## _wp09284

## Source details

**Canonical URL:** [_wp09284](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09284.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09284.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09284.pdf.json)

---

### I. Introduction and research questions
- Objective: test joint growth effects of de jure reforms in domestic finance, trade, and the capital account within a unified empirical framework, using a newly developed database of de jure indicators.
- Major research questions:
  - What do growth regressions indicate on the joint effects of these reforms on economic performance?
  - Are the effects of reforms persistent?
  - Is there heterogeneity of reform effects across countries, and are growth effects constrained by political institutions?
  - Through which channels do reforms affect economic performance — aggregate investment or aggregate economic efficiency (TFP)?

### II. Data, measurement, and empirical model
- Data and indices:
  - Domestic financial reform index (Abiad et al. (2008)): aggregate is average of six sub-indices (five banking-related and one on bond/equity market development), sub-indices coded 0–3 and normalized to unit interval.
  - Capital account index (main regressions): subcomponent from Abiad et al. (2008), coded 0–3 and renormalized to unit interval.
  - Trade liberalization: primary measure is continuous average tariff index (Ostry et al. (2009)) normalized to unit interval; alternative Wacziarg and Welch (2008) binary index used in robustness checks.
- Sample and coverage:
  - Country sample: 90 countries covering low, middle, and high income groups.
  - Time period: 1974-2004.
  - Baseline regression sample observations reported up to 792 depending on table/specification.
- Empirical specifications:
  - Standard panel growth regression: Δy_it = α·y_{i,t-1} + β·Liberalization_{i,t-1} + θ·X_{it-1} + f_i + d_t + ε_it (variables averaged over 3-year periods).
  - Growth accounting extensions: dependent variables Inv/GDP (gross capital formation percent of GDP) and SR (Solow residual) to separate capital accumulation and TFP channels.
  - Dynamic reform episode definition: Reform if ΔI_it > 0; Large_Reform if ΔI_it > Z_I (one standard deviation); dynamic specification includes leads and lags up to 6 years, and reversal indicators.

### III. Key empirical findings (high-level)
- Relative importance of reform types:
  - Domestic financial reforms, and to some extent trade reforms, are most robustly associated with economic growth.
  - No systematic positive relationship between capital account liberalization and economic growth; reversals of capital account liberalization are associated with slower growth.
- Heterogeneity by income group:
  - Positive effects of financial and trade reforms are almost exclusively driven by the group of middle-income countries.
  - Effect of domestic financial reforms on growth in middle-income countries is predominantly explained by improvements in measured aggregate TFP growth rather than by higher aggregate investment.
- Persistence:
  - Growth effects of domestic financial and trade reforms are somewhat persistent, raising growth at a horizon of up to 6 years, but become insignificant at a longer horizon (no persistent effect at 6–9 years).
- Institutions and complementarities:
  - Tentative evidence of complementarities: financial and trade reforms are effective in developing countries with good protection of property rights; sufficiently developed property rights may be a precondition for reaping growth benefits.
- Robustness:
  - Results are robust to a variety of robustness tests and to endogeneity bias (including GMM and IV specifications).

### IV. Quantitative highlights and exact estimates
- Summary statistics (Table 2 — exact values):
  - Real per capita GDP growth: Obs 5900; Mean 0.02; Std. Dev. 0.03; Min -0.13; Max 0.17.
  - TFP growth: Obs 5340; Mean 0.02; Std. Dev. 0.03; Min -0.10; Max 0.10.
  - Investment / GDP: Obs 5790; Mean 0.17; Std. Dev. 0.08; Min 0.02; Max 0.46.
  - Trade liberalization Index: Obs 5840; Mean 0.72; Std. Dev. 0.22; Min 0; Max 1.
  - Domestic Financial Lib Index: Obs 5900; Mean 0.51; Std. Dev. 0.29; Min 0; Max 1.
  - Capital Account Lib Index: Obs 5900; Mean 0.59; Std. Dev. 0.37; Min 0; Max 1.
  - Inflation: Obs 5894; Mean 1%; Std. Dev. 282%; Min -10%; Max 4770%.
- Bivariate correlations (Table 3 — selected exact entries):
  - Correlation of Average annual growth with Inflation: -0.2306.
  - Correlation of Average annual growth with Domestic finance: 0.1484.
  - Correlation of Average annual growth with Capital account: 0.1065.
  - Correlation of Domestic finance with Trade (average tariff): 0.6014.
  - Correlation of Capital account with Domestic finance: 0.7097.
- Standard growth regressions (Table 4 — selected exact coefficients):
  - All countries (Specification (1), Dependent: GDP per capita growth):
    - Domestic finance: 0.0291* [0.0147].
    - Log(initial real GDP per capita): -0.0246*** [0.0064].
    - Initial inflation: -0.0004*** [0.0001].
    - Observations: 792; R-squared: 0.40.
  - Middle income (Specification (3)):
    - Domestic finance: 0.0529** [0.0243].
    - Log(initial real GDP per capita): -0.0275*** [0.0100].
    - Initial inflation: -0.0003*** [0.0001].
    - Observations: 359; R-squared: 0.48.
  - Low income (Specification (4)):
    - Domestic finance: 0.0535* [0.0295].
    - Capital account: -0.0242** [0.0112].
    - Initial inflation: -0.0237*** [0.0051].
    - Observations: 156; R-squared: 0.40.
- Channels (Table 5 — selected exact entries, Middle income group):
  - TFP growth regression:
    - Domestic finance: 0.0489** (0.0236).
    - Domestic finance * socialist legal origin: 0.3723*** (0.0510).
    - Observations: 280; R-squared (TFP growth): 0.35.
  - Investment / GDP regression:
    - Domestic finance: 0.0560** (0.0259).
    - Observations: 317; R-squared (Investment / GDP): 0.83.
- System GMM (Table 6 — selected exact entries):
  - Real per capita GDP growth (column 1): Domestic finance coefficient 0.0566** (0.0288); Observations: 359; Number of countries: 44.
  - Sargan test (p-value) reported as 0.11 for column 1.
- Dynamic effects (Table 7 — selected exact findings):
  - Frequencies of reform episodes:
    - Domestic finance: 8.3 percent.
    - Trade: 7.8 percent.
    - Capital account: 16.2 percent.
    - Large domestic finance reforms: 1.7 percent.
    - Large trade reforms: 4.5 percent.
  - All countries (example estimates):
    - Trade contemporaneous reform: 0.0081** [0.0034]; previous period: 0.0101*** [0.0034].
    - Capital account contemporaneous reversal: -0.0173** [0.0067].
  - Middle-income countries:
    - Episodes of domestic financial reforms increase economic growth by 1.2 to 1.4 percentage points on average in the 3 to 6 years following a reform episode.
    - Large domestic financial reforms increase economic growth by an additional 1.2 percentage points.
    - Trade reforms are positively and significantly associated with faster economic growth with persistence at a six-year horizon (example contemporaneous effect on GDP per capita growth: 0.0165*** [0.0059]).
- Instrumental variables and endogeneity (Tables 11a/11b — selected exact entries):
  - IV strategy: use reform levels and changes of political allies (Entente Alliances index) with up to 3 lags as instruments; for domestic finance use all 6 components as instruments.
  - 2SLS second-stage (example): Domestic financial reform (Real per capita GDP growth, second stage) contemporaneous reform 0.0179** [0.0083].
  - Hansen J test (second-stage): p-values not rejected at 22 percent, 15 percent and 15 percent levels for real GDP growth, TFP growth and investment equations respectively.
  - First-stage partial R2 and F tests (examples): Domestic financial reform contemporaneous reform Partial R2 0.40; F test 8.0. Trade reform contemporaneous reform Partial R2 0.34; F test 11.3.
  - Trade reform coefficients become insignificant in 2SLS; domestic finance coefficients remain significant for real GDP and TFP growth.
- Sequencing and political institutions (Table 12 — selected exact findings):
  - Sample split by median constraint on the executive:
    - Middle-income countries account for 80 percent of countries above the median; 53 percent below the median.
  - In countries with good property rights:
    - Domestic financial reforms are associated with faster economic growth at a 6 years horizon.
  - In countries with bad property rights:
    - No significant association for domestic financial reforms.
  - Example coefficients (Property rights control column, middle-income samples):
    - Domestic financial reform (contemporaneous): 0.0121** [0.0061]; Reform (previous period): 0.0137** [0.0059].
    - Trade reform (contemporaneous): 0.0161*** [0.0060].
    - Capital account contemporaneous reversal: -0.0299*** [0.0104].

### V. Channels, persistence, and interpretation
- Channels:
  - Aggregate TFP growth is identified as the most robust transmission mechanism for the association between domestic financial reforms (and trade reforms) and higher economic growth.
  - The effect of domestic financial reforms remains significant when controlling for private credit to GDP, suggesting improved allocation of credit rather than just increased aggregate credit volume.
- Persistence:
  - Positive growth effects last up to 6 years; adding reform dummies for reforms 6 to 9 years before the observation finds no persistent direct effect at that horizon.
  - Reforms can have lasting effects on the level of output on the order of 10 percent for a full liberalization after 6 years (interpretation provided in text).
- Sequencing and complementarities:
  - Evidence suggests complementarity between political institutions (property rights) and economic reforms: reforms are effective where property rights are stronger.
  - Trade reforms sometimes precede domestic financial reforms; sequencing matters in some theoretical and empirical accounts.

### VI. Robustness, endogeneity checks, and specification notes
- Robustness:
  - Results robust to adding controls one by one: Polity IV “constraint on the executive”, fiscal balance, real exchange rate overvaluation, private credit/GDP, school enrolment, and alternative trade indices.
- Endogeneity:
  - System-GMM (Windmeijer-corrected two-step) and IV (political allies instrument) used to address endogeneity and reverse causality.
  - Evidence of reverse causality/anticipation found for trade reforms (lead of large trade reforms significant), but not for domestic financial reforms.
  - IV/2SLS results: domestic financial reform effects persist for real GDP and TFP growth; trade reform effects become insignificant under IV.

### VII. Main conclusions and open questions
- Main conclusions:
  - Domestic financial reforms are robustly associated with positive real GDP growth and TFP growth for up to 6 years after implementation.
  - Trade reforms show positive associations in OLS but lose significance in IV, indicating possible endogeneity in OLS estimates.
  - Capital account liberalization is not associated with faster growth; reversals of capital account liberalization are associated with slower growth.
  - Growth benefits of domestic financial and trade reforms are concentrated in middle-income countries and are stronger where property rights are well developed.
- Open questions and suggestions for future research:
  - Why do different reforms matter across income groups, and why are middle-income countries the main drivers of estimated effects?
  - How do alternative sequences among economic reforms affect outcomes, including sequencing between economic and political reforms?
  - Further investigation of reform sequencing and institutional prerequisites is warranted.

*Source: _wp09284 - References .............................................................................................................*

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

### _wp09284 - References .............................................................................................................

### I. Introduction
- Over recent decades, the global economy has become increasingly liberalized across multiple dimensions: domestic financial systems have deepened and become less regulated, trade regimes have become more open, and capital accounts have become less restricted.
- Existing empirical literature often:
  - Focused on individual reforms only.
  - Covered different country samples and measures of reforms, limiting comparability.
  - Relied on de facto measures (e.g., financial development, openness to trade) that are outcome measures influenced by a variety of macroeconomic conditions beyond the reform considered.
- This paper’s main contribution:
  - Tests joint growth effects of various types of de jure reforms within a unified empirical framework.
  - Simultaneously considers domestic financial reforms, trade reforms, and capital account reforms.
  - Uses a newly developed database of de jure indicators of structural reform in finance, trade and the capital account.

### Major research questions addressed
- What do growth regressions tell us on the joint effects of these reforms on economic performance?
- Are the effects of reforms persistent?
- Is there heterogeneity of the effects of reforms across countries (and are growth effects constrained by political institutions)?
- What are the main channels through which reforms affect economic performance—do reforms affect economic performance by increasing aggregate investment or by improving aggregate economic efficiency?

### Data and measurement emphasis
- The analysis relies exclusively on de jure indicators (legal and policy restrictions) rather than de facto measures.
- Advantages of the new database:
  - More refined measures than some previous crude, binary indicators, especially for domestic finance and capital account.
  - Wide country and year coverage across all three indicators, allowing better comparability.
  - De jure indicators are emphasized as the more appropriate measures from a policy perspective because they are under policymakers’ direct control.

### Key findings
- Among the three groups of reforms:
  - Domestic financial reforms, and to some extent trade reforms, are most robustly associated with economic growth.
  - No systematic positive relationship is found between capital account liberalization and economic growth.
- Heterogeneity:
  - Positive effects of financial and trade reforms are almost exclusively driven by the group of middle-income countries.
  - The effect of domestic financial reforms on economic growth in these middle-income countries is predominantly explained by improvements in measured aggregate TFP growth, not by higher aggregate investment.
- Persistence:
  - Growth effects of domestic financial and trade reforms are somewhat persistent, raising growth at a horizon of up to 6 years, but become insignificant at a longer horizon.

### Institutions and complementarities
- Tentative evidence indicates interactions between political institutions (quality of property rights) and economic reforms:
  - The heterogeneity of the effectiveness of financial and trade reforms in developing countries seems to be explained by a complementarity between economic reforms and political reforms: financial and trade reforms are effective in developing countries with good protection of property rights.
  - This suggests that sufficiently developed property rights may be a precondition for reaping the growth benefits of reforms.

### Robustness and endogeneity
- Results are robust to a variety of robustness tests.
- Results are robust, in particular, to endogeneity bias.

### Paper structure (overview)
- Section II: overview of the related literature.
- Section III: the empirical model.
- Section IV: the data.
- Section V: the main results.
- Section VI: robustness tests and endogeneity issues.

*Source: _wp09284 - References .............................................................................................................*

### section VII examines the interactions of economic and political reforms; and section VIII

### _wp09284 - section VII examines the interactions of economic and political reforms; and section VIII concludes.

### II. RELATED LITERATURE
- Focus: effects of economic reforms—domestic financial development, trade openness and capital account openness—on economic growth, with emphasis on interactions and sequencing of reforms.
- Financial development literature:
  - Levine (1997, 2005) surveys growth benefits of domestic financial reform.
  - Empirical measures: de facto measures such as the ratio of private domestic credit, M2, or liquid liabilities to GDP.
  - Evidence: higher initial financial development associated with higher subsequent rates of economic growth and economic efficiency improvements (Beck et al., 2000a and 2000b, Aghion et al., 2005).
  - De jure measures: Bekaert et al. (2005) and Henry (2005) find stock market liberalizations positively correlated with growth and investment.
  - Rajan and Zingales (1998), Galindo et al. (2005), Abiad et al. (2008), and Tressel (2008) suggest more developed financial systems improve allocation of capital across firms and industries.
- Financial liberalization → financial development:
  - MacKinnon (1973) and Shaw (1973): financial repression prevents efficient allocation; liberalization should boost financial development and growth.
  - Chinn and Ito (2006) and Tressel and Detragiache (2008): financial deregulation leads to deepening, but sufficiently developed institutions are pre-conditions.
- Trade liberalization:
  - Consensus emerging that reducing trade barriers supports growth (Sachs and Warner (1995), Dollar and Kraay (2004), Wacziarg and Welch (2008)).
  - Divergent views: Rodríguez and Rodrik (2000); Subramanian, Trebbi and Rodrik (2004).
- Capital account liberalization:
  - Less empirical clarity; theoretical case for positive link (Kose et al., 2009; Dell’Ariccia et al., 2008).
  - Equity market liberalizations often show stronger effects (Henry 2000a, 2000b; Bekaert et al., 2003).
  - Kose et al. (2009) and Bonfiglioli (2008): financial integration affects growth mainly through TFP growth, little effect on capital accumulation; positive effects conditional on institutional thresholds.
  - Klein and Olivei (2008): developed countries with more open capital accounts experienced greater financial deepening and higher long-run growth; not found among developing countries.
  - Quinn (1997): improved measurement of capital account restrictiveness can reveal stronger relationships with growth.
  - Rodrik and Subramanian (2009) and Prasad et al. (2007): real exchange rate appreciation from capital inflows may hurt investment opportunities in tradables, obscuring growth benefits.
- Interactions and sequencing literature:
  - Acemoglu and Johnson (2005): property rights quality may constrain reform effects.
  - Rajan and Zingales (2003): interest group theory—incumbents oppose financial development; trade and capital openness can weaken incumbents’ opposition, explaining sequencing where trade reform often precedes domestic financial reform.
  - Hauner and Prati (2008): trade is a leading indicator of domestic finance and capital account reforms; no clear sequencing between domestic finance and capital account reforms.
  - McKinnon (1993): trade and domestic financial reforms should precede capital account liberalization to avoid exacerbating distortions or destabilizing regulated markets.
  - Dell’Ariccia et al. (2008): capital account liberalizations in countries with insufficiently developed financial markets can increase volatility.
  - Braun and Raddatz (2007): domestic financial development has smaller growth effects in countries open to trade and capital flows; explained by reduced importance of domestic finance for tradable sectors in more open countries.
  - Giavazzi and Tabellini (2005): find better growth where economic liberalization preceded democratization (contrasts with findings in this paper).
- Notes on literature coverage and exclusions:
  - Selective review; excludes broader OECD product and labor market reform sequencing literature and macroeconomic stabilization vs structural reform sequencing analyses cited elsewhere.

### III. THE EMPIRICAL MODEL
- Standard panel growth regression specification:
  - Equation (1): Δy_it = α·y_{i,t-1} + β·Liberalization_{i,t-1} + θ·X_{it-1} + f_i + d_t + ε_it (variables averaged over 3-year periods).
  - Interpretation: country fixed effects test whether reform periods associate with acceleration of growth relative to country long-run average; specification (1) does not distinguish whether liberalization affects steady-state level or steady-state growth rate of output per capita.
- Growth accounting extensions:
  - Equation (2) covers investment and TFP channels:
    - Dependent variables: Inv/GDP (gross capital formation in percent of GDP) and SR (Solow residual).
    - Purpose: separate effects on capital accumulation and total factor productivity.
- Dynamic effects and reform episodes:
  - Reform episodes definition: ReformI_it ⇔ ΔI_it > 0 (equation (3)).
  - Large reform episodes: Large_Reform_it ⇔ ΔI_it > Z_I (one standard deviation) (equation (4)).
  - Dynamic specification (5) estimates the temporal profile of reform effects including leads and lags, controls for anticipatory effects, and allows assessment of persistence and symmetry (reversals included).
  - Model includes leads 1it Reform_{+} and 1it Large_Reform_{+} to detect whether reforms follow good performance.

### IV. DATA DESCRIPTION
- Data sources and indices:
  - Domestic financial reform index: Abiad et al. (2008); aggregate index is average of six sub-indices (five banking-related: interest rate controls; credit controls; competition restrictions; degree of state ownership; quality of banking supervision and regulation; sixth sub-index covers development of bond and equity markets). Sub-indices coded 0 (fully repressed) to 3 (fully liberalized); aggregate normalized to unit interval.
  - Capital account index (main regressions): subcomponent from Abiad et al. (2008) covering restrictions on financial credits and personal capital transactions of residents and nonresidents and multiple exchange rates; original coding 0 to 3, renormalized to unit interval.
  - Alternative capital account indices considered but not used in main results: Quinn (1997) index (no significant correlation with growth in most cases) and Schindler (2009) index (short coverage 1995-2005).
  - Trade liberalization indicators:
    - Primary: continuous average tariff index from Ostry et al. (2009), normalized to unit interval (value 0 if average tariff rates are 60 percent or higher; value 1 if tariff rates are zero; linear in between).
    - Alternative: Wacziarg and Welch (2008) binary economic liberalization index based on Sachs and Warner (1995) criteria.
- Construction and coding details:
  - Financial sub-indices aggregated with equal weights; each sub-index coded 0–3 then normalized to unit interval.
  - Trade tariff index constructed using IMF, World Bank, WTO, UN, and Clemens and Williamson (2004) data with extrapolation for missing values via implicit weighted tariff rates.

### IV.B DESCRIPTIVE STATISTICS
- Baseline regression sample:
  - Covers 90 countries over the period 1974-2004.
  - Sample ranges from low-income to high-income countries and includes most emerging markets.
- Summary statistics:
  - Real per capita GDP growth averaged 2 percent annually over the sample period.
  - Investment to GDP averaged 17 percent over the period considered.
- Correlations (Table 3 highlights):
  - Real per capita growth is significantly and positively correlated with:
    - Domestic financial reform index,
    - Wacziarg and Welch index of trade and economic reforms,
    - Indicator of liberalization of cross-border financial credit (capital account index used here).
  - Correlation with Quinn (1997) broader capital account measure is not significant.
  - Per-capita real GDP growth also positively and significantly correlated with macroeconomic indicators (e.g., low inflation or the fiscal balance).
  - No significant correlation with private credit to GDP ratio or with indicators of institutional development (constraint on the executive index or ICRG property rights index).
- Event-study exploration:
  - Defined significant reform episodes as changes in index corresponding to top quartile of distribution of changes.
  - Median per capita real GDP growth plotted around reform events (Figure 1):
    - Domestic financial system reforms: associated with acceleration of economic growth in the five years following the reform.
    - Trade or capital account reforms: no clear changes in economic growth in the years following reforms.

### V. MAIN RESULTS
- Estimation approach:
  - Proceed from descriptive patterns to rigorous empirical specifications described in Sections III and IV.
  - Table 4 reports estimation results for the standard growth regression specification (details of estimates not provided in the excerpt).
- Key empirical themes investigated:
  - Whether reforms in different areas interact and whether complementarities or sequencing matter for growth.
  - Channels: decomposition into capital accumulation (Inv/GDP) and TFP (Solow residual SR) to identify mechanisms.
  - Dynamics and persistence: analysis of leads, lags, large reform episodes, and reversals to detect timing and permanence of effects.
- Preview of contrasts with related work:
  - Paper notes divergence from Giavazzi and Tabellini (2005) who found better growth where economic liberalization preceded democratization; current paper’s findings differ (details in later sections).
  - Authors flag that some existing findings (e.g., Kose et al. (2009) institutional threshold results) are not confirmed in their analysis (footnote: "As discussed below, we fail to confirm this finding.").

*Source: _wp09284 - section VII examines the interactions of economic and political reforms; and section VIII concludes.*

### section III, in which all three liberalization indices are included simultaneously. The first column

### _wp09284 - section III, in which all three liberalization indices are included simultaneously. The first column

### Estimation with all three liberalization indices (cross-section and by income group)
- The first column reports estimation on the complete sample of countries; columns 2-4 report estimations by income groups.
- Control variables in all regressions include:
  - the log of initial real per capita GDP,
  - the initial inflation rate,
  - interactions of each liberalization index with a dummy for transition countries.
- Main findings:
  - Among the three dimensions of liberalization (domestic financial liberalization, capital account liberalization, trade liberalization), only domestic financial liberalization is significantly and positively associated with per capita GDP growth.
  - Capital account liberalization and trade liberalization are not significantly associated with growth in the full sample.
  - The positive association between domestic financial liberalization and economic growth:
    - is strongly significant only for middle-income countries (column 3),
    - is even stronger for transition countries (within the middle-income group),
    - in high income countries (column 2) is driven entirely by two transition countries (the Czech Republic and Estonia),
    - in low income countries (column 4) is weakly statistically significant.
  - Trade reforms seem positively associated with economic growth in high income countries.
  - The coefficient on capital account liberalization is either not significant, or negatively associated with economic growth.
- Footnote details preserved:
  - The group of low-income transition countries—for which the coefficient on capital account liberalization is positive and significant—includes only 3 countries (the Kyrgyz Republic, Uzbekistan and Vietnam).

### Channels: TFP growth and capital accumulation (Table 5)
- Method:
  - Compute the Solow residual from a growth accounting exercise as a proxy for growth of aggregate total factor productivity (TFP).
  - Use the ratio of gross capital formation to GDP as a proxy for capital accumulation.
- Findings:
  - Domestic financial liberalization is positively and significantly associated both with TFP growth and aggregate investment in middle-income countries.
  - Domestic financial liberalization is positively associated with TFP growth in low-income countries.
  - Implication: In developing economies, structural reforms appear associated with both faster TFP growth and higher capital accumulation.
- Methodological note:
  - Growth accounting follows the approach in Bosworth and Collins (2003) by updating their data set based on the latest Penn World Tables.

### Robustness to endogeneity: GMM (Table 6)
- Estimation approach:
  - Use General Method of Moment regressions (two-step system-GMM estimator corrected for heteroscedasticity and small sample bias (Windmeijer, 2004)), following Arellano and Bond (1991) and Blundell and Bond (1998).
  - Identification assumption: lags of first differences of RHS variables are valid instruments for levels; lags of levels are valid instruments for first differences.
  - Instruments used:
    - In real GDP growth and TFP growth regressions: first lag of the first difference of RHS variables and third and fourth lags of levels of RHS variables.
    - In the investment regression: second lag of the level, and third and fourth lags of the first difference as instruments.
  - Report first-step Sargan test of overidentifying restrictions, and autocorrelation tests of order 1, 2 and 3.
- Findings:
  - The positive association between domestic financial reforms and economic performance remains significant in the real GDP per capita growth regression and in the TFP growth regression.
  - Significance falls below 10 percent in the investment regression.

### Dynamic effects of reform episodes (equation (6) specification and Tables 7–8)
- Specification features:
  - Consider reform episodes occurring during the period and in the previous period, assessing effects over a maximum time span of 6 years.
  - Include indicators for reform reversals.
  - Robustness tests include reform episodes two periods earlier (testing effects 6 to 9 years after occurrence).
  - Include leads of reform episodes to capture anticipatory/reverse-causality effects.
  - Additional controls: initial level of real GDP per capita, inflation, dummies for banking crises, conflicts, and episodes of hyperinflation.
  - Reform episode indicators are not correlated because reform episodes rarely occur simultaneously during a three-year period.
- Frequency and correlation notes:
  - Correlations between domestic financial, trade and capital account reform episodes are between 0.05 and 0.13.
  - Frequencies of reform episodes:
    - domestic finance: 8.3 percent,
    - trade: 7.8 percent,
    - capital account: 16.2 percent.
  - Frequencies of large reform episodes:
    - large domestic finance reforms: 1.7 percent,
    - large trade reforms: 4.5 percent.
- Estimation results (Table 7):
  - In the complete sample:
    - Neither domestic finance nor capital account reforms are significantly associated with faster economic growth.
    - Trade reforms are associated with higher economic growth both contemporaneously and in the following period.
    - The coefficient on the lead of large trade reforms is significant and positive, suggesting reverse causality or anticipation effects.
    - Reversals of capital account liberalization are significantly and negatively associated with contemporaneous economic growth; capital account reforms are not symmetrically related to growth improvements.
  - Subsample: middle-income countries (column 3)
    - Significant associations are driven by middle-income countries.
    - Trade reforms are positively and significantly associated with faster economic growth, with persistence at a six-year horizon.
    - Large trade reforms are not associated with higher economic growth than small trade reforms (consistent with possible reverse causality).
    - Episodes of domestic financial reforms are significantly associated with faster economic growth:
      - Economic growth increases by 1.2 to 1.4 percentage points on average in the 3 to 6 years following a reform episode.
      - Episodes of large domestic financial reforms increase economic growth by an additional 1.2 percentage points.
- Channels (Table 8):
  - In middle-income countries, the effect of domestic financial reforms on economic growth seems driven mostly by better use of factors of production (TFP) rather than sustained higher aggregate investment, which increases only temporarily and less significantly.
  - Domestic finance reversals appear preceded by higher aggregate investment in middle-income countries (consistent with anticipation effects or booms preceding reversals).
  - Trade reforms are preceded by faster TFP growth and aggregate investment in middle-income countries, and are associated with faster TFP growth; large trade reforms are associated with higher aggregate investment.
  - No clear pattern between reform episodes and performance in high income and low income countries.

### Robustness tests and endogeneity (Section VI)
- Strategy:
  - Conduct robustness tests by adding additional control variables one by one.
  - Develop an instrumental variable strategy to further address endogeneity concerns.
- Example robustness checks (Table 9, middle-income countries):
  - Add an indicator of quality of property rights: the Polity IV “constraint on the executive” index (column 1), in addition to basic controls.
  - Add additional measures of macroeconomic management (in addition to inflation) in columns 2 and 3: the fiscal balance, and a measure of real exchange rate overvaluation.

*Source: _wp09284 - section III, in which all three liberalization indices are included simultaneously. The first column*

### conclusions remain unchanged.

### _wp09284 - conclusions remain unchanged.

### Main empirical findings
- Domestic financial reforms are robustly associated with positive real GDP growth and TFP growth for up to 6 years after implementation.
- Trade reforms are associated with similar growth benefits in OLS regressions, but this association weakens and becomes insignificant in 2SLS (IV) regressions, suggesting possible endogeneity bias in the OLS results for trade reforms.
- Capital account reforms are not associated with faster growth; reversals of capital account reforms are associated with slower growth.
- Improvements in aggregate TFP growth — rather than higher aggregate investment rates — are identified as the most robust channel through which domestic financial and trade reforms are associated with higher economic growth.
- The growth benefits of domestic financial and trade reforms are concentrated in middle-income countries.

### Channels and mechanisms
- Controlling for the ratio of private credit to GDP (Levine, 1997, 2005) shows that the effect of domestic financial reforms on economic growth remains significant, implying an effect independent of the aggregate volume of bank-intermediated credit and consistent with improved allocation of credit.
- Controlling for the quality of education (proxied by school enrolment rates) and using a broader measure of trade and economic liberalization (Sachs and Warner; extended by Wacziarg and Welch, 2009) only slightly reduces the significance of domestic financial reform coefficients.
- Channel regressions (Table 10) confirm aggregate TFP growth — rather than capital accumulation — as the most robust transmission mechanism for the association between reforms and higher growth.

### Persistence of effects
- Adding reform episode dummies for reforms undertaken 6 to 9 years before the observation period (column 7) finds no persistent effect at this horizon, suggesting:
  - Domestic financial reforms may not have direct long-run effects on growth beyond 6 years.
  - Reforms can have lasting effects on the level of output on the order of 10 percent for a full liberalization after 6 years.
- The absence of persistent direct long-run effects does not preclude indirect long-run effects via deepening of domestic financial systems (Tressel and Detragiache, 2008).

### Endogeneity and instrumental variable strategy
- GMM regressions and the inclusion of leads of reform episode dummies address some endogeneity and reverse causality concerns; evidence of reverse causality is found for trade reforms but not for domestic financial reforms.
- IV hypothesis: reforms diffuse across countries with close political ties via learning or imitation. Instrument: the reform process of political allies weighted by an “Entente Alliances” index that takes the value of 0 or 1 when two countries are common members of, or signatories to, an entente or alliance in a given period.
- Instrument construction details:
  - For each sector S (domestic finance, capital account, trade), instruments include levels I_pol^{S}_{u,t} and changes ΔI_pol^{S}_{u,t} of the index for political allies for the same period t and previous periods, with a maximum of 3 lags (u = 0,1,2,3).
  - For the domestic financial reform index, all 6 components of the index are used instead of the aggregate index.
- Exclusion restrictions: levels and changes of the liberalization index of political allies are assumed uncorrelated with unobserved factors affecting economic performance, conditional on the country reform process and control variables.
- 2SLS (Table 11a) results:
  - Coefficients on domestic financial reform variables are significant for real GDP growth and TFP growth for contemporaneous and previous period reform episodes.
  - Stronger effects on TFP growth are evident for large reforms in the previous period.
  - Trade reform coefficients become insignificant in 2SLS; capital account coefficients remain insignificant.
- Specification tests:
  - Hansen J test: null hypothesis of joint validity of instruments is not rejected at the 22 percent level, 15 percent level and 15 percent level respectively in the real per capita GDP growth, TFP growth and investment equations.
  - First-stage F tests (Table 11b) indicate instruments are, in general, reasonably strong.
  - Stock and Yogo (2005) threshold recommendation noted: about 10 for first-stage F test to avoid weak instrument bias.

### Reform sequencing and political institutions
- The effectiveness of economic reforms may depend on the institutional environment, particularly political institutions and protection of property rights.
- Sample split by median value of the constraint on the executive variable:
  - Middle-income countries are disproportionately represented among countries above the median, accounting for 80 percent of countries with above-median developed political institutions.
  - In the below-median sample, middle-income countries account for 53 percent, including Indonesia, Morocco and Tunisia.
- Regression results by subsample (Table 12):
  - In countries with good property rights, domestic financial reforms are associated with faster economic growth at a 6 years horizon.
  - No significant association for domestic financial reforms in countries with bad property rights.
  - Some positive association between trade reforms and growth in the good property rights sample.
  - No significant impact of capital account reforms in either subsample.
  - Results for domestic financial reforms (and to some extent trade reforms) are robust to inclusion of controls and to endogeneity bias (appendix/IV results).
  - No sustained impact of domestic financial reforms on economic growth beyond a 6 year horizon (column 7).
- Interpretation: sufficiently developed political institutions (property rights) appear to be a precondition for reaping the benefits of economic reforms.

### Conclusions and implications for future research
- Summary conclusions:
  - Domestic financial reforms are robustly associated with positive growth for up to 6 years after reform.
  - Trade reforms show similar benefits in OLS but lose significance in IV estimations, suggesting endogeneity concerns.
  - Capital account liberalization is not associated with faster growth; reversals harm growth.
  - Well-developed property rights may be necessary for reforms to be effective.
- Most results are driven by middle-income countries; understanding why different reforms matter across income groups is an important area for future research.
- Open questions for further work:
  - Do different reforms matter in high- and low-income countries compared to middle-income countries?
  - How do other sequences among economic reforms affect outcomes?
  - Additional investigation of sequencing between economic and political reforms and other reform orderings is warranted and is the subject of ongoing research.

*Source: _wp09284 - conclusions remain unchanged.*

### References

### _wp09284 - References

### Major datasets, sample, and period
- Country sample: high, middle, and low income lists as shown in Table 1 (examples include Australia, Austria, Belgium, Canada, Czech Rep, Japan, Korea, US for High Income; Argentina, Belarus, Brazil, China, India, Mexico, Russia, Turkey for Middle Income; Bangladesh, Burkina Faso, Cote D'Ivoire, Ethiopia, Ghana, Kenya, Nepal, Nigeria, Pakistan, Tanzania, Uganda, Viet Nam, Zimbabwe for Low Income).
- Time period covered in regressions and analysis: 1974-2004.
- Observations reported in various specifications: up to 792, 359, 280, 317, 590, 230, 125, 267, 198 depending on table and subsample (see respective tables).
- Note on reform definitions (from table notes): “A reform is a strictly positive change in the liberalization index over the 3-year period, a large reform is a change greater or equal to the third quartile of the distribution of the change in the index, and a reversal is a strictly negative change in the index.”
- Dependent variable construction: non-overlapping three-year averages of growth in real GDP per capita (and similarly for TFP growth and Investment/GDP in channel regressions).

### Summary statistics (Table 2 — exact values)
- Real per capita GDP growth: Obs 5900; Mean 0.02; Std. Dev. 0.03; Min -0.13; Max 0.17.
- TFP growth: Obs 5340; Mean 0.02; Std. Dev. 0.03; Min -0.10; Max 0.10.
- Investment / GDP: Obs 5790; Mean 0.17; Std. Dev. 0.08; Min 0.02; Max 0.46.
- Trade liberalization Index: Obs 5840; Mean 0.72; Std. Dev. 0.22; Min 0; Max 1.
- Domestic Financial Lib Index: Obs 5900; Mean 0.51; Std. Dev. 0.29; Min 0; Max 1.
- Capital Account Lib Index: Obs 5900; Mean 0.59; Std. Dev. 0.37; Min 0; Max 1.
- Inflation: Obs 5894; Mean 1%; Std. Dev. 282%; Min -10%; Max 4770%.
- Fiscal balance / GDP: Obs 517; Mean -3.42; Std. Dev. 4.48; Min -25.11; Max 13.81.
- Banking crisis dummy: Obs 5900; Mean 0.16; Std. Dev. 0.32; Min 0; Max 1.
- External conflict dummy: Obs 5900; Mean 0.07; Std. Dev. 0.23; Min 0; Max 1.
- Private credit / GDP: Obs 5330; Mean 0.41; Std. Dev. 0.35; Min 0.01; Max 1.97.
- Constraint on the Executive: Obs 5835; Mean 0.04; Std. Dev. 2.16; Min 0; Max 7.
- Dummy inflation > 100 percent: Obs 5900; Mean 0.03; Std. Dev. 0.18; Min 0; Max 1.
- RER overvaluation: Obs 590; Mean -0.054; Std. Dev. 0.43; Min -153.84; Max 160.50.

### Bivariate correlations (Table 3 — selected exact entries)
- Correlation of Average annual growth with:
  - Inflation: -0.2306
  - Domestic finance: 0.1484
  - Trade (average tariff): 0.0385
  - Capital account: 0.1065
  - Private credit/GDP: 0.0273
- Correlation of Domestic finance with Trade (average tariff): 0.6014
- Correlation of Capital account with Domestic finance: 0.7097
- Correlation of Private credit/GDP with Average annual investment to GDP ratio: 0.5661

### Standard growth regressions (Table 4 — key coefficients and exact values)
- Specification (1) All countries (Dependent: GDP per capita growth)
  - Log(initial real GDP per capita): -0.0246*** [0.0064]
  - Initial inflation: -0.0004*** [0.0001]
  - Domestic finance: 0.0291* [0.0147]
  - Domestic finance * socialist legal origin: 0.1629*** [0.0418]
  - Trade: 0.0050 [0.0100]
  - Trade * socialist legal origin: 0.0351 [0.0659]
  - Capital account: 0.0073 [0.0060]
  - Capital account * socialist legal origin: -0.0248 [0.0158]
  - Observations: 792; R-squared: 0.40
- Specification (2) High income:
  - Log(initial real GDP per capita): -0.0303*** [0.0100]
  - Initial inflation: -0.0015 [0.0043]
  - Domestic finance: 0.0004 [0.0126]
  - Domestic finance * socialist legal origin: 0.1423*** [0.0399]
  - Trade: 0.0268* [0.0148]
  - Trade * socialist legal origin: 0.0546 [0.3752]
  - Capital account: -0.0012 [0.0071]
  - Capital account * socialist legal origin: -0.0101 [0.0087]
  - Observations: 270; R-squared: 0.45
- Specification (3) Middle income:
  - Log(initial real GDP per capita): -0.0275*** [0.0100]
  - Initial inflation: -0.0003*** [0.0001]
  - Domestic finance: 0.0529** [0.0243]
  - Domestic finance * socialist legal origin: 0.1644*** [0.0573]
  - Trade: 0.0015 [0.0142]
  - Trade * socialist legal origin: 0.0431 [0.0636]
  - Capital account: 0.0119 [0.0098]
  - Capital account * socialist legal origin: -0.0254 [0.0223]
  - Observations: 359; R-squared: 0.48
- Specification (4) Low income:
  - Log(initial real GDP per capita): -0.0059 [0.0278]
  - Initial inflation: -0.0237*** [0.0051]
  - Domestic finance: 0.0535* [0.0295]
  - Domestic finance * socialist legal origin: -0.0319 [0.0482]
  - Trade: -0.0014 [0.0159]
  - Trade * socialist legal origin: -0.3610*** [0.1216]
  - Capital account: -0.0242** [0.0112]
  - Capital account * socialist legal origin: 0.3662** [0.1494]
  - Observations: 156; R-squared: 0.40

(Note: Robust standard errors in brackets; ***, **, * indicate significance at the one, five, and ten percent levels, respectively.)

### Channels: TFP and Investment (Table 5 — selected exact entries)
- Panel reporting TFP growth and Investment / GDP with separate regressions for High, Middle, Low income groups.
- Example coefficients (Middle income group, TFP growth):
  - Log(initial real GDP per capita): -0.0341*** (0.0110)
  - Initial inflation: -0.0004*** (0.0001)
  - Domestic finance: 0.0489** (0.0236)
  - Domestic finance * socialist legal origin: 0.3723*** (0.0510)
  - Trade: -0.0046 (0.0136)
  - Trade * socialist legal origin: -0.1798*** (0.0261)
  - Capital account: 0.0097 (0.0114)
  - Capital account * socialist legal origin: 0.0602** (0.0255)
  - Observations: 280; R-squared (TFP growth): 0.35

- Example coefficients (Middle income group, Investment / GDP):
  - Log(initial real GDP per capita): 0.0350 (0.0282)
  - Initial inflation: 0.0002** (0.0001)
  - Domestic finance: 0.0560** (0.0259)
  - Domestic finance * socialist legal origin: 0.0508 (0.0600)
  - Trade: -0.0129 (0.0194)
  - Trade * socialist legal origin: 0.0631 (0.0450)
  - Capital account: 0.0195 (0.0161)
  - Capital account * socialist legal origin: 0.0369* (0.0198)
  - Observations: 317; R-squared (Investment / GDP): 0.83

### System GMM regressions (Table 6 — selected exact entries)
- Dependent variables: Real per capita GDP growth, TFP growth, Investment / GDP.
- Specification highlights:
  - Domestic finance coefficient (column 1, Real per capita GDP growth): 0.0566** (0.0288); Domestic finance * socialist legal origin: -0.0492 (0.0578).
  - Trade coefficient (column 1): -0.0402 (0.0250); Trade * socialist legal origin: 0.0905** (0.0429).
  - Capital account coefficient (column 1): 0.0134 (0.0109); Capital account * socialist legal origin: 0.0059 (0.0224).
  - Observations: column (1) 359; column (2) 280; column (3) 317.
  - Number of countries: 44, 31, 44 in respective columns.
  - Sargan test (p-value) reported as 0.11, 0.96, 0.99.
  - ar1 and ar2 statistics reported (e.g., ar1 -2.76, ar2 0.55 in column 1).

### Dynamic effects of reforms (Table 7 and subsequent tables — selected exact findings)
- Baseline dynamic effects (Table 7):
  - Example: For All countries, Trade contemporaneous reform: 0.0081** [0.0034]; Reform (previous period): 0.0101*** [0.0034]; Large reform (following period): 0.0187*** [0.0067].
  - Capital account contemporaneous reform reversal (All countries): -0.0173** [0.0067].
  - Controls noted include banking crisis, conflict, hyperinflation; hyperinflation defined as inflation exceeding 100 percent.
  - Observations in baseline GDP per capita growth regressions: 590 (All), 125 (Low), 267 (Middle), 198 (High). R-squared: 0.53, 0.51, 0.69, 0.55 respectively.

- Channels dynamic effects (Table 8 — TFP and Investment):
  - Domestic financial reform (example, All countries, contemporaneous reform on Investment/GDP): 0.0122** [0.0060].
  - Trade reform (example, Middle income, contemporaneous reform on GDP per capita growth): 0.0165*** [0.0059].
  - Capital account reform (example, Middle income, contemporaneous reform reversal on GDP per capita growth): -0.0234*** [0.0079].
  - Reporting of many lag structures: contemporaneous, previous period, following period; large reforms and reversals similarly included.

### Robustness, IV, and sequencing analyses (Tables 9–12 — selected exact values)
- Robustness tests (Table 9) report alternative controls, sample sizes, and R-squared values (examples):
  - Regressions including constraint on the executive, central government balance, overvaluation (Penn World Table), private credit to GDP, and school enrolment measures.
  - Observations across robustness specifications: 267, 230, 267, 233, 242, 198, 225. R-squared reported around 0.69–0.73.
- Instrumental variables regressions (Table 11a and 11b):
  - Second-stage results (example): Domestic financial reform (Real per capita GDP growth, second stage): contemporaneous reform 0.0179** [0.0083].
  - First-stage statistics (Table 11b, partial R2 and F test):
    - Example partial R2 and F test entries: Domestic financial reform contemporaneous reform Partial R2 0.40; F test 8.0. Trade reform contemporaneous reform Partial R2 0.34; F test 11.3. Capital account contemporaneous reform Partial R2 0.48; F test 7.5.
  - First-stage Hansen J test p-values and instrument strength reported: e.g., Hansen J test 0.22, 0.15, 0.15 for three second-stage columns; F-test values and partial R2s listed by endogenous variable and lag structure.
- Sequencing economic and political reforms (Table 12):
  - Controls and interaction samples include “Good property rights sample” and “Bad property rights sample.”
  - Examples of coefficients (Middle-income samples):
    - Domestic financial reform (contemporaneous reform) coefficients across controls: 0.0121** [0.0061] (column with Property rights control); Reform (previous period): 0.0137** [0.0059].
    - Trade reform (contemporaneous reform) 0.0161*** [0.0060] (Property rights control column).
    - Capital account contemporaneous reform: -0.0111* [0.0057] (Property rights control column); contemporaneous reversal: -0.0299*** [0.0104].
  - Notes: regressions include country and year fixed effects, interactions with transition-country dummy, and robustness to inclusion of long lags and alternative control sets.

### Miscellaneous methodological and reference notes (from the References list)
- Extensive citation list covering financial reform databases, measures of financial openness, capital account liberalization literature, growth empirics, and econometric methods (e.g., Arellano and Bond 1991; Blundell and Bond 1998; Stock and Yogo 2005; Windmeijer 2004).
- Key referenced datasets and measures: new database of financial reforms; measures by Chinn and Ito; Quinn capital account measures; Wacziarg and Welch trade liberalization measures; private credit and macro controls from standard sources referenced in the bibliography.

*References and supporting tables as contained in the source PDF.*

### Appendix table. 2 SLS regressions

### Appendix table. 2 SLS regressions

### Sample
- Good property rights sub-sample
- Time period: 1974-2004
- Dependent variable: non-overlapping three-year averages of growth in real GDP per capita
- Observations163159163
- R-squared0.520.580.90
- J test0.230.140.28

### Regression results (coefficients with robust standard errors in brackets)
- Capital account reform:
  - Contemporaneous reform -0.0073 [0.0072]
  - Contemporaneous reform -0.0048 [0.0067]
  - Contemporaneous reform -0.0116** [0.0058]

- Capital account reversal:
  - Contemporaneous reversal -0.0216* [0.0115]
  - Contemporaneous reversal -0.0083 [0.0105]
  - Contemporaneous reversal -0.0367*** [0.0085]

- Reform lagged 3 years:
  - reform lagged 3 years -0.0132* [0.0070]
  - reform lagged 3 years -0.0156*** [0.0053]
  - reform lagged 3 years -0.0043 [0.0045]

- Reversal lagged 3 years:
  - reversal lagged 3 years -0.0290** [0.0126]
  - reversal lagged 3 years -0.0128 [0.0109]
  - reversal lagged 3 years -0.0204** [0.0099]

- Reform 3 years before:
  - reform 3 years before -0.0077 [0.0059]
  - reform 3 years before -0.0029 [0.0060]
  - reform 3 years before -0.0002 [0.0053]

- Reversal 3 years before:
  - reversal 3 years before -0.0140 [0.0117]
  - reversal 3 years before 0.0011 [0.0110]
  - reversal 3 years before -0.0168* [0.0089]

### Statistical significance notation
- ***, **, * indicate significance at the one, five, and ten percent levels, respectively.

### Controls and specification
- A reform is a strictly positive change in the liberalization index over the 3-year period.
- A large reform is a change greater or equal to the third quartile of the distribution of the change in the index.
- A reversal is a strictly negative change in the index.
- The regression includes contemporaneous, previous period and following period reform episodes.
- Control variables include initial log real GDP per capita, initial log inflation, and dummy variables for banking crisis, internal conflicts and hyperinflation (defined as inflation exceeding 100 percent).
- Country and year fixed effects are included in all specifications.
- Robust standard errors (in brackets).

*Source: Appendix table. 2 SLS regressions (from the provided content).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09284.pdf_