointegration at the 1 percent level.
  - Error-correction: about a quarter of the gap between actual and equilibrium RER is corrected within one year (δ ≈ 0.25), implying a half-life of around 2½ years.
- Forecast performance:
  - Forecast standard error of RER ≈ 12 percent (standard error of regression ≈ 11 percent).
  - Imposing cointegrating vector in an error-correction framework reduces forecast standard error to about 7–8 percent.

### V. Additional empirical observations on productivity measures and regional patterns
- Average cross-section growth (country annual averages, in percent):
  - All sample: tradables 0.03; Non-tradables -0.11; differential 0.14.
  - Industrial: tradables -0.04; Non-tradables -0.16; differential -0.12.
  - Developing: tradables 0.10; Non-tradables -0.06; differential 0.16.
  - Asian: tradables 1.20; Non-tradables 0.78; differential 0.43.
  - Latin American: tradables -2.19; Non-tradables -1.58; differential -0.61.
  - Transition: tradables 2.10; Non-tradables 0.70; differential 1.40.
  - Other: tradables -1.50; Non-tradables -0.66; differential -0.84.
- Correlations with log real Effective Exchange Rate (significant at 5 percent level where starred):
  - Log relative prod. tradables: 0.12*.
  - Log productivity differential: 0.11*.
  - Log commodity terms of trade: 0.21*.
  - Net foreign assets to trade: 0.10*.
  - Government consumption to GDP: 0.32*.
  - Trade restriction index: 0.10*.
  - Price controls: -0.08*.
- Relation to GDP per worker (Table 4):
  - Log relative GDP per worker coefficient on tradables: 0.80 (p-value (0.00)***).
  - Log relative GDP per worker coefficient on non-tradables: 0.80 (p-value (0.00)***).
  - Log relative GDP per worker coefficient on differential: -0.00 (p-value (0.97)).
  - Implication: relative GDP per worker is strongly correlated with sectoral levels but the difference (tradables minus nontradables) is not correlated with relative GDP per worker, indicating the need for sectoral productivity measures to proxy Balassa-Samuelson effects.

### VI. Robustness checks and alternative normalizations
- FMOLS produces similar long-run results.
- Excluding agricultural sector from productivity measures (Table 6):
  - Main coefficients remain similar; adjusted R2 ≈ 0.60 (large sample).
  - Long-run signs and significance preserved for trade restriction index, commodity terms of trade, net foreign assets to trade, government consumption, and sectoral productivity terms.
- Different net foreign asset ratios (Table 7):
  - Net foreign assets to exports: coefficient 0.03 (p-value (0.03)**) in one specification.
  - Net foreign assets to GDP: coefficient 0.11 (p-value (0.01)*** ) in one specification.
  - Net foreign assets to imports: coefficients 0.05 (p-value (0.00)*** ) and 0.04 (p-value (0.01)**) in reported specifications.
  - Other key coefficients (commodity terms of trade, government consumption, trade restriction index, price controls, productivity terms) remain positive/negative and significant across specifications.
  - Adjusted R2 ranges from 0.56 to 0.61 across specifications; observations typically 861 (large sample) and 645 (small sample).

### VII. Empirical implications and key takeaways
- A parsimonious set of fundamentals (net external position, productivity differential tradables vs. nontradables relative to trading partners, commodity terms of trade, government consumption, trade restrictions, share of administered prices) explains long-run behavior of CPI-based real effective exchange rates in a 48-country panel (1980–2004).
- Estimated long-run relationships are statistically and economically significant:
  - Commodity terms of trade: coefficients around 0.52–0.64 (p-value (0.00)***).
  - Government consumption to GDP: coefficients around 2.79–3.27 (p-value (0.00)***).
  - Trade restriction index: coefficients around 0.10–0.17 (significant).
  - Net foreign assets (various normalizations): generally positive association with RER, magnitudes vary by normalization.
  - Sectoral productivity: higher relative tradable productivity → appreciation (coefficients ~0.17–0.25); higher relative nontradable productivity → depreciation (coefficients ~-0.19 to -0.21).
- Adjustment dynamics: error-correction Gap(-1) coefficients around -0.23 to -0.21 (highly significant), implying partial mean reversion and a misalignment half-life ≈ 2½ years.
- Methodological implication: importance of using sectoral productivity measures for both tradables and non-tradables relative to trading partners to proxy Balassa-Samuelson; the difference matters and is not well proxied by relative GDP per worker.

*Source: IMF Working Paper chapter “7. Robustness: Different Net Foreign Asset Ratios” (excerpt provided) — _wp0813.*

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

### _wp0813 - References................................................................................................................................13

### References
- References................................................................................................................................13

### Appendix: Data Sources and Construction
- Appendix: Data Sources and Construction ..............................................................................17

### Tables
- 1. Summary Statistics...............................................................................................................20
- 2. Correlations..........................................................................................................................21
- 3. Average Growth of Productivity Measures (Cross-Section Averages of Country 
 Annual Averages, in percent........................................................................................22
- 4. The Relation Between Productivity Measures and GDP per worker ..................................22
- 5. Main Regressions.................................................................................................................23
- 6. Robustness: Excluding Agricultural Sector from Productivity Measures ...........................24

*Source: _wp0813 - References..............................................................................................................*

### 7. Robustness: Different Net Foreign Asset Ratios .................................................................25

### 7. Robustness: Different Net Foreign Asset Ratios

### I. Introduction
- World trade to world GDP increased more than 40 percent since the early 1990s; international financial cross-holdings to world GDP more than doubled.
- Emerging markets’ share of world trade rose from 27 percent in 1990 to 37 percent in 2004.
- Paper objective: estimate a reduced-form panel cointegrating relationship for the CPI-based real effective exchange rate using a novel dataset of 48 countries (covering 1980–2004) and a comprehensive set of fundamentals including productivity (tradables vs. nontradables), net foreign assets, and commodity-based terms of trade.

### III. Data and econometric methodology
- Sample: 48 industrial countries and emerging markets; alternative sample treats the Euro area as one country. Period: 1980–2004.
- Variables (constructed as described in Appendix): net foreign assets (scaled by trade), productivity differential (output per worker in tradables minus nontradables, relative to trading partners), commodity terms of trade (weighted main commodity export prices divided by weighted main commodity import prices, relative to price of manufacturing exports of advanced countries), government consumption (ratio to GDP), trade restriction index (dummy: 1 before liberalization, 0 after), price controls (share of administered prices in CPI basket).
- Unit root and cointegration: Levin, Lin, and Chu (2002) and Im, Pesaran, and Shin (2003) panel unit root tests indicate unit roots; Kao (1999) panel cointegration test finds long-run relation between RER and fundamentals.
- Estimation method: Dynamic OLS (DOLS) with fixed country effects:
  - log(RER_it) = α_i + β' X_it + Σ_{s=-p}^{p} γ_s ΔX_{i,t+s} + u_it
  - Leads and lags of first differences (one lead and lag in baseline) included; Newey-West correction applied.
- Robustness checks: FMOLS panel cointegration; alternative productivity measures excluding agriculture; different normalizations of net foreign assets; different leads/lags; alternative government consumption sources; crisis-slope checks; OLS on 3-year averages. Results broadly robust.

### II. Theoretical background (rationale for variables)
- Net foreign assets: debtor countries require more depreciated RER to service liabilities; creditor countries can sustain more appreciated RER. Net foreign assets scaled by trade.
- Productivity differential (Balassa-Samuelson): higher productivity growth in tradables vs. nontradables leads to higher relative price of nontradables → real appreciation. Measured as difference in output per worker in tradables and nontradables (relative to trading partners).
- Commodity terms of trade: higher commodity terms of trade appreciate RER via income/wealth effects. Constructed from main commodity export/import prices relative to manufactured goods price.
- Government consumption: higher government consumption to GDP likely appreciates RER if consumption falls on nontradables.
- Trade restriction index: pre-liberalization = 1, post-liberalization = 0; liberalization associated with real depreciation episodes.
- Price controls: share of administered prices in CPI; reduction associated with real appreciation.

### IV. Estimation results — long-run coefficients and magnitudes (large sample, Column 2)
- Net foreign asset variable (scaled by trade):
  - Estimated coefficient ≈ 0.04.
  - Interpretation example: a deterioration of the ratio of net foreign assets to trade of about 50 percentage points (as experienced by the United States between 2001 and 2005) would imply a depreciation of the equilibrium exchange rate by about 2 percent.
  - Comment: estimate somewhat lower than Lane and Milesi-Ferretti (2004); weakening over 199-2004 may reflect rise in domestic-currency liabilities and currency composition effects.
- Productivity differential (tradables vs. nontradables):
  - A ten percent increase in domestic productivity of tradables relative to nontradables (relative to trading partners) tends to appreciate equilibrium exchange rate by about 2 percent.
  - Estimated coefficient ≈ 0.2 (noted as low relative to theoretical priors; measurement noise likely).
- Commodity terms of trade:
  - A ten percent increase implies an equilibrium appreciation of 5½ percent.
  - Coefficient in line with Chen and Rogoff (2004), Cashin, Céspedes, and Sahay (2004).
- Government consumption:
  - An increase in government consumption to GDP ratio of 1 percentage point is associated with an appreciation of the equilibrium RER of close to 3 percent.
  - Noted as somewhat higher than the 1.5–2 percent range found in De Gregorio, Giovannini, and Wolf (1994).
- Trade liberalization dummy:
  - Move to a liberalized trade regime depreciates equilibrium RER by about 12 percent.
- Price controls (administered prices share):
  - Elimination of administered prices accounting for 7 percent of the price basket is associated with an appreciation of about 4 percent.
  - Example: Slovak Republic liberalization of prices accounting for 20 percent of price basket between 1997 and 2004 would imply equilibrium appreciation ≈ 12 percent.
- Cointegration tests: Kao (1999) widely rejects null of no cointegration at the 1 percent level.
- Speed of adjustment (error-correction specification):
  - About a quarter of the gap between actual and equilibrium RER is corrected within one year (δ ≈ 0.25), implying a half-life of around 2½ years.
- Forecast performance:
  - Forecast standard error of RER ≈ 12 percent (standard error of regression ≈ 11 percent).
  - Imposing cointegrating vector in an error-correction framework reduces forecast standard error to about 7–8 percent.
- Robustness:
  - FMOLS estimations produce similar results.
  - Results robust to alternative productivity measures (excluding agriculture) and alternative normalizations of net foreign assets.

### IV. Additional empirical observations on productivity measures
- Average country over sample experienced:
  - Modest negative growth of productivity of non-tradables relative to trading partners.
  - Slight positive growth of relative productivity of tradables → modest positive Balassa-Samuelson effect on average.
- Regional heterogeneity:
  - Asia: annual average growth of relative productivity of tradables ≈ 1 percent; Balassa-Samuelson effect positive.
  - Latin America: opposite pattern.
- Correlations:
  - Tradables and non-tradables productivities relative to trading partners are very strongly correlated with GDP per worker.
  - The difference between the two productivities is not strongly correlated with GDP per worker.
  - Implication: relative GDP per worker may be a poor proxy for the Balassa-Samuelson effect; the difference in sectoral productivities matters.

### V. Concluding remarks (key takeaways)
- A parsimonious set of fundamentals (net external position, productivity differential tradables vs. nontradables relative to trading partners, commodity terms of trade, government consumption, trade restrictions, share of administered prices) explains long-run behavior of CPI-based real effective exchange rates in a 48-country panel (1980–2004).
- Estimated long-run relationships are statistically and economically significant.
- Estimated speed of adjustment: about 0.25 per year → half-life of misalignment ≈ 2½ years.
- Importance of using sectoral productivity measures for both tradables and non-tradables relative to trading partners to proxy Balassa-Samuelson: the difference (not level) matters and is not captured well by relative GDP per worker.

*Source: IMF Working Paper chapter “7. Robustness: Different Net Foreign Asset Ratios” (excerpt provided).*

### References

### _wp0813 - References

### Sample and data coverage
- Sample includes 48 countries for the period 1980 to 2004.
- Country groups:
  - Advanced economies: Australia, Canada, Denmark, Japan, New Zealand, Norway, Sweden, Switzerland, United Kingdom, United States; and 11 Euro-area countries comprising Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Portugal, and Spain.
  - Newly industrialized or emerging markets: Argentina, Brazil, Chile, China, Colombia, Czech Republic, Hong Kong SAR, Hungary, India, Indonesia, Korea, Malaysia, Mexico, Morocco, Pakistan, Peru, Philippines, Poland, Russia, Singapore, Slovak Republic, Slovenia, South Africa, Taiwan Province of China, Thailand, Turkey, Venezuela.
- Notes on time series: Euro-area series are employed in some regressions instead of individual countries; Central and Eastern European countries generally have data from 1993 onwards; data for China, Hungary, and Poland for the 1980s was dropped; Peru’s 1980s data dropped due to hyperinflation.

### Data construction and variable definitions
- Real effective exchange rate: based on CPI and new competitiveness weights constructed from 1999–2001 data (Bayoumi and others (2006)); nominal exchange rate and CPI from IFS; Euro-area data (prior to 1999) from Global Data Source (GDS).
- Productivity of tradables and nontradables relative to trading partners:
  - Productivity measured as output per worker.
  - New dataset for output and employment for a 6-sector classification (or 3-sector when 6-sector not available).
  - 6-sector tradables: agriculture, hunting, forestry, and fishing; mining, manufacturing, and utilities; transport, storage, and communication.
  - 6-sector nontradables: construction; wholesale and retail trade; other services.
  - 3-sector tradables: agriculture and industry.
  - Sources: United Nations Statistics Division, International Labor Office Bureau of Statistics, Eurostat, World Bank, Groningen Growth and Development Centre, CEIC Database, and national authorities.
  - Missing observations filled using sectoral shares for adjacent years and aggregate data; trading partner series constructed by applying competitiveness weights and extended when missing using three-year adjacent trends.
  - Robustness checks: alternative measures excluding agricultural sector for specified countries and employment series smoothed with Hodrick-Prescott filter.
- Commodity-based terms of trade: ratio of weighted average price of main commodity exports to weighted average price of main commodity imports; index from six commodity categories (food, fuels, agricultural raw materials, metals, gold, beverages) measured against the manufacturing unit value index (MUV) of WEO; weights are time averages over 1980-2001 of export/import shares.
- Net foreign assets to trade: ratio of net foreign assets (end of previous period) to average exports and imports (goods and non-factor services) of previous period; net foreign assets from Lane and Milesi-Ferretti (2007); trade data from IFS and extended using WEO; Singapore exports adjusted for re-exports.
- Government consumption to GDP: ratio of government consumption (purchases of goods and services plus government wages) to GDP; main source OECD, Annual National Income Accounts; missing observations spliced using IFS or WEO.
- Trade restriction index: takes value 0 during years of liberalization and 1 during years of restriction; constructed on the basis of liberalization years suggested by Sachs and Warner, extended by Wacziarg and Welch (2003).
- Share of administered prices (transition economies only): EBRD construct as number of categories with administered prices out of a basket of 15 categories; available for Czech Republic, Hungary, Poland, Russia, Slovak Republic, Slovenia; takes value 0 for other countries.

### Key summary statistics (Table 1)
- Observations: 861 for listed variables.
- Log real Effective Exchange Rate: Mean 4.60; Std. Dev. 0.18; Min 3.82; Max 5.41.
- Log relative prod. tradables: Mean 0.14; Std. Dev. 0.14; Min -0.51; Max 0.75.
- Log relative prod. nontradables: Mean 0.03; Std. Dev. 0.12; Min -0.43; Max 0.64.
- Log productivity differential: Mean 0.10; Std. Dev. 0.14; Min -0.65; Max 0.50.
- Log commodity terms of trade: Mean 4.61; Std. Dev. 0.11; Min 4.34; Max 5.49.
- Net foreign assets to trade: Mean -0.81; Std. Dev. 1.28; Min -5.11; Max 3.12.
- Net foreign assets to GDP: Mean -0.15; Std. Dev. 0.42; Min -1.65; Max 2.10.
- Net foreign assets to exports: Mean -0.86; Std. Dev. 1.29; Min -5.49; Max 3.02.
- Net foreign assets to imports: Mean -0.79; Std. Dev. 1.29; Min -5.77; Max 3.23.
- Government consumption to GDP: Mean 0.17; Std. Dev. 0.05; Min 0.05; Max 0.30.
- Trade restriction index: Mean 0.10; Std. Dev. 0.30; Min 0; Max 1.
- Price controls: Mean 0.13; Std. Dev. 0.60; Min 0; Max 5.

### Correlations (Table 2) — selected significant correlations (5 percent level)
- Log real Effective Exchange Rate with:
  - Log relative prod. tradables: 0.12*.
  - Log relative prod. nontradables: 0.10 (not starred).
  - Log productivity differential: 0.11*.
  - Log commodity terms of trade: 0.21*.
  - Net foreign assets to trade: 0.10*.
  - Government consumption to GDP: 0.32*.
  - Trade restriction index: 0.10*.
  - Price controls: -0.08*.
- Notes: Residuals of each variable on fixed effect; sample based on main regression sample for the large sample (Table 5 columns 1 and 2). A star (*) denotes significance at the 5 percent level.

### Productivity growth (Table 3)
- Cross-section averages of country annual averages (in percent):
  - All sample: tradables .03; Non-tradables -.11; differential .14.
  - Industrial: tradables -.04; Non-tradables -.16; differential -.12.
  - Developing: tradables .10; Non-tradables -.06; differential .16.
  - Asian: tradables 1.20; Non-tradables .78; differential .43.
  - Latin American: tradables -2.19; Non-tradables -1.58; differential -.61.
  - Transition: tradables 2.10; Non-tradables .70; differential 1.40.
  - Other: tradables -1.50; Non-tradables - .66; differential -.84.

### Relation between productivity measures and GDP per worker (Table 4)
- Dependent variable: Log Relative Productivity
  - Log relative GDP per worker coefficient on tradables: 0.80 (p-value (0.00)***).
  - Log relative GDP per worker coefficient on non-tradables: 0.80 (p-value (0.00)***).
  - Log relative GDP per worker coefficient on differential: -0.00 (p-value (0.97)).
- Observations: 861 for each regression.
- Estimation: Dynamic OLS with fixed effects; only long-term relationship reported. Significance indicators: *, **, *** for 10, 5, 1 percent.

### Main regression results (Table 5) — long-run relations (Dynamic OLS with fixed effects; p-values in parentheses)
Dependent variable: Log real Effective Exchange Rate
- Large sample (columns reported):
  - Trade restriction index: 0.13 (0.01)***.
  - Price controls: -0.04 (0.02)**.
  - Log commodity terms of trade: 0.56 (0.00)***.
  - Net foreign assets to trade: 0.04 (0.01)***.
  - Government consumption to GDP: 2.84 (0.00)***.
  - Log relative prod. Tradable: 0.17 (0.01)***.
  - Log relative prod. Nontradables: -0.21 (0.03)**.
  - Adjusted R2: 0.60.
  - Observations: 861.
  - HO: prod. t = - prod. non-t (p-val): 0.65.
- Small sample and balanced sample show broadly similar signs and significance; examples:
  - Small sample Trade restriction index: 0.12 (0.01)***; Log commodity terms of trade: 0.55 (0.00)***; Government consumption to GDP: 2.91 (0.00)***; Net foreign assets to trade: 0.04 (0.00)***.
  - Balanced sample Log productivity differential: 0.21 (0.04)**; Government consumption to GDP: 3.11 (0.00)***; Observations: 440; Adjusted R2: 0.26/0.29 in reported columns.
- Error-correction / mechanism (Dependent variable: Change in log real Effective Exchange Rate):
  - Gap(-1): coefficients around -0.23 to -0.21 with p-values (0.00)*** to (0.00)**.

### Robustness checks
- Excluding agricultural sector from productivity measures (Table 6):
  - Main coefficients remain similar:
    - Trade restriction index: 0.13 (0.01)*** (large sample).
    - Log commodity terms of trade: 0.56 (0.00)***.
    - Net foreign assets to trade: 0.04 (0.01)***.
    - Government consumption to GDP: 2.79 to 2.94 (0.00)*** range.
    - Log relative prod. tradables: 0.18 (0.00)***; Log relative prod. nontradables: -0.21 (0.03)**.
  - Observations: 861 (large sample) and 645 (small sample).
  - Adjusted R2: 0.60 (large sample), 0.59–0.60 (small sample).
  - HO: prod. t = - prod. non-t (p-val): 0.75 and 0.55 in reported columns.

- Different net foreign asset ratios (Table 7):
  - Trade restriction index remains positive and significant (e.g., 0.12 (0.01)***, 0.14 (0.00)***).
  - Price controls remain negative and significant (e.g., -0.04 (0.02)**).
  - Log commodity terms of trade stays positive and significant (e.g., 0.56 (0.00)***, 0.52 (0.00)***).
  - Government consumption to GDP remains strongly positive and significant (e.g., 2.86 (0.00)***, 3.09 (0.00)***).
  - Net foreign assets expressed to exports/GDP/imports:
    - Net foreign assets to exports: 0.03 (0.03)** in one specification.
    - Net foreign assets to GDP: 0.11 (0.01)*** in one specification.
    - Net foreign assets to imports: 0.05 (0.00)*** and 0.04 (0.01)** in reported specifications.
  - Observations: 861 (large sample) and 645 (small sample).
  - Adjusted R2 ranges from 0.56 to 0.61 across specifications.

### Empirical implications (drawn from reported estimates)
- Commodity terms of trade have a consistently positive and highly significant association with the real effective exchange rate (coefficients around 0.52–0.64 with (0.00)*** p-values).
- Government consumption to GDP is positively associated with real effective exchange rate (coefficients around 2.79–3.27 with (0.00)*** p-values).
- Trade restriction index is positively associated with the real effective exchange rate (coefficients around 0.10–0.17 with (0.01)*** to (0.03)** p-values).
- Net foreign asset positions (variously scaled) are generally positively associated with the real effective exchange rate, though magnitudes and significance vary by specification.
- Relative productivity patterns matter:
  - Higher relative tradable productivity is associated with appreciation (positive coefficient ~0.17–0.25, significant).
  - Higher relative nontradable productivity is associated with depreciation (negative coefficient ~-0.19 to -0.21, significant).
  - Productivity differential specifications show positive coefficients (~0.18–0.24, significant).
- Error-correction estimates indicate partial adjustment: Gap(-1) coefficients around -0.23 to -0.21 (highly significant), implying mean reversion in changes of the log real effective exchange rate.

*Source: _wp0813 - References (PDF)._

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0813.pdf_