## _wp1120 — 8. Exchange Rate Misalignment

## Source details

**Canonical URL:** [_wp1120 — 8. Exchange Rate Misalignment](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1120.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1120.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1120.pdf.json)

---

### I. Introduction: purpose and methods
- Purpose: assess whether Armenia’s real exchange rate is close to its equilibrium value to inform adjustment needs and trajectories of economic fundamentals.
- Methodologies used:
  - Macroeconomic Balance approach (MB) — current account misalignment-based.
  - External Sustainability approach (ES) — accounting-based current account norm to stabilize the net external position.
  - Equilibrium Real Exchange Rate approach (ERER) — reduced-form long-run relationship between the real exchange rate and fundamentals (single-country ARDL).
  - Panel estimation techniques addressing cross-section dependence using common correlated effects estimators (Pesaran, 2006) and a panel cointegration test accommodating common factors (Pesaran and Tosetti, 2010).
- Headline result: ERER and panel approaches indicate the dram was overvalued as of end 2009 by 7–11 percent.

### II. Recent developments and competitiveness evidence
- Export performance and structure:
  - Export volume: increased between 2001 and 2008, then fell sharply at end-2008; exports started a modest recovery but remained at 2003 levels.
  - Export composition: precious and non-precious metals, precious and semi-precious stones and minerals constitute more than 65 percent of exports between 2000 and 2009.
  - Sectoral shifts: away from precious stones and metals to non-precious metals/minerals; declines in machines/equipment, textile products, plastics and rubber — interpreted as signs of loss of competitiveness.
- Current account and tradables:
  - Current account: deteriorated markedly since 2004, reaching more than -16 percent in 2009.
  - Non-commodity exports: declined as a percent of GDP until 2008.
  - World export share and exports per capita: Armenia’s share of world exports decreased significantly since 2007; exports per capita declined while output per capita grew until 2008.
- Price competitiveness (REER developments):
  - REER: appreciated steadily from 2006 until March 2009.
  - In March 2009 the dram depreciated by 22 percent vis-à-vis the US dollar after returning to a floating exchange rate regime.
  - Between December 2009 and July 2010 the dram appreciated the most (both real and nominal) compared to major competitors.
- Non-price competitiveness (surveys/indices):
  - World Economic Forum 2010: Armenia ranked 98 out of 139 countries (GCI score 3.76); outperformed only Tajikistan and the Kyrgyz Republic in the sample.
  - Specific constraints (2010): lack of local competition (rank 136/139), extent of market dominance (rank 133/139), ineffectiveness of anti-monopoly policy (rank 138/139).
  - World Bank Ease of Doing Business 2010: “Starting a business” rank = 21 out of 183; “Protecting investors” rank = 93; “Paying taxes” rank = 153.
  - Enterprise surveys: high shares of firms identify corruption, crime/theft/disorder and access to finance as major constraints; Armenia’s percentages exceed transition-economy averages.
- Conclusion from indicators: multiple price and non-price indicators point to a loss of external competitiveness in recent years.

### III. Exchange rate assessment — methods, key parameters, and results

- Common data/method notes and elasticities (preserved exactly):
  - Imports elasticity to RER = 1.22.
  - Exports elasticity to RER = -1.08 (Tokarick, 2010).
  - Implied elasticity of the current account to the real exchange rate = 0.45 (formula and average import/export ratios 1999–2009 used).
  - ERER single-country estimation uses quarterly data with 45 observations from 1999Q1 to 2010Q1; ARDL(3,0,4,2,3,0,0,4) selected by Schwarz Bayesian Criterion.
  - Panel estimation covers 27 transition economies with annual data mostly from early 1990s to 2009; cross-section dependence tested and addressed using CCE estimators.

- A. Macroeconomic Balance (MB) approach — findings and caveats
  - Estimation approach: panel regression of current account on fundamentals (fiscal balance, old age dependency ratio, population growth, relative income, per capita GDP growth, lagged current account) using CGER panels.
  - Misalignment magnitudes:
    - Medium-term dram misalignment: slight overvaluation with MB estimates pointing to a misalignment of 3–4 percent depending on panel specification (full panel vs emerging-market-only panel).
  - Limitations:
    - Sensitive to WEO projection assumptions, price elasticities of exports/imports, and pooled OLS assumptions (coefficient homogeneity and exogeneity).

- B. External Sustainability (ES) approach — scenarios and sensitivity
  - Method: compute current account that stabilizes the net external position at a benchmark using CA* = (g + Π) * b where g = potential real GDP growth, Π = inflation, b = benchmark NEP/GDP.
  - Sensitivity and scenario results:
    - Estimates very sensitive to assumptions on potential GDP growth and inflation.
    - Depending on assumptions, 2015 misalignment ranges from an undervaluation of -6 percent to an overvaluation of 10 percent (Table A2).
    - Under g = 4 percent and inflation = 5 percent (within central bank target of 4 ± 1.5 percent), the CA norm is close to the actual current account projection and a real depreciation of only 1 percent is needed to close the gap between the actual current account in 2015 and the norm.
  - Limitations:
    - Inherits MB problems: sensitivity to elasticities, benchmark choice for NEP, and reliability of current account forecasts.

- C. ERER — single-country ARDL estimation (selected results, preserved exactly)
  - Model and data:
    - Fundamentals: openness, relative per capita GDP (Balassa-Samuelson), government consumption and government investment (relative), NIIP, terms of trade, remittances.
    - ARDL(3,0,4,2,3,0,0,4) estimated with 45 quarterly observations (1999Q1–2010Q1); dependent variable = LREER (log REER).
  - Long-run coefficient estimates (selected from Table 4; exact values preserved):
    - LTOT = 0.032 [T-Ratio 0.392; Prob 0.699]
    - LGC = 0.368 [T-Ratio 3.251; Prob 0.004]
    - LGFC = -0.061 [T-Ratio -0.808; Prob 0.428]
    - LOPEN = -0.684 [T-Ratio -6.041; Prob 0.000]
    - LRGDPPC = 0.064 [T-Ratio 1.077; Prob 0.294]
    - LREMIT = -0.013 [T-Ratio -0.291; Prob 0.774]
    - NIIP = 0.902E-4 [T-Ratio 4.335; Prob 0.000]
    - CONSTANT = 7.998 [T-Ratio 16.446; Prob 0.000]
  - Error-correction representation (selected coefficients from Table A4):
    - ecm(-1) = -0.919 (T-Ratio -5.5447[.000]) indicating a significant cointegrating relationship.
    - Goodness-of-fit: R-Squared = .88858; R-Bar-Squared = .76655.
  - Misalignment result:
    - ARDL single-country estimation: dram became overvalued sometime in 2007 and remained overvalued after the March 2009 depreciation.
    - End-2009 long-term overvaluation estimate = 11 percent.
  - Caveat: relatively small sample size (45 observations) limits inference robustness.

- D. REER — panel estimation with CCE (cross-country) — rationale and highlights
  - Panel sample: 27 transition economies (Albania, Armenia, Azerbaijan, Belarus, Bosnia & Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Macedonia, Moldova, Poland, Romania, Russia, Serbia, Slovak Republic, Slovenia, Tajikistan, Turkmenistan, Ukraine, Uzbekistan). Oil-producers identified: Azerbaijan, Kazakhstan, Russia, Turkmenistan, Uzbekistan.
  - Variables (logged): REER (dependent), TRT (terms of trade), OPN (openness), GDP per capita, GOV (government spending), INV (national investment).
  - Cross-section dependence:
    - Pesaran (2004) CD test strongly rejects cross-section independence; large positive CD statistics reported (e.g., fixed effects: 8.18***, 9.69***, 12.24***, 15.35***, 15.52***, 16.94***).
  - Estimation approach:
    - Common Correlated Effects estimators used: CCEMG (mean group) and CCEP (pooled) to accommodate unobserved common factors.
  - CCE results (selected long-run coefficient highlights from Table 6; exact values preserved where shown):
    - TOT (examples): CCEMG (1a) -0.03 (.15); CCEP (1b) 0.36*** (.08); CCEP (3b) 0.36*** (0.07).
    - OPN (examples): CCEMG (1a) -0.27 (.23); CCEP (1b) -.19*** (.06); CCEP (3b) -0.19*** (0.07).
    - GDP (examples): CCEMG (1a) 0.14 (.10); CCEP (1b) 0.07** (.04); CCEP (3b) 0.07* (0.04).
    - GOV (examples): CCEMG (1a) 0.28 (.18); CCEP (1b) 0.16* (.09).
  - Panel cointegration and unit-root evidence:
    - CADF/CIPS and IPS tests on residuals and variables show strong evidence rejecting unit roots in many specifications (selected CIPS and IPS statistics preserved in Appendix B).
    - Preferred panel models (by CADF and IPS): Model 2b (cointegration at the 5 percent level); Models 1b and 3b perform best overall.
  - Armenia misalignment from panel CCE (selected conclusions and figures):
    - Models 1b and 3b summary: dram undervalued between 2002 and 2006 by an average of 7–9 percent; began to be overvalued in 2007.
    - For 2009 results point to an overvaluation of about 7.3–9.2 percent (selected annual misalignment figures for 2009 across models include 7.3, 9.2, 7.5, 9.3, 9.0, 11.3 etc., as shown in Table 8).
  - Note: Most CCE models generate consistent direction of misalignment except Models 1a and 2a; CCEP methodology shows particular stability in direction of misalignment.

### IV. Synthesis of findings and policy-relevant points
- Convergent evidence of loss of external competitiveness:
  - Multiple indicators (exports share, current account deterioration, sectoral export declines, non-price competitiveness indices) point to a loss of external competitiveness in recent years.
- Exchange rate misalignment ranges by method (preserved exactly):
  - MB approach: dram estimated slightly overvalued by 3–4 percent (medium term, depending on panel specification).
  - ES approach: misalignment highly sensitive to g and Π; range across scenarios is from -6 percent to +10 percent in 2015; under g = 4 percent and inflation = 5 percent, a 1 percent real depreciation needed by 2015.
  - ERER single-country (ARDL): end-2009 long-term overvaluation estimated at 11 percent.
  - Panel CCE and ERER summary: ERER and panel approaches indicate overvaluation of 7–11 percent as of end 2009.
- Key parameters and data points (preserved exactly):
  - Imports elasticity to RER = 1.22.
  - Exports elasticity to RER = -1.08.
  - Implied elasticity of the current account to the real exchange rate = 0.45.
  - ARDL estimation: 45 observations from 1999Q1 to 2010Q1; ARDL(3,0,4,2,3,0,0,4).
  - ES scenario example: g = 4 percent and inflation = 5 percent.
- Methodological caveats and implications for policy interpretation:
  - MB and ES sensitivity to WEO projections, elasticity choices, and NEP benchmark selections implies wide scenario uncertainty for policy guidance.
  - Single-country ERER precision limited by sample size.
  - Panel estimators must account for cross-sectional dependence; CCE estimators mitigate bias from common shocks and provide robust panel cointegration evidence supporting the overvaluation conclusion.
- Policy-relevant takeaways:
  - Multiple methods and indicators point to a deterioration in external competitiveness and a real exchange rate that was overvalued in the late 2000s.
  - Magnitudes vary: modest overvaluation (3–4 percent) under MB; larger overvaluation (7–11 percent) under ERER and panel CCE estimates; ES results depend on macro assumptions (range -6 to +10 percent for 2015 scenarios).
  - Policy diagnostics should account for sensitivity to elasticity and growth/inflation assumptions; consideration of complementary structural reforms is implied by weak non-price competitiveness indicators.

### Appendix B (CCE estimators) — technical summary and robustness evidence
- CCE overview and properties:
  - Basic idea: augment observed regressors with cross-section averages of dependent and explanatory variables to filter unobserved common factors; consistent when N and T are sufficiently large.
  - Consistency: Kapetanios et al. (2009) prove CCE estimators are consistent regardless of stationarity of common factors and for any fixed number of common factors.
  - Practical advantage: no pre-testing required for the number of common factors.
- Testing for panel cointegration and unit roots:
  - Primary unit root test: CADF (Pesaran 2007) augmented with cross-section averages; CIPS is cross-section average of CADF t-ratios.
  - Robustness: IPS test and Pedroni’s panel cointegration tests applied as additional checks.
- Evidence of cross-sectional dependence:
  - Pesaran’s CD test (fixed/random effects) strongly rejects cross-section independence (selected CD statistics include 8.18***, 9.69***, 12.24***, 15.35***, 15.52***, 16.94*** and random effects 8.96***, 10.59***, 12.20***, 16.02***, 15.47***, 17.21***).
- Long-run coefficient patterns (selected highlights from Table 6):
  - CCEP estimates often show statistically significant TOT, OPN, GDP and GOV coefficients (examples: CCEP (1b) TOT = 0.36*** (.08); OPN = -0.19*** (.06); GDP = 0.07** (.04); GOV = 0.16* (.09)).
- Panel cointegration tests and residual stationarity:
  - CIPS and IPS statistics on residuals after long-run effects show widespread rejection of non-stationarity (selected CIPS and IPS values preserved in source).
  - CADF residual tests reject unit roots for most CCE models; IPS residual tests strongly suggest stationarity for all twelve models (significant at 1 percent).
  - Preferred models by these tests: Model 2b (CADF and IPS), Models 1b and 3b (best-performing in coefficients and cointegration evidence).
- Armenia-specific panel misalignment outcomes (selected)
  - Summary: dram undervalued 2002–2006 by an average of 7–9 percent; overvalued from 2007 and overvalued in 2009 by about 7.3–9.2 percent across preferred panel models.
- Model uncertainty and suggested extensions:
  - Sources: weak data quality, short time series, rapidly changing economic structures.
  - Robustness checks performed: CADF and IPS on residuals, Pedroni cointegration tests, alternative cross-section dependence tests (Free and Friedman).
  - Suggested further research: include additional fundamentals (e.g., remittances), model combination techniques (Bayesian averaging), relax long-run homogeneity assumptions using panel techniques allowing fixed differences across units/time.

*Source: IMF Working Paper — chapter "8. Exchange Rate Misalignment" and Appendix B (content as provided).*

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

### _wp1120 - References

### Figures
- 1.         Export         Volume         ................................................................................................................4
- 2.         Armenia’s         Current         Account         and Non-Commodity Exports. ..........................................5
- 3. Indicators of Armenia’s External Competitiveness ........................................................6
- 4. Effective Exchange Rate Developments .........................................................................6
- 5. GDP Dynamics and Global Competitiveness Rankings for Armenia ............................8
- 6. World Bank Ease of Doing Business Indicators .............................................................9
- 7. Results from the MB Approach .....................................................................................13
- 8. Results from the ES Approach .......................................................................................15
- 9.  ARDL Approach, Estimated Equilibrium Real Exchange Rates ...................................19

### Tables
- 1. Export Structure by Products ..........................................................................................5
- 2. World Economic Forum Competitiveness Indicators .....................................................8
- 3. Enterprise Survey of Selected Transition Economies ....................................................10
- 4. Estimated Long Run Coefficients using the ARDL Approach ......................................18
- 5. Pesaran’s Cross-Section Dependence Test ....................................................................22
- 6. Panel Estimation Results: Long-run Coefficients ..........................................................25
- 7. Tests for Panel Co-Integration  ......................................................................................26

### Appendix
- Appendix ....................................................................................................................................32

*Source: _wp1120 - References .............................................................................................................*

### 8. Exchange Rate Misalignment  .......................................................................................27

### 8. Exchange Rate Misalignment

### I. Introduction
- Purpose: assess whether Armenia’s real exchange rate is close to its equilibrium value to inform adjustment needs and trajectories of economic fundamentals.
- Methodologies used:
  - Macroeconomic Balance approach (MB) — current account misalignment-based.
  - External Sustainability approach (ES) — accounting-based current account norm to stabilize the net external position.
  - Equilibrium Real Exchange Rate approach (ERER) — reduced-form long-run relationship between the real exchange rate and fundamentals (single-country ARDL).
  - Panel estimation techniques addressing cross-section dependence using common correlated effects estimators (Pesaran, 2006) and a panel cointegration test accommodating common factors (Pesaran and Tosetti, 2010).
- Key headline result: ERER and panel approaches indicate the dram was overvalued as of end 2009 by 7-11 percent.

### II. Recent Developments (Competitiveness Evidence)
- Export performance and structure:
  - Armenia’s export volume increased between 2001 and 2008, then fell sharply at end-2008; exports started a modest recovery but remained at 2003 levels.
  - Precious and non-precious metals, precious and semi-precious stones and minerals constitute more than 65 percent of exports between 2000 and 2009.
  - Shift observed away from precious stones and metals to non-precious metals and minerals; notable declines in the share of machines, equipment and devices, textile products, plastic items and rubber — interpreted as clearer signs of a loss of competitiveness.
- Current account and tradables:
  - Current account as percent of GDP deteriorated markedly since 2004, reaching more than -16 percent in 2009.
  - Non-commodity exports declined as a percent of GDP until 2008.
  - Armenia’s share of world exports decreased significantly since 2007; exports per capita declined while output per capita grew until 2008.
- Price competitiveness (REER developments):
  - From 2006 until March 2009 the real effective exchange rate (REER) steadily appreciated.
  - In March 2009 the dram depreciated by 22 percent vis-à-vis the US dollar after returning to a floating exchange rate regime.
  - Between December 2009 and July 2010 the dram appreciated the most (both real and nominal) compared to major competitors.
- Non-price competitiveness (survey/indices):
  - World Economic Forum: Armenia ranked 98 out of 139 countries in 2010 (GCI score 3.76); outperformed only Tajikistan and the Kyrgyz Republic in the sample.
  - Key constraints (2010): lack of local competition, extent of market dominance, ineffectiveness of anti-monopoly policy — Armenia ranked 136, 133, and 138 out of 139 on these indicators respectively.
  - World Bank Ease of Doing Business: “Starting a business” rank = 21 out of 183 in 2010; “Protecting investors” rank = 93; “Paying taxes” rank = 153.
  - Enterprise surveys: high shares of firms identifying corruption, crime/theft/disorder and access to finance as major constraints; Armenia’s percentages exceed the transition-economy averages in Table 3.
- Conclusion from indicators: multiple price and non-price indicators point to a loss of external competitiveness in recent years.

### III. Exchange Rate Assessment — Methods, Key Parameters, and Results

- Common data/method notes:
  - MB and ES approaches rely on current account norms and the elasticity of the current account to the real exchange rate.
  - Armenia-specific elasticities used: imports elasticity to RER = 1.22; exports elasticity to RER = -1.08 (Tokarick, 2010); implied elasticity of the current account to the real exchange rate = 0.45 (formula and average import/export ratios 1999–2009 used).
  - ERER single-country estimation uses quarterly data with 45 observations from 1999Q1 to 2010Q1; ARDL cointegration (Pesaran and Shin, 1999) chosen with lag selection by Schwarz Bayesian Criterion.
  - Panel estimation covers 27 transition economies (listed in the source) with annual data mostly from early 1990s to 2009; cross-section dependence tested and addressed using CCE estimators.

- A. Macroeconomic Balance (MB) Approach
  - Estimation approach: panel regression of current account on fundamentals (fiscal balance, old age dependency ratio, population growth, relative income, per capita GDP growth, lagged current account), using both full CGER panel (54 advanced and emerging economies, 1969–2008) and an emerging-markets-only panel (42 countries).
  - Results and magnitudes:
    - With projected improvements in the actual current account, the CA norm is close to the current account projection.
    - Medium-term estimated dram misalignment: slight overvaluation with MB estimates pointing to a misalignment of 3-4 percent depending on whether the full panel or emerging-market-only panel is used.
  - Limitations highlighted:
    - MB results sensitive to WEO projection assumptions (policy adjustments), to the price elasticities of exports and imports, and to pooled OLS assumptions of coefficient homogeneity and exogeneity.

- B. External Sustainability (ES) Approach
  - Method: compute current account that stabilizes the net external position (NEP) at a benchmark level (often last observation) using formula CA* = (g + Π) * b where g = potential real GDP growth, Π = inflation, b = benchmark NEP/GDP.
  - Sensitivity and scenario results:
    - Estimates of misalignment are very sensitive to assumptions on potential GDP growth and inflation.
    - Depending on assumptions, 2015 misalignment ranges from an undervaluation of -6 percent to an overvaluation of 10 percent (Table A2).
    - Under assumption g = 4 percent and inflation = 5 percent (within central bank target of 4 ± 1.5 percent), the CA norm is close to the actual current account projection and a real depreciation of only 1 percent is needed to close the gap between the actual current account in 2015 and the norm.
  - Limitations noted:
    - ES inherits MB problems: sensitivity to elasticities, benchmark choice for NEP, and reliability of current account forecasts.

- C. Equilibrium Real Exchange Rate (ERER) — Single-Country ARDL Estimation
  - Fundamentals considered (per Bussiere et al., 2010): openness (trade/GDP), relative per capita GDP (Balassa-Samuelson), government consumption and government investment (relative), net international investment position (NIIP), terms of trade (TOT), remittances.
  - ARDL(3,0,4,2,3,0,0,4) estimated using 45 quarterly observations (1999Q1–2010Q1); dependent variable = LREER (log REER).
  - Long-run coefficient estimates (selected, exact values from Table 4):
    - LTOT coefficient = 0.032 [T-Ratio 0.392; Prob 0.699]
    - LGC (government consumption) coefficient = 0.368 [T-Ratio 3.251; Prob 0.004]
    - LGFC (government investment) coefficient = -0.061 [T-Ratio -0.808; Prob 0.428]
    - LOPEN coefficient = -0.684 [T-Ratio -6.041; Prob 0.000]
    - LRGDPPC coefficient = 0.064 [T-Ratio 1.077; Prob 0.294]
    - LREMIT coefficient = -0.013 [T-Ratio -0.291; Prob 0.774]
    - NIIP coefficient = 0.902E-4 [T-Ratio 4.335; Prob 0.000]
    - CONSTANT = 7.998 [T-Ratio 16.446; Prob 0.000]
  - Statistical inferences:
    - Government consumption, openness, NIIP and the constant are significant at the 1 percent level.
    - Diagnostics indicate evidence of a long-run cointegrating relationship (significant error correction term reported in Appendix A3).
  - Misalignment result:
    - Estimated exchange rate misalignment from ARDL: dram became overvalued sometime in 2007 and remained overvalued after the March 2009 depreciation.
    - End-2009 long-term overvaluation estimate = 11 percent (using ARDL single-country estimation).
  - Caveat: relatively small sample size (45 observations) limits inference robustness.

- D. Real Effective Exchange Rate — Panel Estimation (Cross-country)
  - Panel sample: 27 transition economies (Albania, Armenia, Azerbaijan, Belarus, Bosnia & Herzegovina, Bulgaria, Croatia, Czech Republic, Estonia, Georgia, Hungary, Kazakhstan, Kyrgyz Republic, Latvia, Lithuania, Macedonia, Moldova, Poland, Romania, Russia, Serbia, Slovak Republic, Slovenia, Tajikistan, Turkmenistan, Ukraine, Uzbekistan). Oil producers identified among them: Azerbaijan, Kazakhstan, Russia, Turkmenistan, Uzbekistan.
  - Variables in panel: REER (dependent), TRT (terms of trade), OPN (openness), GDP per capita, GOV (government spending), INV (national investment). All variables converted to natural logs.
  - Cross-section dependence:
    - Pesaran (2004) cross section dependence (CD) test strongly rejects cross-section independence in the sample.
    - Because of cross-sectional dependence, Common Correlated Effects (CCE) estimators are used: CCEMG (mean group) and CCEP (pooled).
  - Rationale: CCE models accommodate cross-section correlations that are likely given trade linkages and common shocks among these countries.
  - (Detailed panel estimation results and misalignment magnitudes are presented in the paper’s subsequent sections and appendices; panel approach results in the paper align with ERER findings indicating overvaluation in the late 2000s.)

### IV. Synthesis of Findings and Policy-Relevant Points
- Convergent evidence of loss of external competitiveness:
  - Multiple indicators (exports share, current account deterioration, sectoral export declines, non-price competitiveness indices) point to a loss of external competitiveness in recent years.
- Exchange rate misalignment estimates vary by method and assumptions:
  - MB approach: dram estimated slightly overvalued by 3-4 percent (medium term, depending on panel specification).
  - ES approach: misalignment is highly sensitive to assumed potential growth and inflation; range across scenarios is from -6 percent (undervaluation) to +10 percent (overvaluation) in 2015; under g = 4 and inflation = 5 percent, a 1 percent real depreciation needed by 2015.
  - ERER single-country (ARDL): end-2009 long-term overvaluation estimated at 11 percent.
  - Introductory summary: ERER and panel approaches indicate overvaluation of 7-11 percent as of end 2009.
- Key parameters and elasticities used (preserved exactly):
  - Imports elasticity to RER = 1.22.
  - Exports elasticity to RER = -1.08.
  - Implied elasticity of the current account to the real exchange rate = 0.45 (computed using average import to export ratios between 1999 and 2009).
  - ARDL estimation: 45 observations from 1999Q1 to 2010Q1; ARDL(3,0,4,2,3,0,0,4).
  - ES scenarios example: g = 4 percent and inflation = 5 percent used as a plausible medium-term scenario.
- Methodological caveats emphasized in the analysis:
  - MB and ES are sensitive to WEO projection assumptions, elasticity choices, and arbitrary benchmark selection for NEP.
  - Single-country ERER inference limited by sample size.
  - Panel estimators must address cross-sectional dependence; CCE estimators (CCEMG and CCEP) are applied because the Pesaran CD test rejected independence.

*Italic: Source — IMF Working Paper chapter "8. Exchange Rate Misalignment" (content as provided).*

### Appendix B provides a technical discussion of CCE estimators. This section will provide a

### _wp1120 - Appendix B provides a technical discussion of CCE estimators. This section will provide a

### Overview of the CCE estimation approach
- Basic idea: filter individual-specific regressors by means of cross-section averages so that asymptotically the differential effects of unobserved common factors are accommodated.
- Implementation: observed regressors are augmented by cross-section (weighted) averages of the dependent variable and the individual specific regressors and estimation can be conducted by ordinary least squares (OLS) applied to an auxiliary regression.
- Key property: invariance to the (unknown but fixed) number of unobserved common factors when N and T are sufficiently large.
- Consistency results: Kapetanios et al. (2009) prove CCE estimators are consistent regardless of whether the common factors are stationary or nonstationary and produce consistent estimates with any fixed number of common factors.
- Practical advantage: no pre-testing required for the number of common factors.

### Testing for panel cointegration and unit roots
- Strategy: test for panel cointegration using residuals from CCE estimation; absence of a unit root in residuals establishes panel cointegration.
- Primary unit root test: cross-section augmented Dickey Fuller (CADF) tests (Pesaran 2007) that augment standard ADF regressions with cross-section averages to filter cross-sectional dependence (second-generation unit root tests).
- Robustness check: Im-Pesaran-Shin (IPS) test (Im et al., 2003) as a first-generation panel unit root test (does not account for cross-sectional correlations).
- Additional robustness: Pedroni’s (1999, 2001 and 2004) panel cointegration test used as further robustness check (assumes single common time effect with similar responses across units).

### Evidence of cross-sectional dependence (justifying CCE)
- Pesaran’s CD test results (fixed effects and random effects) strongly reject cross-section independence across model specifications and samples (with and without oil-producers).
- Selected CD statistics (fixed effects): 8.18***, 9.69***, 12.24***, 15.35***, 15.52***, 16.94***.
- Selected CD statistics (random effects): 8.96***, 10.59***, 12.20***, 16.02***, 15.47***, 17.21***.
- Note: H0: cross-section independence. Large positive test statistics reject the null. *** denotes significance at 1 percent, ** at 5 percent and * at 10 percent levels.

### CCE model specifications and long-run coefficient estimates (summary)
- Model sets: five-regressor model (TRT, OPN, GDP, GOV, INV), four-regressor model (TRT, OPN, GDP, GOV), three-regressor model (TRT, OPN, GDP).
- Estimators reported: CCEMG and CCEP versions across full sample and sub-samples (oil-producers only, non oil-producers only).
- Selected long-run coefficient highlights from Table 6 (standard errors in parentheses; significance denoted by ***/**/*):
  - TOT coefficients (examples): CCEMG (1a) -0.03 (.15); CCEP (1b) 0.36*** (.08); CCEMG (2a) -0.02 (.29); CCEP (2b) 0.23** (.10); CCEP (3b) 0.36*** (0.07).
  - OPN coefficients (examples): CCEMG (1a) -0.27 (.23); CCEP (1b) -.19*** (.06); CCEP (3b) -0.19*** (0.07); CCEP (5b) -0.23*** (.07).
  - GDP coefficients (examples): CCEMG (1a) 0.14 (.10); CCEP (1b) 0.07** (.04); CCEP (3b) 0.07* (0.04); CCEP (5b) 0.07 (.05).
  - GOV coefficients (examples): CCEMG (1a) 0.28 (.18); CCEP (1b) 0.16* (.09); CCEP (3b) 0.15* (0.08).
  - INV coefficient reported in five-regressor: CCEMG (1a) 0.05 (.08); CCEP (1b) 0.03 (0.07).
- Sample sizes reported in Table 6:
  - Obs: 461, 461, 384, 384, 461, 461, 384, 384, 475, 475, 395, 395 (across models).
  - Countries: 27, 27, 22, 22, 27, 27, 22, 22, 27, 27, 22, 22.

### Panel cointegration test results (Table 7)
- CIPS test for residuals after long-run effect and country specific intercept (A1) — selected statistics:
  - lag(0) examples: -1.85***, -2.02***, -1.91***, -2.40***, -1.89***, -1.88***, -1.97***, -2.25***, -1.48, -1.89***, -1.16, -2.00***.
  - lag(1) examples: -2.76***, -2.72***, -2.26***, -1.84**, -2.86***, -2.72***, -2.61***, -2.40***, -2.68***, -2.55***, -1.71**, -2.19***.
  - With const: lag(1) examples: -2.05, -2.56***, -1.70, -2.34**, -2.07*, -2.54***, -2.08, -1.12*, -1.68, -2.15*, -2.18**, -1.96.
- IPS test for residuals after long-run effect and country specific intercept (A2) — selected statistics:
  - const examples: -10.45***, -10.2***, -7.09***, -9.33***, -9.95***, -9.95***, -8.90***, -9.33***, -9.65***, -9.83***, -9.25***, -7.8***.
  - const + trend examples: -8.61***, -6.50***, -6.22***, -5.76***, -6.07***, -6.07***, -6.46***, -5.46***, -5.88***, -6.16***, -6.07***, -7.23***.
- Note: H0: unit root exists (non-stationary); large negative values imply rejection of the null. *** denotes rejection at the 1 percent significance level, ** at 5 percent and * at 10 percent. CIPS statistics are based on a balanced panel (data 1999-2009). IPS statistics are based on full (unbalanced) panel.

### Residual stationarity and model selection guidance
- CADF tests on residuals from twelve CCE models: CADF rejects unit roots in the residuals for all twelve CCE models in at least one specification.
  - Model (2b) rejects non-stationarity at the five or one percent level in all four test specifications.
  - Models (1b), (3a), (3b), (4b), and (5b) reject non-stationarity at conventional levels in three of four specifications.
- IPS tests on residuals: IPS results strongly suggest all residuals are stationary; all twelve models yield test statistics significant at the 1 percent level.
- Model candidates based on tests:
  - Preferred by CADF and IPS: Model 2b (all CADF and IPS residual tests show cointegration at the 5 percent level).
  - Models with most significant long-run coefficients: Model 3b (all four regressors significant) and Model 1b (four of five regressors significant).
  - Overall best-performing models for estimation and cointegration: Models 1b and 3b (perform best in significant coefficients and show significant evidence of cointegration).

### Exchange rate misalignment results for Armenia (selected)
- Summary conclusion from models 1b and 3b:
  - Armenian dram undervalued between 2002 and 2006 by an average of 7–9 percent.
  - Beginning in 2007 dram became overvalued.
  - For 2009 results point to an overvaluation of about 7.3–9.2 percent.
- Selected annual misalignment figures for Armenia from Table 8 (in percent), examples across models:
  - 2000: 2.3, 3.4, -10.8, 6.1, 2.8, 3.3, 4.2, 4.9, 4.4, 4.9, 4.5, 6.2.
  - 2001: 1.4, 3.9, -18.9, 5.7, -0.8, 4.0, 0.8, 5.0, 0.2, 3.5, 0.6, 5.8.
  - 2002: -3.6, -1.2, -19.2, -2.4, -4.7, -0.1, -4.4, -0.9, -5.8, -2.6, -5.4, 0.6.
  - 2003: -10.7, -9.7, -23.0, -12.1, -11.1, -8.8, -11.1, -10.6, -11.6, -10.2, -11.5, -10.1.
  - 2004: -12.0, -11.8, -15.6, -13.6, -13.6, -11.3, -14.0, -12.3, -15.2, -13.5, -14.6, -11.7.
  - 2005: -7.5, -8.7, -9.0, -9.3, -8.0, -8.1, -7.6, -8.5, -7.4, -8.7, -7.8, -8.2.
  - 2006: -3.9, -9.1, 5.6, -9.2, -7.2, -9.0, -6.7, -8.2, -6.9, -9.4, -6.9, -8.7.
  - 2007: 4.8, 1.1, -0.1, 1.6, 0.9, 0.8, 2.0, 2.0, 1.5, 0.6, 2.0, 2.5.
  - 2008: 7.0, 7.9, 21.5, 7.1, 11.2, 7.3, 8.5, 5.6, 9.7, 7.8, 8.7, 3.7.
  - 2009: -0.8, 7.3, -18.0, 9.2, 1.4, 7.5, 3.6, 9.3, 3.0, 9.0, 4.1, 11.3.
- Note: Most CCE models generate consistent direction of misalignment except Models 1a and 2a; most differing results occur between 2005 and 2009. CCEP methodology shows particular stability in direction of misalignment.

### Model uncertainty, robustness, and extensions
- Sources of model uncertainty: weak data quality, short time series, rapidly changing economic structure across sample countries.
- Robustness checks performed: CADF and IPS on residuals, Pedroni cointegration tests, alternative cross-section dependence tests (Free and Friedman) confirming CD test results.
- Suggested further research directions:
  - Experiment with different economic fundamentals to expand cointegration space (example suggested: remittances).
  - Explore model combination techniques (estimating many specifications and combining with Bayesian averaging).
  - Relax long-run homogeneity assumption by employing panel techniques allowing fixed differences across units and time (Hsiao and Pesaran, 2008).

### Key conclusions relevant for policy and interpretation
- Panel CCE estimation explicitly models cross-country common effects and is more suitable than first-generation panel approaches when cross-sectional dependence is present.
- Across four independent approaches (Macroeconomic Balance, External Sustainability, ERER single-country, and CCE panel), findings are consistent:
  - Armenia dram was undervalued between 2003 and late 2006.
  - Dram became overvalued sometime in 2007 and remained so after March 2009 depreciation.
  - Current overvaluation estimated at around 10 percent and corroborated by Armenia’s loss in world export share and deteriorating competitiveness indicators.

*Source: Excerpt from the specified IMF working paper section on CCE estimators and panel estimation results.*

### REFERENCES

### _wp1120 - REFERENCES

### Key cited methodology and empirical literature
- Breitung, Joerg, and Hashem Pesaran, 2008, “Unit Roots and Cointegration in Panels.”
- Bussiere, Matthieu, Michele Ca’ Zorzi, Alexander Chudik, and Alistair Dieppe, 2010, “Methodological Advances in the Assessment of Equilibrium Exchange Rates,” ECB Working Paper Series No 1151.
- De Hoyos, Rafael E., and Vasilis Sarafidis, 2006, “Testing for Cross–Sectional Dependence in Panel–Data Models,” The Stata Journal, Vol. 6, No.4, pp. 482-496.
- Fisheresque and panel-data methodology references emphasized: Pesaran (2004, 2006, 2007), Pedroni (1999, 2004), Im, Pesaran, and Shin (2003), Phillips and Sul (2003), and Pesaran and Shin (1999, 2001).
- Applied IMF/World Bank/ECB working papers and case studies on exchange rate assessment, remittances, aid, and external wealth: Lane and Milesi-Ferretti (2007); Lee et al. (2008); Mongardini and Rayner (2009); Lee, Kyung-woo, Markus Haacker, and Raju Singh (2009); Ooomes, Stepanyan, and Minasyan (2009).

### Appendix A — Tables: scope, data sources, and key indicators
- Table A1: MB Approach, Data sources and Variable Construction
  - Variables constructed include:
    - Fiscal Balance (FB): general government balance relative to trading partners; weights w(i)= (0.47, 0.16, 0.15, 0.11, 0.10) for i=(Russia, China, Germany, Ukraine, USA).
    - Old Age Dependency (OAD): ratio of old age (65 and over) to middle age (30-64).
    - Population Growth (PG): ln(POP_t) - ln(POP_{t-15}) aggregated across trading partners.
    - Relative Income (RI): Real GDP per capita relative to US real GDP per capita.
    - Per Capita GDP Growth (PCG): relative to trading partners.
    - Current account (CA): CA/GDP (WEO).
  - Primary sources noted: WEO, UN.

- Table A2: ES Approach — Exchange Rate Misalignment (in percent) under different assumptions for g and Π
  - Presents misalignment percentages for years 2005–2015 under combinations of Π = 2.5, 4, 5, 6, 7 and growth scenarios high (g=6), normal (g=4), low (g=2).
  - Example figures (selected cells preserved as in source):
    - Π = 2.5, high growth(g=6): 2005 = 0.1; 2006 = 2.6; 2007 = 12.6 22.7 29.5 23.2 18.9 13.9 9.1 3.9 2.0 (table formatting as in source).
    - Π = 4, normal(g=4): 2005 = 0.4; 2006 = 2.9; 2007 = 12.8 23.0 30.0 23.7 19.6 14.6 9.9 4.8 2.9 (table formatting as in source).
  - (Full matrix of yearly misalignment values is provided in the source table for each Π and g combination.)

- Table A3: ERER Approach, Data Sources and Variable Construction
  - Variables and sources include:
    - Real effective exchange rate (LREER) — Central Bank of Armenia.
    - Openness (LOPEN): ln(exports + imports)/GDP, seasonally adjusted — IFS.
    - Productivity (LRGDPPC): ln per capita real GDP relative to trading partners, seasonally adjusted — IFS.
    - Government consumption (LGC): ln government consumption/nominal GDP relative to trading partners — IFS.
    - Investment (LGFC): ln government investment/nominal GDP relative to trading partners — IFS.
    - NIIP, Terms of Trade (LTOT), Remittances (LREMIT) — national authorities and staff calculations.

- Table A4: Error Correction Representation for the Selected ARDL Model
  - Model: ARDL(3,0,4,2,3,0,0,4) selected based on Schwarz Bayesian Criterion.
  - Dependent variable: dLREER; 45 observations used for estimation from 1999Q1 to 2010Q1.
  - Selected estimated coefficients (coefficient, standard error, T-Ratio[Prob]):
    - dLREER1: 0.586, 0.129, 4.5249[.000]
    - dLREER2: -0.337, 0.147, -2.3007[.030]
    - dLGC1: -0.305, 0.083, -3.6587[.001]
    - dLGC2: -0.236, 0.068, -3.4697[.002]
    - dLGC3: -0.148, 0.053, -2.7985[.010]
    - dLGFC: -0.302, 0.093, -3.2516[.003]
    - dLOPEN1: 0.520, 0.128, 4.0599[.000]
    - dLOPEN2: 0.233, 0.105, 2.2121[.036]
    - dNIIP1: 0.000, 0.000, -4.5068[.000]
    - dNIIP3: 0.000, 0.000, -3.1450[.004]
    - dCONSTANT: 7.349, 1.514, 4.8551[.000]
    - ecm(-1): -0.919, 0.166, -5.5447[.000]
  - Error-correction term ecm specified as:
    - ecm = LREER  -.032270*LTOT   -.36783*LGC +  .061397*LGFC +   .68406*LOPEN  -.064338*LRGDPPC +  .013104*LREMIT -.9017E-4*NIIP   -7.9977*CONSTANT
  - Goodness-of-fit and diagnostics:
    - R-Squared = .88858
    - R-Bar-Squared = .76655
    - S.E. of Regression = .021772
    - F-stat. F( 19,  25) = 8.8146[.000]
    - Mean of Dependent Variable = .0022387
    - S.D. of Dependent Variable = .045061
    - Residual Sum of Squares = .0099546
    - Equation Log-likelihood = 125.5163
    - Akaike Info. Criterion = 101.5163
    - Schwarz Bayesian Criterion = 79.8364
    - DW-statistic = 2.3666

- Table A5: ERER Panel Approach, Data Sources and Variable Construction
  - Variables and sources include:
    - Real effective exchange rate (RER) — IMF.
    - Openness (OPN): ln(exports + imports)/GDP — IFS.
    - Productivity (GDP): ln per capita real GDP relative to trading partners — IFS.
    - Government consumption (GOV), Investment (INV) — IFS.
    - Terms of Trade (TOT) — IMF; Oil Price index (2005=100) — IMF.

- Table A6: CADF Test for Unit Roots (selected reported values)
  - CADF level and first-difference statistics (selected entries preserved exactly):
    - Level, constant, lag(1): RER = -3.39***; TOT = 0.362; OPN = -3.42***; GDP = -2.36**; GOV = -3.44***; INV = -4.83***.
    - First diff, constant, lag(1): RER = -8.30***; TOT = -5.47***; OPN = -4.86***; GDP = -7.70***; GOV = -8.73***; INV = -4.98***.
  - IPS test (level and first diff):
    - Level, IPS: RER = -1.16; TOT = 0.17; OPN = -4.43 ***; GDP = -0.24; GOV = -3.54***; INV = -.71.
    - First diff, IPS (Lag # chosen by AIC): RER = -18.14***; TOT = -15.93***; OPN = -22.30***; GDP = -21.31***; GOV = -15.91***; INV = -11.81***.
  - Note on interpretation: H0: unit root exists; large negative values imply rejection. Significance markers: ***, **, * as in source.

- Table A7: Pedroni’s panel co-integration test (with common time trend)
  - Selected test statistics (preserving values and significance):
    - Five-regressor model, full sample: Panel PP = -2.08***; Panel ADF = -0.72; Group PP = -3.70***; Group ADF = -2.87***.
    - Five-regressor model, oil-producers only: Panel PP = -2.01***; Panel ADF = -0.44; Group PP = -3.19***; Group ADF = -2.49***.
    - Four-regressor model, full sample: Panel PP = -1.68*; Panel ADF = -2.33***; Group PP = -2.67***; Group ADF = -3.39**.
    - Three-regressor model, full sample: Panel PP = -2.83***; Panel ADF = -2.71***; Group PP = -2.31***; Group ADF = -2.49***.
  - Note: H0: no cointegration; large negative values imply rejection. First statistic in each column may be Panel V-statistic (see source footnote).

### Appendix B — Technical Appendix: Estimation methods and panel tests
- Common Correlated Effects (CCE) estimators — conceptual summary and assumptions (equations preserved from source):
  - Model structure (as in source):
    - y_{i,t} = alpha_i' x_{i,t} + u_{i,t}              (7)
    - u_{i,t} = lambda_i' f_t + epsilon_{i,t}          (8)
    - x_{i,t} = delta_i' f_t + v_{i,t}                  (9)
  - Unobserved common factors f_t (t = 1,...,T) are m-dimensional; m unknown but fixed.
  - The approach uses cross-section (weighted) averages as observable proxies for f_t; cross-section averages of dependent and explanatory variables are used to filter out differential effects of unobserved common factors.
  - Asymptotic argument: f_t ≈ (xi' xi)^{-1} xi' xbar_t as N,T → ∞ (equation references preserved from source).

- Panel Unit Root Tests — CADF and CIPS
  - CADF regression (as in source):
    - Δy_{i,t,τ} = a_{i,τ} + b_{i,τ} t + c_{i,τ} y_{i,t-1} + d_{i,τ} ybar_{t-1} + Σ_{j=1}^p φ_{i,τ,j} Δy_{i,t-j} + e_{i,t,τ}      (13)
  - CIPS statistic: simple cross-section average of OLS t-ratios of c_{i,τ} from CADF regressions.
  - Notes and assumptions in source regarding idiosyncratic shocks e_{i,t,τ} and optional inclusion of constant and trend.

*Source: _wp1120 - REFERENCES (PDF chapter/section and appendices)*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1120.pdf_