## _wp05192

## Source details

**Canonical URL:** [_wp05192](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05192.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05192.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05192.pdf.json)

---

### I. Introduction — context and main question
- The 15 post-Soviet countries reoriented from a centralized command economy toward more decentralized, diversified market economies, weakening former strong interconnections and increasing links to the rest of the world.
- The paper tests whether adding Russian real GDP growth to a standard growth regression for the other CIS and Baltic countries shows a change in the relationship related to the Russian crisis, using annual data for 1993–2003 pooled across the 13 CIS and Baltic countries (excluding Turkmenistan and Russia) in a one-way error component regression model.
- Econometric evidence supports the hypothesis of a weakening of the link that broadly coincided with the crisis; possible transmission channels considered include declines in trade and financial flows.

### II. Data, patterns, and limitations
- Data sources:
  - Annual real GDP growth, CPI inflation, and government expenditures as a percentage of GDP from the IMF World Economic Outlook database.
  - Initial conditions from Havrylyshyn and van Rooden (2000); reform index from EBRD reform indices.
  - Merchandise trade data from IMF Direction of Trade database.
- Stylized patterns:
  - Real GDP: most CIS countries exhibit a “smile” pattern; Baltics show an increasing output trajectory.
  - CPI inflation: rapid disinflation in early 1990s and low inflation thereafter in many cases.
  - Government expenditure as percent of GDP: varied movements across countries.
- Data weaknesses and consequences:
  - GDP data likely include biases from underreporting by private enterprises.
  - Government expenditure data may not be fully comparable across countries due to transparency, definitions, and coverage differences.
  - CPI inflation data quality may vary.
  - Years prior to 1993 excluded; Turkmenistan excluded due to data quality.

### III. Econometric model and identification of a structural break
- Model specification:
  - One-way error component regression model with y_it = real GDP growth for country i in year t; y_Rt = real GDP growth for Russia in year t; x_it = vector of exogenous determinants; u_it = μ_i + v_it.
  - When x_it includes current and one-period lags, model is a first-order autoregressive distributed lag model.
- Structural break approach:
  - Full specification interacts explanatory variables with a dummy t_d equal to 0 prior to the Russian crisis and 1 thereafter to capture shifts in coefficients.
  - Uncertainty about exact timing: search within a neighborhood of the crisis for the most likely shift point; 1998 identified as the most likely break point.
- Estimation issues:
  - Inclusion of a lagged dependent variable gives rise to bias in standard estimators of fixed or random effects models.
  - Under fixed country effects, initial conditions must be excluded from x_it because they are collinear with country effects (can be included under random effects).
  - Arellano–Bond instrumental variables estimators used to address bias from lagged dependent variables.

### IV. Main econometric findings (empirical specification and coefficients)
- Column (1) (general specification excluding lagged independent variables):
  - Coefficient on Russian growth: 0.96 and significant.
  - Coefficient on Russian crisis dummy interacted with Russian growth: –0.91 and highly significant.
  - Interpretation: before the Russian crisis, a one percentage point increase in Russian growth was associated with a 0.96 percentage point increase in another country’s growth rate; after the crisis the effect dropped to 0.05 percentage points and was not significantly different from zero.
  - Own-country effect rose in 9 of 12 cases following the crisis (reflected in many positive and significant coefficients on interactions between the Russian crisis dummy and country dummies).
  - CPI inflation: coefficient not significant and near zero; interaction with crisis dummy also near zero and not significant.
  - Government expenditure (percent of GDP): insignificant and essentially near zero; its interaction with the crisis dummy is –0.5 and statistically significant.
  - Growth in the EU: coefficient –2.9 and significant prior to 1998; insignificant and close to zero in 1998–2003.
  - Lagged Russian growth: not significant.
- Column (2) (alternative specification including initial condition measure IC2 but not country effects):
  - Estimated coefficient on Russian growth: 0.67 and highly significant prior to 1998; falls to 0.19 and is insignificant thereafter.
  - CPI and government expenditure statistically significant in this specification, though CPI coefficient remains near zero.
  - IC1 not significant; IC2 significant.
- Regional dummies and other specifications:
  - Regional-dummies specification produces results broadly similar to column (1).
  - Excluding the Baltics (column (4)): coefficient on Russian growth is 1.37 and highly significant; coefficient on Russian crisis dummy interacted with Russian growth is –1.32 and highly significant; their difference is 0.06 and is insignificant.
  - Column (7) (alternative real exchange rate measure): coefficient on Russian growth is 0.78 and highly significant prior to 1998; falls to 0.06 and becomes insignificant thereafter.
  - Substituting world growth for EU growth (column (10)): coefficient on world growth is insignificant; coefficients on Russian growth similar to earlier results.
- Arellano–Bond vs LSDV:
  - Arellano–Bond estimates yield coefficients on Russian growth broadly comparable to LSDV results; coefficient significant in one of two Arellano–Bond specifications reported.
  - Arellano–Bond estimates of interaction between crisis dummy and Russian growth comparable and significant in both specifications reported.
  - Sargan tests do not reject overidentifying restrictions; Arellano–Bond serial correlation tests do not reject null of no second-order serial correlation.
  - Caution: Arellano–Bond may exhibit large bias in finite samples and higher standard errors; endogeneity-corrected coefficients should be treated with caution.

### V. Structural break timing and robustness
- Structural break timing:
  - Structural change likely occurred in either 1998 or 1999; tests and model-fit comparisons favor 1998 as the break point.
  - Regressions using a break point of 1999 have lower log-likelihood, higher AIC, fewer significant t-statistics, and less precise coefficients compared with regressions using a break point of 1998.
  - Coefficient on Russian crisis dummy interacted with Russian growth is not significant in any 1999-break specification.
  - Andrews’s (1993) Sup W, Sup LM, and Sup LR tests did not reject parameter constancy, but results are inconclusive due to limited observations and contamination from imprecisely estimated control variables.
  - Rationale for choosing 1998: expectations converged by early 1998 that macroeconomic imbalances were emerging in Russia; links may have been disrupted even before the August 17, 1998 ruble float.
- Robustness summary:
  - Main finding of a weakened link post-1998 is robust to choice of explanatory variables and estimation methods, though there is substantial variation across countries.
  - Results are insensitive to inclusion/exclusion of the Baltics in many specifications.

### VI. Possible transmission mechanisms (empirical evidence)
- Trade:
  - Merchandise exports from CIS and Baltic countries to Russia (as percent of total merchandise exports) were generally on a declining trend during 1993–2003.
  - Drop in exports to Russia as share of total exports during 1997–99 pronounced for the Baltics; more gradual for the CIS and started prior to the crisis.
  - The share of exports to Russia in total exports was statistically insignificant in alternative growth regressions.
  - Aggregate growth decompositions provide little support for net external demand as a crucial driver of real GDP growth in these countries; exports to Russia generally contributed only a small amount to real GDP growth and little evidence that this contribution declined following the Russian crisis.
  - Indirect trade effects (domestically produced inputs for export production) may be important and are difficult to capture.
- Capital flows and remittances:
  - Net capital inflows slowed markedly during the crisis.
  - Data on capital inflows from Russia are unavailable.
  - It is reasonable to suppose some additional capital inflows into Russia in the post-crisis period may have been invested in other CIS countries, which could strengthen rather than weaken growth linkages.
  - Additional data collection on capital inflows from Russia and on worker remittances would help clarify financial transmission channels.
- Transition-related effects:
  - Common output decline 1993–98 and differential rebounds thereafter may account for high pre-crisis correlations and lower post-crisis correlations.

### VII. Main empirical summary and key statistics
- Pre-crisis (before 1998) relationship:
  - Russian economic growth was a significant determinant of growth in other CIS and Baltic countries.
  - Each percentage point of additional growth in Russia led to around 0.8–0.9 percentage points of additional growth in the other countries on average, holding other factors constant.
- Post-crisis (beginning in 1998) relationship:
  - A 1 percentage point increase in Russian real GDP appears to have induced an increase in real GDP of the other CIS and Baltic countries of less than 0.2 percentage points on average.
  - The link was statistically insignificant in the period starting in 1998 (identified as the most likely break point).
- Sample numerical highlights (from tables and appendices):
  - Simple correlation coefficients between Real GDP growth in Russia and other CIS and Baltic countries, 1993–2003 (Average across countries):
    - CIS: 0.81 (1993–97); 0.27 (1998–2003); 0.74 (1993–98)
    - Energy exporters: 0.91 (1993–97); 0.32 (1998–2003); 0.82 (1993–98); –0.02 (1999–2003)
    - Energy importers: 0.77 (1993–97); 0.25 (1998–2003); 0.70 (1993–98)
    - Baltics: 0.73 (1993–97); 0.06 (1998–2003); 0.70 (1993–98); 0.03 (1999–2003)
  - Selected country correlations (Appendix Tables 1a–1d): examples include RUS—UKR: 0.98 (1993–97), RUS—EU: 0.61 (1999–2003).
  - Merchandise exports to Russia as percent of total merchandise exports (selected country-year values):
    - Belarus: 1997 64.2; 1998 65.2; 1999 54.5; 2003 49.1
    - Estonia: 1993 22.6; 1998 13.3; 1999 9.2; 2003 11.4
    - Latvia: 1993 28.5; 1998 12.0; 1999 6.6; 2003 5.4
    - CIS excl. Russia (aggregate): 1994 40.1; 1998 33.8; 2003 22.7
  - Merchandise imports from Russia as percent of total merchandise imports (selected country-year values):
    - Belarus: 1993 46.0; 1998 54.8; 2003 65.8
    - Estonia: 1993 17.2; 1998 11.1; 2003 10.2
    - CIS excl. Russia (aggregate): 1994 47.2; 1998 41.4; 2003 39.0
  - Appendix Table 3 selected real GDP series (percentage changes):
    - Russia — Real GDP: 1993 -8.7; 1994 -12.7; 1995 -4.1; 1996 -3.6; 1997 1.4; 1998 -5.3; 1999 6.3; 2000 10.0; 2003 7.3
    - Estonia — Real GDP: 1993 -8.2; 1997 10.5; 1999 -0.1; 2003 5.1
    - Kazakhstan — Real GDP: 1993 -9.2; 1998 -1.9; 1999 2.7; 2000 9.8; 2001 13.5
  - Averages (Appendix Table 3 footer): 3.3; -0.3; 1.0; 0.5

### VIII. Policy and research implications / recommendations
- Further investigation needed into the links between trade and financial flows in the region, preferably at a disaggregated level, to explain the significant break in growth linkages observed in the post-1998 period.
- Consideration of indirect trade effects (value added via domestically produced inputs for export production) should be part of future work because indirect effects may exceed direct export contributions.
- Additional data collection on capital inflows from Russia and on worker remittances would help clarify financial transmission channels.
- Given estimator biases in short panels with lagged dependent variables:
  - Apply instrumental-variable methods (e.g., Arellano–Bond) with caution and complement them with alternative estimators.
  - Interpret finite-sample results carefully.

*Italic: IMF Direction of Trade and World Economic Outlook databases; Fund staff estimates.*

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

### _wp05192 - References..............................................................................................................

### I. Introduction — context and main question
- The 15 post-Soviet countries reoriented from a centralized command economy toward more decentralized, diversified market economies, weakening former strong interconnections and increasing links to the rest of the world.
- There was a precipitous drop in the simple correlations between real GDP growth in Russia and growth in the other CIS and Baltic countries roughly coincident with the 1998 Russian crisis.
- The paper tests whether adding Russian real GDP growth to a standard growth regression for the other CIS and Baltic countries shows a change in the relationship related to the Russian crisis, using annual data for 1993–2003 pooled across the 13 CIS and Baltic countries (excluding Turkmenistan and Russia) in a one-way error component regression model.
- Econometric evidence supports the hypothesis of a weakening of the link that broadly coincided with the crisis; the paper also considers possible transmission channels including declines in trade and financial flows.

### Key statistic from Table 1 — Simple correlation coefficients between Real GDP growth in Russia and other CIS and Baltic countries, 1993–2003 (Average across countries)
- CIS: 0.81 (1993–97); 0.27 (1998–2003); 0.74 (1993–98); — (1999–2003)
- Energy exporters: 0.91 (1993–97); 0.32 (1998–2003); 0.82 (1993–98); –0.02 (1999–2003)
- Energy importers: 0.77 (1993–97); 0.25 (1998–2003); 0.70 (1993–98); —
- Baltics: 0.73 (1993–97); 0.06 (1998–2003); 0.70 (1993–98); 0.03 (1999–2003)

### II. Previous studies — literature positioning
- Prior literature mainly analyzed transmission of business cycles among industrial countries; some studies address developing countries or regional influences by including a major partner’s growth as an explanatory variable (examples cited: Arora and Vamvakidis 2004, 2005a, 2005b).
- Empirical studies of the former Soviet Union have largely focused on determinants of growth within each country rather than regional growth linkages and mainly refer to the period prior to the Russian crisis.
- Determinants identified in past studies fall into three categories:
  - Initial conditions (including institutions) — two “clusters” used: IC1 (macroeconomic distortions and unfamiliarity with market processes) and IC2 (level of socialist development and associated distortions), plus initial income.
  - Macroeconomic variables — including CPI inflation and government expenditure as a percentage of GDP.
  - Structural reform indices — EBRD reform indices.
- Other candidate explanatory variables: real GDP growth in trading partners (EU and world), CPI-based real exchange rate vis-à-vis Russia, trade-weighted real effective exchange rate, share of exports to Russia, openness (exports + imports divided by GDP).
- Related studies document shifts in trade away from intra-CIS trade toward the rest of the world and offer demand-side decompositions and growth-accounting explanations for individual country recoveries.

### III. Data — sources, patterns, and weaknesses
- Annual data for real GDP growth, CPI inflation, and government expenditures as a percentage of GDP were obtained from the IMF World Economic Outlook database.
- Figures referenced (1–3) display:
  - Real GDP: most CIS countries exhibit a “smile” pattern; Baltics show an increasing output trajectory.
  - CPI inflation: many cases show rapid disinflation in the early 1990s and low inflation thereafter.
  - Government expenditure as a share of GDP: varied movements across countries.
- Initial conditions from Havrylyshyn and van Rooden (2000); reform index from EBRD reform indices.
- Data weaknesses and consequences:
  - GDP data likely include biases from underreporting by private enterprises.
  - Government expenditure data may not be fully comparable across countries due to transparency, definitions, and coverage differences (example: sizeable and time-varying quasi-fiscal expenditures in Belarus vs. higher fiscal transparency in the Baltics).
  - Consumer price inflation data quality may vary across countries.
  - Owing to poor data quality, years prior to 1993 were excluded and Turkmenistan was not included in the analysis.

### IV. Unit root tests — outcomes and implications
- Table 3 tests the null hypothesis that the level of real GDP exhibits a unit root using DF and ADF tests including a constant and a time trend.
- Test outcomes:
  - The null hypothesis of a unit root is rejected in 7 of 13 cases at the 5 percent level based on the DF test.
  - The null hypothesis is rejected in 2 of 13 cases based on the ADF test.
  - Using the specification that minimizes the Akaike Information Criterion (AIC), the tests reject the null hypothesis in 5 of 13 cases.
  - The Im, Pesaran, and Shin (2003) t-bar test strongly rejects the null hypothesis that all cross-sectional units have a unit root based on either the DF or ADF test statistics.
- Given that levels of real GDP for most countries appear to have a unit root, regressions are estimated using real GDP growth rates.

### V. Econometric model — specification and structural break
- General model form (Equation (1)):
  - y_it = real GDP growth for country i in year t
  - y_Rt = real GDP growth for Russia in year t
  - x_it = vector of exogenous determinants of growth in country i
  - u_it = random disturbance term with u_it = μ_i + v_it
  - One-way error component regression model; if x_it includes current and one-period lags, model is a first-order autoregressive distributed lag model.
- Allowing for a possible shift after the Russian crisis (Equation (2)):
  - Full specification includes interactions with a dummy t_d equal to 0 prior to the Russian crisis and 1 thereafter to capture shifts in coefficients; uncertainty about exact timing implies searching within a neighborhood of the crisis for the most likely shift point.
- Estimation issues:
  - Inclusion of a lagged dependent variable gives rise to bias in standard estimators of fixed or random effects models.
  - Under fixed country effects, restrictions that sums of μ_i and sums of d_t μ_i equal zero must be imposed to avoid the dummy variable trap.
  - Initial conditions must be excluded from x_it under fixed effects because they are collinear with country effects (but can be included under random effects).

### VI. Model selection and sensitivity analysis — practical choices and rationale
- Starting from a general fixed effects model with first-order lagged dependent and independent variables, the full set of explanatory variables yields uninformative estimates due to close correlation among variables common to all countries in a panel and the short time series (11 observations).
- Key practical findings:
  - With first-order lags for all explanatory variables included, coefficients for most explanatory and dummy variables are insignificant.
  - If lagged values of explanatory variables are omitted, coefficients on most explanatory and dummy variables become significant.
  - The econometric analysis therefore proceeds from the general model including a lagged dependent variable but excluding all lagged independent variables.
- Alternative model evaluation and selection methods considered but deemed less suitable for this estimation problem:
  - General-to-specific approach associated with David Hendry and others.
  - Bayesian Averaging of Classical Estimates (BACE) approach developed by Sala-i-Martin, Doppenhofer, and Miller (2004).
  - Reasons for unsuitability are discussed in Box 1 (Model Evaluation and Selection Methods).

*Source: _wp05192 - References (excerpt) — IMF working paper content (pages 3–10).*

### Box 1. Model Evaluation and Selection Methods

### Box 1. Model Evaluation and Selection Methods

### General-to-specific approach
- Starts with a general model (such as equation (2)) including current and lagged values of the dependent and independent variables.
- Considers multiple paths along which combinations of explanatory variables may be excluded based on standard statistical tests.
- A priori agnostic as to which combination of explanatory variables should be included.
- Berg and others (1999) applied the methodology “loosely” by always simplifying first among time constants, then initial conditions, and finally policy variables.
- In the present study, once specification is narrowed to exclude dynamics that appear excessively demanding given the short time series, the need to narrow further is limited.
- Results for the general model (including lags for all explanatory variables) are available from the authors upon request.

### BACE approach
- Developed to sort through a large number of possible explanatory variables suggested by alternative theories of economic growth.
- Forms weighted averages of coefficient estimates based on all possible combinations of regressors.
- Agnostic as to the correct specification except for model size.
- Main issue in the present paper: insufficient time-series observations to permit estimation of dynamics.

### Empirical specification and main econometric findings
- Column (1) of Table 4: general specification excluding lagged independent variables includes lagged own-country growth, country dummies, the CPI, government expenditure in percent of GDP, the EBRD structural reform index, EU growth, the real exchange rate, Russian growth, the trade openness ratio, and interactions between a post-Russian-crisis dummy (0 prior to 1998, 1 in 1998 and onwards) and all other explanatory variables.
- Key coefficient estimates and interpretations:
  - Coefficient on Russian growth: 0.96 and significant.
  - Coefficient on Russian crisis dummy interacted with Russian growth: –0.91 and highly significant.
  - Interpretation: before the Russian crisis, a one percentage point increase in Russian growth was associated with a 0.96 percentage point increase in another country’s growth rate, holding other factors constant; after the crisis, this effect dropped to 0.05 percentage points and was not significantly different from zero.
  - Own-country effect rose in 9 of 12 cases following the crisis (reflected in many positive and significant coefficients on interactions between the Russian crisis dummy and country dummies).
  - Coefficient on CPI inflation: not significant and near zero; interaction with the Russian crisis dummy also near zero and not significant.
  - Coefficient on government expenditure (percent of GDP): insignificant and essentially near zero; its interaction with the Russian crisis dummy is –0.5 and statistically significant.
  - Growth in the EU: coefficient –2.9 and significant prior to 1998; insignificant and close to zero in the 1998–2003 period.
  - Lagged Russian growth: not significant.
- Column (2) (alternative specification including initial condition measure IC2 but not country effects):
  - Estimated coefficient on Russian growth: 0.67 and highly significant prior to 1998; falls to 0.19 and is insignificant thereafter.
  - CPI and government expenditure (percent of GDP) are statistically significant in this specification, though CPI coefficient remains near zero.
  - Note: IC1 was not significant; IC2 was significant.
- Initial conditions:
  - Estimated coefficient for initial conditions in the present study is positive and highly significant (contrasting with Havrylyshyn and others (1998) who found a negative sign).
  - Rerunning Havrylyshyn and others (1998) regressions on the present sample yields a positive impact of initial conditions.
- Regional dummies specification (column (3)):
  - Regression includes regional dummies (Baltics; the Caucasus and Moldova; and Central Asia—reference: Belarus and Ukraine) and regional dummies interacted with the Russian crisis dummy.
  - Results broadly similar to column (1).
- Sample sensitivity:
  - Results are insensitive to inclusion/exclusion of the Baltics.
  - Column (4) (excluding the Baltics): coefficient on growth in Russia is 1.37 and highly significant; coefficient on Russian crisis dummy interacted with Russian growth is –1.32 and highly significant; their difference is 0.06 and is insignificant.
  - Column (5) reports estimates using data only for the Baltics (results presented in the table).
- Real exchange rate and bilateral measures:
  - Two variants include a CPI-based bilateral real exchange rate vis-à-vis Russia either instead of, or in addition to, the multilateral real effective exchange rate index.
  - Caveats: real exchange rate measures based on official exchange rates may be problematic where multiple exchange rates operated; CPI may be an inaccurate index of price competitiveness; PPI not available for all countries and years.
  - Column (7): coefficient on Russian growth is 0.78 and highly significant prior to 1998; falls to 0.06 and becomes insignificant thereafter.
- Other specifications and robustness:
  - Column (8): specification that excludes both country fixed effects and initial conditions—key variables of interest remain broadly similar.
  - Column (9): explores possible nonlinearity in the response of growth to inflation using INF, the natural logarithm of percent changes in CPI inflation, in place of CPI; coefficient on (log) CPI inflation remains insignificant; coefficients on Russian growth similar to earlier results.
  - Column (10): substitutes world growth for EU growth; coefficient on world growth is insignificant; coefficients on Russian growth similar to earlier results.

### Structural break timing and sensitivity analysis
- Structural change likely occurred in either 1998 or 1999.
- To assess the choice of 1998 as the structural break point, estimates are presented in Table 5 for the same specifications as in Table 4 but assuming the break occurred in 1999 rather than 1998 (discussion follows in the paper).

*Source: _wp05192 - Box 1. Model Evaluation and Selection Methods*

### 1998. The fit of these regressions is (except for the Baltics-only regression) uniformly worse

### _wp05192 - 1998. The fit of these regressions is (except for the Baltics-only regression) uniformly worse

### Structural break selection and robustness tests
- Regressions using a break point of 1999 have:
  - lower values of the log-likelihood function,
  - higher values of the Akaike information criterion (AIC),
  - a smaller number of significant t-statistics,
  - and generally less precisely estimated coefficients compared with regressions using a break point of 1998.
- The coefficient on Russian growth is broadly similar in magnitude between 1998- and 1999-break specifications and is highly significant in all but one specification when using a 1999 break.
- The coefficient on the Russian crisis dummy interacted with Russian growth:
  - ranges widely across specifications when using a 1999 break,
  - is no longer significant in any specification with a 1999 break.
- Effect of Russian growth on other CIS and Baltic countries following the break point:
  - not significantly different from zero with a 1999 break in any specification,
  - magnitude of the post-crisis effect varies widely.
- Andrews’s (1993) Sup W, Sup LM, and Sup LR tests (allowing for an unknown break point and gradual coefficient changes):
  - did not reject the null hypothesis of parameter constancy for either the entire coefficient vector or for the coefficient on Russian growth,
  - owing partly to the limited number of observations, the data set provides insufficient information either to reject or to accept parameter constancy based on these tests,
  - test indeterminacy is partly due to contamination from imprecision in estimating coefficients on many control variables (tests are based on sums of squared regression residuals).
- The paper concludes that standard tests of constancy of the coefficient on Russian real GDP growth (presented elsewhere in the document) are not contaminated in the same way and are considered more reliable in this context.
- Taken together, these results provide support for choosing 1998 as the structural break point.
  - Rationale noted: expectations converged by early 1998 that macroeconomic imbalances were emerging in Russia; links may have been disrupted even before the August 17, 1998 ruble float by macroeconomic imbalances and financial sector vulnerabilities.

### Estimation methods and robustness (LSDV, Arellano–Bond, random effects)
- Inclusion of a lagged dependent variable in an error components model biases least-squares dummy variable (LSDV) estimators.
- Arellano and Bond (1991) provide consistent instrumental variables estimators for this model.
- Comparison of estimates (Table 6):
  - LSDV estimates presented in column (1).
  - Arellano–Bond one-step estimates (using one-period lags of independent variables as instruments) presented in column (11); alternative specification in column (12).
  - Coefficients on Russian growth from Arellano–Bond are broadly comparable to LSDV results, although the coefficient is significant only in one of two specifications.
  - Arellano–Bond estimates of the interaction between the crisis dummy and Russian growth are comparable and significant in both specifications reported.
- Sargan test results:
  - do not reject the null that overidentifying restrictions underlying Arellano–Bond estimation are satisfied, suggesting instruments are valid (Table 6 memorandum items).
- Arellano–Bond serial correlation tests:
  - the null hypotheses of second-order serially uncorrelated errors are not rejected, fulfilling a necessary condition for consistency of the Arellano–Bond procedure.
- Endogeneity correction:
  - Column (13) presents Arellano–Bond estimates correcting for possible endogeneity of CPI and EXP using one-period lagged values as instruments; these estimates are very similar to those assuming exogeneity.
- Cautions:
  - Arellano–Bond estimator, while consistent under stated assumptions, may exhibit large bias in finite samples and have higher standard errors than ordinary least-squares.
  - Endogeneity-corrected coefficients in Table 6 should be treated with caution and are not necessarily superior to LSDV estimates from Table 4.
- Random effects model:
  - estimates of the variance of the random country effect μi were negative,
  - as discussed in Baltagi (2001), this occurs only when the true variance is small and close to zero; generalized least squares estimator reduces to ordinary least squares in this case.

### Possible transmission mechanisms (trade, finance, remittances, transition effects)
- Candidate channels for the drop in the effect of Russian growth on other CIS and Baltic growth:
  - disruptions in trade flows,
  - disruptions in financial flows,
  - reductions in worker remittances from Russia,
  - transition-related effects (common output decline 1993–98 followed by differential rebounds).
- Trade:
  - Figure 4 and Appendix Table 2 show merchandise exports from CIS and Baltic countries to Russia (as percent of total merchandise exports) were generally on a declining trend during 1993–2003.
  - Drop in exports to Russia as share of total exports during 1997–99:
    - pronounced for the Baltics,
    - more gradual for the CIS and started prior to the crisis.
  - The share of exports to Russia in total exports was statistically insignificant in alternative growth regressions (Section VII).
  - Aggregate growth decompositions (Appendix Table 3) provide little support for net external demand as a crucial driver of real GDP growth in these countries; external demand usually was not a large factor.
  - Exports to Russia generally contributed only a small amount to real GDP growth and there is little evidence that this contribution declined following the Russian crisis.
  - Indirect trade effects (domestically produced inputs for export production) may be important and are difficult to capture.
- Capital flows:
  - Net capital inflows slowed markedly during the crisis (Figure 5).
  - Data on capital inflows from Russia are unavailable.
  - It is reasonable to suppose some additional capital inflows into Russia in the post-crisis period may have been invested in other CIS countries, which could strengthen rather than weaken growth linkages.
- Transition-related explanation:
  - In 1993–98, real GDP in CIS countries may have fallen together due to transition effects; once these effects played out, output may have rebounded at different rates, lowering correlations.

### Main empirical findings and key statistics
- Pre-crisis (before 1998) relationship:
  - Russian economic growth was a significant determinant of growth in other CIS and Baltic countries.
  - Each percentage point of additional growth in Russia led to around 0.8–0.9 percentage points of additional growth in the other countries on average, holding other factors constant.
- Post-crisis (beginning in 1998) relationship:
  - A 1 percentage point increase in Russian real GDP appears to have induced an increase in real GDP of the other CIS and Baltic countries of less than 0.2 percentage points on average.
  - The link was statistically insignificant in the period starting in 1998 (identified as the most likely break point).
- Robustness:
  - Findings are robust to choice of explanatory variables and estimation methods, though there is substantial variation across countries.
- Data and inference limitations:
  - Substantial weaknesses in the data and limited number of time-series observations warrant cautious interpretation.
  - Andrews’s tests were inconclusive due to limited information and contamination from estimation imprecision on control variables.

### Policy and research implications / recommendations
- Further investigation is needed into the links between trade and financial flows in the region, preferably at a disaggregated level, to explain the significant break in growth linkages observed in the post-1998 period.
- Consideration of indirect trade effects (value added via domestically produced inputs for export production) should be part of future work because indirect effects may exceed direct export contributions.
- Additional data collection on capital inflows from Russia and on worker remittances would help clarify financial transmission channels.
- Given estimator biases in short panels with lagged dependent variables, apply instrumental-variable methods (e.g., Arellano–Bond) with caution and complement them with alternative estimators; interpret finite-sample results carefully.

*IMF Working Paper content unit: _wp05192 - 1998. The fit of these regressions is (except for the Baltics-only regression) uniformly worse*

### Appendix Table 1a. Simple Correlation Coefficients Between Real GDP Growth Rates, 1993–97

### Appendix Table 1a. Simple Correlation Coefficients Between Real GDP Growth Rates, 1993–97

### Pairwise correlation highlights (selected entries)
- EST—EST: 1.00
- EST—LVA: 0.89
- EST—LTU: 0.96
- BLR—KAZ: 0.95
- UKR—RUS: 0.98
- KGZ—UZB: 0.99
- ARM—EU: 0.93
- MDA—ARM: -0.45
- EU—MDA: -0.46
- RUS—UKR: 0.98

*Source: IMF World Economic Outlook database; Fund staff estimates.*

---

### Appendix Table 1b. Simple Correlation Coefficients Between Real GDP Growth Rates, 1998–2003

### Pairwise correlation highlights (selected entries)
- EST—EST: 1.00
- LVA—UKR: 0.92
- LTU—BLR: 0.84
- TJK—LVA: 0.98
- KAZ—UKR: 0.93
- UZB—UZB: 1.00
- ARM—GEO: 0.82
- GEO—UZB: -0.74
- EU—ARM: -0.91

*Source: IMF World Economic Outlook database; Fund staff estimates.*

---

### Appendix Table 1c. Simple Correlation Coefficients Between Real GDP Growth Rates, 1993–98

### Pairwise correlation highlights (selected entries)
- EST—EST: 1.00
- EST—GEO: 0.95
- LTU—UZB: 0.86
- KAZ—KGZ: 0.97
- UKR—BLR: 0.91
- ARM—AZE: 0.58
- MDA—LVA: -0.04
- EU—MDA: -0.36
- RUS—KGZ: 0.91

*Source: IMF World Economic Outlook database; Fund staff estimates.*

---

### Appendix Table 1d. Simple Correlation Coefficients Between Real GDP Growth Rates, 1999–2003

### Pairwise correlation highlights (selected entries)
- EST—EST: 1.00
- LVA—UKR: 0.97
- LTU—BLR: 0.93
- BLR—LTU: 0.93
- TJK—LVA: 0.98
- MDA—LTU: 0.95
- ARM—LTU: 0.93
- UZB—UZB: 1.00
- EU—ARM: -0.92
- RUS—EU: 0.61

*Source: IMF World Economic Outlook database; Fund staff estimates.*

---

### Appendix Table 2a. Merchandise Exports to Russia as a Percent of Total Merchandise Exports, 1993–2003

### Selected country-year values (percent)
- Armenia: 1993 34.2; 1994 34.7; 1995 25.4; 1996 33.1; 1997 27.0; 1998 18.1; 1999 14.6; 2000 14.8; 2001 17.7; 2002 14.8; 2003 12.0
- Azerbaijan: 1993 22.0; 1994 21.9; 1995 18.1; 1996 17.6; 1997 23.1; 1998 17.4; 1999 8.9; 2000 5.6; 2001 3.4; 2002 4.4; 2003 4.4
- Belarus: 1993 40.7; 1994 47.1; 1995 44.4; 1996 53.2; 1997 64.2; 1998 65.2; 1999 54.5; 2000 50.6; 2001 53.1; 2002 49.6; 2003 49.1
- Estonia: 1993 22.6; 1994 23.0; 1995 17.6; 1996 16.4; 1997 18.8; 1998 13.3; 1999 9.2; 2000 6.8; 2001 8.6; 2002 10.0; 2003 11.4
- Georgia: 1993 45.5; 1994 33.6; 1995 31.0; 1996 28.5; 1997 29.8; 1998 17.4; 1999 12.5; 2000 21.0; 2001 23.4; 2002 17.4; 2003 17.7
- Kyrgyz Republic: 1993 31.5; 1994 17.2; 1995 23.7; 1996 26.5; 1997 16.2; 1998 16.5; 1999 15.6; 2000 12.9; 2001 13.6; 2002 16.5; 2003 16.7
- Latvia: 1993 28.5; 1994 28.1; 1995 24.9; 1996 23.1; 1997 21.0; 1998 12.0; 1999 6.6; 2000 4.2; 2001 5.7; 2002 5.8; 2003 5.4
- Lithuania: 1993 4.2; 1994 28.2; 1995 20.4; 1996 23.8; 1997 24.5; 1998 16.5; 1999 7.0; 2000 7.1; 2001 11.0; 2002 12.1; 2003 10.1
- Moldova: 1993 35.6; 1994 51.1; 1995 48.3; 1996 54.0; 1997 58.1; 1998 53.7; 1999 41.3; 2000 44.5; 2001 43.7; 2002 37.2; 2003 39.0
- Tajikistan: 1993 17.9; 1994 9.4; 1995 12.7; 1996 10.2; 1997 7.9; 1998 8.0; 1999 16.7; 2000 33.6; 2001 16.1; 2002 11.9; 2003 6.6
- Ukraine: 1993 …; 1994 40.3; 1995 39.8; 1996 38.7; 1997 26.2; 1998 23.0; 1999 20.7; 2000 24.1; 2001 22.6; 2002 17.8; 2003 17.8
- Uzbekistan: 1993 …; 1994 38.9; 1995 29.7; 1996 22.6; 1997 31.9; 1998 20.5; 1999 21.6; 2000 27.6; 2001 25.4; 2002 20.0; 2003 22.0
- CIS excl. Russia (aggregate): 1993 …; 1994 40.1; 1995 39.2; 1996 39.6; 1997 36.5; 1998 33.8; 1999 27.4; 2000 27.1; 2001 26.7; 2002 22.9; 2003 22.7
- Baltics (aggregate): 1993 17.6; 1994 26.6; 1995 20.5; 1996 21.4; 1997 21.8; 1998 14.4; 1999 7.8; 2000 6.4; 2001 9.1; 2002 10.1; 2003 9.7

*Source: IMF Direction of Trade database; Fund staff estimates.*

---

### Appendix Table 2b. Merchandise Imports from Russia as a Percent of Total Merchandise Imports, 1993–2003

### Selected country-year values (percent)
- Armenia: 1993 30.1; 1994 28.4; 1995 19.4; 1996 14.6; 1997 24.2; 1998 21.2; 1999 21.5; 2000 15.5; 2001 19.8; 2002 11.9; 2003 11.6
- Azerbaijan: 1993 23.1; 1994 15.1; 1995 13.2; 1996 16.5; 1997 19.1; 1998 18.0; 1999 21.9; 2000 21.3; 2001 10.7; 2002 16.9; 2003 15.5
- Belarus: 1993 46.0; 1994 62.9; 1995 56.1; 1996 50.8; 1997 53.6; 1998 54.8; 1999 56.4; 2000 64.8; 2001 65.2; 2002 65.1; 2003 65.8
- Estonia: 1993 17.2; 1994 16.6; 1995 16.1; 1996 13.4; 1997 14.4; 1998 11.1; 1999 13.5; 2000 13.4; 2001 12.5; 2002 12.0; 2003 10.2
- Georgia: 1993 4.1; 1994 7.8; 1995 12.4; 1996 18.5; 1997 13.4; 1998 9.2; 1999 7.4; 2000 13.8; 2001 12.4; 2002 15.4; 2003 14.0
- Kyrgyz Republic: 1993 35.4; 1994 21.9; 1995 26.8; 1996 21.9; 1997 26.9; 1998 24.3; 1999 17.9; 2000 23.9; 2001 18.3; 2002 19.9; 2003 24.7
- Latvia: 1993 28.1; 1994 23.5; 1995 21.6; 1996 20.3; 1997 15.6; 1998 11.8; 1999 10.5; 2000 11.6; 2001 9.2; 2002 8.8; 2003 8.7
- Lithuania: 1993 23.7; 1994 39.3; 1995 31.2; 1996 26.0; 1997 25.3; 1998 21.1; 1999 20.1; 2000 27.4; 2001 25.3; 2002 21.2; 2003 22.0
- Moldova: 1993 35.2; 1994 46.9; 1995 33.1; 1996 30.0; 1997 28.6; 1998 22.3; 1999 23.6; 2000 15.4; 2001 16.1; 2002 14.8; 2003 13.0
- Tajikistan: 1993 15.7; 1994 11.1; 1995 16.8; 1996 11.1; 1997 15.3; 1998 14.4; 1999 13.9; 2000 15.7; 2001 19.0; 2002 23.0; 2003 20.2
- Ukraine: 1993 …; 1994 54.1; 1995 37.8; 1996 50.1; 1997 45.8; 1998 48.1; 1999 47.2; 2000 41.7; 2001 36.8; 2002 37.2; 2003 35.9
- Uzbekistan: 1993 …; 1994 37.4; 1995 29.9; 1996 24.5; 1997 21.2; 1998 18.2; 1999 10.6; 2000 14.6; 2001 17.5; 2002 24.0; 2003 22.3
- CIS excl. Russia (aggregate): 1993 …; 1994 47.2; 1995 39.4; 1996 43.5; 1997 41.5; 1998 41.4; 1999 40.2; 2000 43.5; 2001 40.1; 2002 40.2; 2003 39.0
- Baltics (aggregate): 1993 23.3; 1994 28.3; 1995 24.2; 1996 20.6; 1997 19.5; 1998 15.5; 1999 15.4; 2000 18.4; 2001 17.1; 2002 15.3; 2003 14.9

*Source: IMF Direction of Trade database; Fund staff estimates.*

---

### Appendix Table 3. Contribution of External Demand and Exports to Russia to Real GDP Growth in the CIS and Baltic Countries, 1993–2003

Note: Figures in Appendix Table 3 are reported in percentage points of real GDP. The table provides annual Real GDP and decompositions into Domestic demand, Net exports, Exports (o/w Russia), and Imports (o/w Russia) for each country.

### Selected country-level time series (Real GDP, and selected components)
- Armenia — Real GDP:
  - 1993 -14.1; 1994 5.4; 1995 6.9; 1996 5.9; 1997 3.3; 1998 7.3; 1999 3.3; 2000 6.0; 2001 9.6; 2002 13.2; 2003 13.9
  - Domestic demand (selected): 1999 3.7; 2000 17.5; 2001 14.5
  - Net exports (selected): 1999 19.4; 2000 -22.2; 2001 5.9

- Belarus — Real GDP:
  - 1993 -7.6; 1994 -11.7; 1995 -10.4; 1996 2.8; 1997 11.4; 1998 8.4; 1999 3.4; 2000 5.8; 2001 4.7; 2002 5.0; 2003 6.8
  - Exports (selected): 1997 17.4; 1998 27.5; 1999 -11.5; 2000 11.4
  - Exports o/w Russia (selected): 1997 13.7; 1998 24.8; 1999 -6.7; 2000 -0.9

- Estonia — Real GDP:
  - 1993 -8.2; 1994 0.9; 1995 4.5; 1996 4.5; 1997 10.5; 1998 5.2; 1999 -0.1; 2000 7.8; 2001 6.4; 2002 7.2; 2003 5.1
  - Domestic demand (selected): 1993 -8.6; 1994 5.2; 1995 5.3; 1996 7.5; 1997 12.5
  - Net exports (selected): 1993 0.4; 1994 -4.3; 1995 -0.8; 1996 -3.0; 1997 -2.0
  - Exports o/w Russia (selected): 1994 3.0; 1995 -2.6; 1996 -0.4; 1997 4.6; 1998 -2.7

- Kazakhstan — Real GDP:
  - 1993 -9.2; 1994 -12.6; 1995 -8.3; 1996 0.5; 1997 1.6; 1998 -1.9; 1999 2.7; 2000 9.8; 2001 13.5; 2002 9.8; 2003 9.2
  - Domestic demand (selected): 1999 21.5; 2000 7.7; 2001 4.7
  - Net exports (selected): 1999 19.5; 2000 10.7; 2001 -2.9

- Kyrgyz Republic — Real GDP:
  - 1993 -13.0; 1994 -19.8; 1995 -5.8; 1996 7.1; 1997 9.9; 1998 2.1; 1999 3.7; 2000 5.4; 2001 5.3; 2002 0.0; 2003 6.9
  - Domestic demand (selected): 1994 -22.6; 1995 -7.3; 1996 20.2
  - Exports o/w Russia (selected): 1994 -5.9; 1995 0.8; 1996 2.1

- Latvia — Real GDP:
  - 1993 -11.4; 1994 2.2; 1995 -0.9; 1996 3.8; 1997 8.3; 1998 4.7; 1999 3.3; 2000 6.9; 2001 8.0; 2002 6.4; 2003 7.5
  - Domestic demand (selected): 1993 -13.9; 1994 6.6; 1995 4.0; 1996 7.8; 1997 6.1
  - Net exports (selected): 1993 2.5; 1994 -4.4; 1995 -4.9

- Lithuania — Real GDP:
  - 1993 -16.2; 1994 -9.8; 1995 3.3; 1996 4.7; 1997 7.0; 1998 7.3; 1999 -1.7; 2000 3.9; 2001 6.4; 2002 6.8; 2003 9.7
  - Domestic demand (selected): 1997 10.6; 1998 8.5; 1999 -0.3; 2000 2.2
  - Exports o/w Russia (selected): 1997 2.4; 1998 -3.7; 1999 -5.4; 2000 0.3

- Moldova — Real GDP:
  - 1993 -1.2; 1994 -30.9; 1995 -15.3; 1996 -5.9; 1997 1.6; 1998 -6.5; 1999 -3.4; 2000 2.1; 2001 6.1; 2002 7.8; 2003 6.3
  - Domestic demand (selected): 1993 23.5; 1994 -38.4; 1995 -9.5; 1996 8.8
  - Exports (selected): 1993 -94.0; 1994 -5.1; 1995 12.3; 1996 -5.1
  - Exports o/w Russia (selected): 1994 2.3; 1995 4.8; 1996 0.7; 1997 3.0

- Russia — Real GDP:
  - 1993 -8.7; 1994 -12.7; 1995 -4.1; 1996 -3.6; 1997 1.4; 1998 -5.3; 1999 6.3; 2000 10.0; 2001 5.1; 2002 4.7; 2003 7.3
  - Domestic demand (selected): 1993 -10.1; 1994 -12.2; 1995 -4.2; 1996 -4.6
  - Net exports (selected): 1993 1.4; 1994 -0.5; 1995 0.1; 1996 1.0
  - Exports (selected): 1993 -0.6; 1994 1.4; 1995 1.9; 1996 1.4
  - Imports (selected): 1993 1.9; 1994 -1.9; 1995 -1.8; 1996 -0.4

- Ukraine — Real GDP:
  - 1993 -14.2; 1994 -22.9; 1995 -12.2; 1996 -10.0; 1997 -3.0; 1998 -1.9; 1999 -0.2; 2000 5.9; 2001 9.2; 2002 5.3; 2003 9.4
  - Domestic demand (selected): 1994 -23.7; 1995 -20.7; 1996 -9.2; 1997 -3.4
  - Net exports (selected): 1994 0.8; 1995 8.6; 1996 -0.9; 1997 0.4
  - Exports o/w Russia (selected): 1999 0.5; 2000 1.0; 2001 -6.2; 2002 -2.8

### Averages (table footer)
- Average (selected): 3.3; -0.3; 1.0; 0.5

*Source: IMF Direction of Trade and World Economic Outlook databases; Fund staff estimates.*

---

*Italic: IMF Direction of Trade and World Economic Outlook databases; Fund staff estimates.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp05192.pdf_