## wpiea2019049 - 2012. Some recent studies adopt treatment effect techniques to address the endogeneity ## Source details **Canonical URL:** [wpiea2019049 - 2012. Some recent studies adopt treatment effect techniques to address the endogeneity](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019049.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2019/wpiea2019049.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019049.pdf.json) --- ### Contribution and main results - Develops a new instrumental variables strategy exploiting adoption of fiscal rules in neighboring countries as an instrument for domestic fiscal rule adoption. - Two main empirical results: - The effect of (any) rule on the fiscal deficit is statistically insignificant once potential endogeneity is adequately controlled for. - Using a continuous fiscal rule strength index (Schaechter and others 2012), better designed fiscal rules have a significant impact on the fiscal balance: moving from a relatively weakly designed fiscal rule to a better designed fiscal rule can increase the fiscal balance by some 0.6 percent of GDP. ### Rationale for the instrumental variable and construction - Identification assumption: - Fiscal rule adoption in neighboring countries may induce domestic adoption through diffusion channels (economic competition, learning, socialization, mimicking, coercion). - Neighboring countries’ rules provide an exogenous source of variation in domestic rules that does not directly impact the fiscal balance (relevance tested; exogeneity cannot be directly tested). - Instrument definition (geographical diffusion / contiguity instrument): - Instrument equals the number of fiscal rules in place in countries with common borders relative to the domestic country. - Formal expression uses 퐹퐹𝑅𝑅𝑗𝑗,𝑑𝑑 (dummy for country j having a fiscal rule at time t) and 푋푋𝑗𝑗,𝑖𝑖,𝑑𝑑 (takes value 0 when countries have no common borders and otherwise sums number of countries with common borders). - Data on common borders obtained from Frankel and Rose, 2002 and Rose, 2007. - Channels of diffusion cited: coercion, competition, imitation, learning, socialization; regional waves and leader–laggard dynamics; anecdotal examples (Chilean model in Latin America; Swiss debt brake influence on Germany; Czech Republic following Poland and Slovakia). ### Empirical specification and estimators - Baseline augmented fiscal reaction function: - balancei,t = Ξ²1 rulei,t + Ξ²2 Xi,t + Ξ±i + Ξ»t + Ξ΅it - "balance" = nominal budget balance for country i at time t (percent of GDP). - "rule" = dummy equal to 1 if the country has a fiscal rule. - Xi,t includes lags of the budget balance, lagged debt, GDP per capita, GDP growth, output gap, and terms of trade movements. - Country fixed effects (Ξ±i) and year fixed effects (Ξ»t) included. - Estimators: - OLS with country and year fixed effects and lagged balances (baseline). - IV: two-stage least squares using contiguity instrument (contemporaneous, lagged by one year, and 3-year moving average). - Weak-instrument tests: Kleinbergen-Paap rk test; Anderson-Rubin (AR) identification-robust inference. ### OLS and IV estimation results (selected exact figures) - OLS fiscal rule coefficients (Table 1 Columns 1–4): - 1.36*** (0.43) - 1.03** (0.46) - 0.70*** (0.23) - 0.80*** (0.22) β€” preferred specification (with full fixed effects and dynamics): countries with fiscal rules have fiscal balance on average 0.8 percent higher. - Other OLS (Column 4): Real GDP growth = 0.12*** (0.02); Delta terms of trade = 0.04*** (0.01); Observations = 2,823; R-squared = 0.71. - IV fiscal rule coefficients (contiguity instrument, Table 2 Columns 1–3): - 0.86 (0.76) - 1.38 (0.87) - 1.17 (0.90) β€” coefficients lose statistical significance relative to OLS though magnitudes comparable or larger. - First-stage contiguity coefficients (Table 3): - Contiguity = 0.11*** (0.03) - Lag 1 contiguity = 0.08*** (0.03) - Contiguity MA = 0.103*** (0.031) β€” neighboring countries’ adoption increases domestic adoption probability by around 10 percent. - Kleinbergen-Paap rk test statistics for contiguity IV (Table 2): 15.24, 11.23, 12.46 (above Staiger and Stock (1997) rule-of-thumb value of 10). - Stock-Wright p-value = 0.00 for all three contiguity instrument variants. - Anderson-Rubin (AR) confidence intervals closely match standard Wald intervals (Figure 3), supporting that non-significant IV estimates are not driven by weak instruments. ### Robustness, identification concerns, and additional checks - Exclusion restriction concerns (possible direct spillovers from neighbors’ rules via trade) are addressed by augmenting regressions with: - Trade openness (trade as % of GDP). - Average growth in the region. - Robustness results using lag 1 contiguity instrument (Table 4): - Including trade openness and regional growth: rule coefficient remains similar in magnitude and not significant; regional growth significant in some specifications (e.g., 0.11**). - Dividing contiguity by border length: results unchanged. - Excluding all European countries or EU countries: rule coefficients remain insignificant. - Excluding the global financial crisis and removing year fixed effects: non-significant effect persists. - IV without any control variables: magnitude and significance unaffected. - First-stage robustness (Table A4): lag 1 contiguity remains strongly associated with domestic adoption across robustness samples; example: Full sample Lag 1 contiguity = 0.084*** (0.025); No Europe = 0.116*** (0.024); Kleinbergen-Paap rk in robustness exercises often remains above rule-of-thumb thresholds (e.g., Column 3 = 23.36). - Alternative instruments used in literature perform poorly in this global sample (Table 5): - Instruments: inflation targeting (IT), government fragmentation (Govfrac), checks and balance (Checks). - Kleinbergen-Paap rk test statistics: 7.65, 0.51, 0.65 respectively (below 10 except IT marginal). - The paper notes theoretical concerns that these instruments may directly affect fiscal balance. ### Fiscal rule strength index: estimation and findings - Fiscal Rule Strength (FRS) index: - Equals 0 for countries without rules; ranges from 0.1 to 1 for countries with rules; sample average reported elsewhere as 0.26 (Table A3). - Table A3 summary statistics: Mean = 0.294; Std. Dev. = 0.140; 25% = 0.181; 50% = 0.259; 75% = 0.375; distribution skewed. - IV for strength index: average strength in bordering countries (lagged) β€” first-stage strong: - L1 strength contiguity = 0.018*** (0.004) in full-sample first stage. - Kleinbergen-Paap rk test for strength instrument = 19.01 (Table 6, Column 1). - IV results for strength index (Table 6): - Column (1) Full sample: Strength Index coefficient = 3.91** (1.69) - Column (2): 3.83** (1.84) - Column (3): 4.43** (1.97) - Interpretation example in text: - Using a coefficient of 3.4 and moving from the 25th percentile to the 75th percentile (actual change 0.19) implies improvement of budget balance by 0.64 percent of GDP (3.4 * 0.19 = 0.646). - Robustness of strength results: - Generally robust to earlier checks but lose significance when excluding European countries (European countries provide much of the variation in strong rules; standard deviation of index reduces from 1.4 to 0.65 when Europe excluded as noted). ### Data, sample, and summary statistics - Sample: panel of up to 142 countries over 1985–2015; IMF fiscal rules dataset provides information for 96 countries with fiscal rules in place. - Dependent variable: nominal fiscal balance as a percent of GDP. - Main treatment variables: - Fiscal rule dummy (presence/absence). - Fiscal rule strength index (FRS): equals 0 for countries without rules, ranges from 0.1 to 1 for countries with rules; sample average 0.26 (Table A3). - Sample sizes: Observations for OLS and various IVs range from 1,921 to 3,309 depending on specification; Number of id up to 157 (Tables 1–6). - Selected summary statistics (Table A2 and text): - Fiscal balance mean = -1.99; Std. Dev. = 6.05. - Real GDP growth mean = 4.02. - Trade openness mean = 84.02. - Contiguity IV mean = 1.04. - Fiscal rule strength index moments: Mean = 0.294; Std. Dev. = 0.140; 25% = 0.181; 50% = 0.259; 75% = 0.375. ### Caveats and interpretation limits - Instrument exogeneity cannot be directly tested; paper discusses potential channels that could invalidate exclusion restrictions (e.g., trade spillovers). - Strength index measures design features (institutional coverage, monitoring/enforcement independence, legal base, flexibility, correction mechanisms and sanctions) but does not measure implementation, public/political support, or quality of budgeting practices. - IV results indicate average presence of a fiscal rule shows no statistically significant causal effect on fiscal balance, while rule design (strength) matters β€” results sensitive to exclusion of regions (notably Europe). ### Conclusions and policy implications - Endogeneity is a major concern: countries may adopt rules due to fiscal preferences or during periods of stress. - The contiguity-based diffusion instrument provides a viable exogenous source of variation; first-stage relevance is strong. - Key empirical implications: - Any-type fiscal rule: no statistically significant causal effect on fiscal balance in the IV framework (despite positive OLS correlations). - Fiscal rule design: better designed rules (higher strength index) are positively and significantly associated with improved fiscal balances in IV estimation. - Policy implication: focusing on rule design features (strength) is more likely to yield fiscal improvements than mere adoption of any rule, while recognizing measurement caveats (design vs implementation). *Source: wpiea2019049 - 2012. Some recent studies adopt treatment effect techniques to address the endogeneity* ### 2012. Some recent studies adopt treatment effect techniques to address the endogeneity ### wpiea2019049 - 2012. Some recent studies adopt treatment effect techniques to address the endogeneity ### Contribution and main results - Develops a new instrumental variables strategy for estimating the effectiveness of fiscal rules, exploiting adoption of fiscal rules in neighboring countries as an instrument. - Two main empirical results: - The effect of (any) rule on the fiscal deficit is statistically insignificant once potential endogeneity is adequately controlled for. - Using an index capturing a continuous measure of fiscal rule strength (Schaechter and others 2012), better designed fiscal rules have a significant impact on the fiscal balance: moving from a relatively weakly designed fiscal rule to a better designed fiscal rule can increase the fiscal balance by some 0.6 percent of GDP. ### Rationale for the instrumental variable - Key identifying assumption: fiscal rule adoption in neighboring countries may induce domestic adoption through diffusion channels documented in political science (economic competition, learning, socialization, mimicking, coercion). - The presence of rules in neighboring countries provides an exogenous source of variation in domestic rules that does not directly impact the fiscal balance (authors perform weak instrument tests that confirm relevance; exogeneity cannot be directly tested and potential caveats and exclusion restriction threats are discussed). ### Channels of diffusion (literature grounding) - Cited channels and examples: - Coercion, competition, imitation, learning, socialization (Dobbin, Simmons, and Garrett 2007; Shipan and Volden 2008). - Regional waves and leader–laggard dynamics in policy adoption (Persson and Tabellini 2009; Acemoglu and others 2016; Huntington 1991; Markoff 1996). - Anecdotal examples: Chilean model influencing Latin America; Germany adopting a debt break influenced by Swiss debt brake (Kirchgassner 2017); Czech Republic adoption following Poland and Slovakia. - Empirical suggestion that fiscal rules’ β€œpopularity” (aggregate number of debt rules) predicts adoption (Altunbas and Thornton 2017). ### Construction of the instrumental variable - Instrument defined to capture geographical diffusion: number of fiscal rules in place in countries with common borders relative to the domestic country. - Formal expression provided: - 푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐푐_퐼퐼퐼퐼 ν‘–ν‘–,ν‘‘ν‘‘ = �퐹퐹푅푅 ν‘—ν‘—,ν‘‘ν‘‘ βˆ— ν‘‹ν‘‹ ν‘—ν‘—,ν‘–ν‘–,ν‘‘ν‘‘ ν‘—ν‘—β‰ ν‘–ν‘– - Where 퐹퐹𝑅𝑅𝑗𝑗,𝑑𝑑 is a dummy equal to 1 when country j has a fiscal rule at time t, and 0 otherwise. - 푋푋𝑗𝑗,𝑖𝑖,𝑑𝑑 takes the value 0 when countries have no common borders and sums the number of countries with common borders. - Data on common borders obtained from Frankel and Rose, 2002 and Rose, 2007. ### Empirical specification - Baseline augmented fiscal reaction function (following Debrun and others 2008 and Baum and others 2017): - ν‘ν‘Žν‘™ν‘Žν‘›ν‘ν‘’π‘–,𝑑 = ν›½1 ν‘Ÿν‘’π‘™π‘’π‘–,𝑑 + ν›½2 𝑋𝑖,𝑑 + 훼𝑖 + νœ†π‘‘ + νœ€π‘–π‘‘ - Where: - "balance" is the nominal budget balance for country i at time t. - "rule" is a dummy equal to 1 if the country has a fiscal rule, and 0 otherwise. - 𝑋𝑖,𝑑 includes standard determinants: lags of the budget balance, lagged debt, GDP per capita, GDP growth, and the output gap, plus terms of trade movements to account for commodity price effects especially in LICs. - Country fixed effects (훼𝑖) and year fixed effects (νœ†π‘‘) are included. ### Robustness and caveats discussed - Weak instrument tests confirm the instrument’s relevance. - The paper explicitly acknowledges inability to test the instrument’s exogeneity directly and discusses channels that could invalidate exclusion restrictions. - Robustness exercises and further checks are presented in subsequent sections (Sections IV–V) to address potential weaknesses. *Source: wpiea2019049 - 2012. Some recent studies adopt treatment effect techniques to address the endogeneity* ### introduction of fiscal rules, a dummy for periods when the countries are under Fund ### wpiea2019049 - introduction of fiscal rules, a dummy for periods when the countries are under Fund ### Data and empirical strategy - Sample: panel of up to 142 countries over 1985–2015; IMF fiscal rules dataset provides information for 96 countries with fiscal rules in place. - Dependent variable: nominal fiscal balance as a percent of GDP. - Main treatment variables: - Fiscal rule dummy (presence/absence). - Fiscal rule strength index (FRS): equals 0 for countries without rules, ranges from 0.1 to 1 for countries with rules; sample average 0.26 (Table A3). - Instrumental variable strategy: - New instrument: diffusion of fiscal rule adoption in neighboring countries (contiguity instrument). - Instrument for strength index: average strength of fiscal rules in bordering countries (Equation (4)). - Estimators: - Baseline: OLS with country and year fixed effects and lagged balances. - IV: two-stage least squares using contiguity instrument (contemporaneous, lagged by one year, and 3-year moving average). - Tests for weak instruments: Kleinbergen-Paap rk test and Anderson-Rubin (AR) identification-robust inference. ### OLS results (conditional correlates) - Baseline OLS suggests a positive correlation between fiscal rules and the budget balance (Table 1): - Column (1) Country FE only: Fiscal rule coefficient = 1.36*** (0.43) - Column (2) + Year FE: Fiscal rule coefficient = 1.03** (0.46) - Column (3) + Lags of balance: Fiscal rule coefficient = 0.70*** (0.23) - Column (4) preferred specification: Fiscal rule coefficient = 0.80*** (0.22) - Interpretation: with full fixed effects and dynamics, countries with fiscal rules have fiscal balance on average 0.8 percent higher than countries without fiscal rules (per Column 4). - Other notable OLS coefficients (Column 4, Table 1): - Real GDP growth = 0.12*** (0.02) - Delta terms of trade = 0.04*** (0.01) - Observations = 2,823; R-squared = 0.71; Country FE = Yes; Year FE = Yes. ### Instrumental variable estimation using contiguity - IV baseline (Table 2): instrumenting fiscal rule dummy with contiguity measures. - Column (1) contemporaneous contiguity: Fiscal rule = 0.86 (0.76) - Column (2) lag 1 contiguity: Fiscal rule = 1.38 (0.87) - Column (3) 3-year moving average contiguity: Fiscal rule = 1.17 (0.90) - Coefficients lose statistical significance relative to OLS but magnitudes are comparable or slightly larger. - Relevance of contiguity instrument: - First-stage (Table 3): neighboring countries’ adoption increases domestic adoption probability by around 10 percent: - Contiguity = 0.11*** (0.03) - Lag 1 contiguity = 0.08*** (0.03) - Contiguity MA = 0.103*** (0.031) - Kleinbergen-Paap rk test statistics (Table 2): 15.24, 11.23, 12.46 for the three instrument variants β€” above Staiger and Stock (1997) rule-of-thumb value of 10. - Stock-Wright p-value = 0.00 for all three instrument variants. - Weak-instrument-robust inference: - Anderson-Rubin (AR) confidence intervals closely match standard Wald intervals (Figure 3), supporting that non-significant IV estimates are not driven by weak instruments. ### Robustness checks for contiguity instrument and identification concerns - Exclusion restriction concerns: potential direct spillovers from neighbors’ rules (e.g., trade linkages) addressed by augmenting regressions with: - Trade openness (trade as % of GDP). - Average growth in the region. - Robustness results (Table 4, using lag 1 contiguity instrument): - Including trade openness and regional growth (Columns 1–2): rule coefficient remains similar in magnitude and not significant; growth in the region is significant in some specifications (e.g., 0.11** in Column 2). - Dividing contiguity by border length (not reported): results unchanged. - Excluding all European countries (Column 3) or EU countries (Column 4): rule coefficients remain insignificant. - Excluding the global financial crisis (Column 5) and removing year fixed effects (Column 6): non-significant effect of the rule persists. - IV without any control variables (Column 7): magnitude and significance unaffected. - First-stage robustness (Table A4): lag 1 contiguity remains strongly associated with domestic adoption across robustness samples; coefficient examples: - Full sample: Lag 1 contiguity = 0.084*** (0.025) (Column 1) - No Europe: 0.116*** (0.024) (Column 3) - Kleinbergen-Paap rk test values in robustness exercises remain above rule-of-thumb thresholds in most cases (e.g., Column 3 = 23.36). - Alternative instruments used in literature perform poorly in this global sample (Table 5): - Instruments tested: inflation targeting (IT), government fragmentation (Govfrac), checks and balance (Checks). - Kleinbergen-Paap rk test statistics: 7.65, 0.51, 0.65 respectively (all below 10 except IT marginally); Stock-Wright p-values indicate weak instrument concerns. - The paper argues potential direct effects of these instruments on fiscal balance (theoretical concerns): government fragmentation, checks and balance, and inflation targeting may directly affect fiscal outcomes. ### Extensions: fiscal rule strength index (design features) - Replacement of fiscal rule dummy with continuous fiscal rule strength index (FRS) (Equation (3) and construction described): index averages 0.294 mean, std. dev. 0.140 (Table A3); distribution skewed with median 0.259 and 75th percentile 0.375. - IV for strength index: average strength in bordering countries (lagged) β€” first-stage confirmed strong correlation (Table A5 examples): - L1 strength contiguity = 0.018*** (0.004) in full-sample first stage. - Kleinbergen-Paap rk test for strength instrument = 19.01 (Table 6, Column 1). - IV results for strength index (Table 6): - Column (1) Full sample: Strength Index coefficient = 3.91** (1.69) - Column (2): 3.83** (1.84) - Column (3): 4.43** (1.97) - Interpretation example provided in text: coefficient on strength index = 3.4; moving from 25th percentile to 75th percentile (actual change 0.19) implies improvement of budget balance by 0.64 percent of GDP (3.4 * 0.19 = 0.646). - Robustness of strength results: - Results generally robust to earlier robustness checks but lose significance when excluding European countries (likely because European countries generally exhibit the strongest rules and removing them reduces variation; standard deviation of index reduces from 1.4 to 0.65 as noted). - Caveats on the strength index: - Index measures design features (broad institutional coverage, independence of monitoring/enforcement, legal base, flexibility, correction mechanisms and sanctions) but not implementation or public/political support, nor the quality of budgeting practices. ### Conclusion and implications - Endogeneity is a major concern when estimating the link between fiscal rules and fiscal outcomes; countries may adopt rules due to fiscal preferences or during periods of stress. - The contiguity-based diffusion instrument provides a viable exogenous source of variation in rule adoption: - First-stage relevance is strong; IV estimates for the average rule become statistically insignificant despite OLS positive correlations. - This aligns with a meta-analysis result: any type of fiscal rules have no statistically significant impact on the fiscal balance once endogeneity is adequately addressed (Heinemann and others, 2018). - Crucial finding: the average presence of a fiscal rule does not show a statistically significant causal effect on fiscal balance in the IV framework, but the design (strength) of rules matters: - Better designed rules (higher strength index) are positively and significantly associated with improved fiscal balances in IV estimation. - Well-designed rules tend to have broad institutional coverage, clear links to sustainability, are easy to monitor, and support countercyclical policy (consistent with Eyraud and others, 2018). - Policy implication (as interpreted from results): focusing on rule design features (strength) is more likely to yield fiscal improvements than mere adoption of any rule, although the analysis highlights measurement caveats (design vs implementation). ### Key statistics and test results (selected exact values) - OLS fiscal rule coefficients: 1.36*** (0.43), 1.03** (0.46), 0.70*** (0.23), 0.80*** (0.22) (Table 1 Columns 1–4). - IV fiscal rule coefficients (contiguity): 0.86 (0.76), 1.38 (0.87), 1.17 (0.90) (Table 2 Columns 1–3). - First-stage contiguity coefficients: 0.11*** (0.03), 0.08*** (0.03), 0.103*** (0.031) (Table 3). - Kleinbergen-Paap rk test (contiguity IV): 15.24, 11.23, 12.46 (Table 2). - Strength index IV coefficients (Table 6 Column 1): Strength Index = 3.91** (1.69); example coefficient used in text = 3.4 for interpretation. - Strength index percentiles and moments (Table A3): Mean = 0.294; Std. Dev. = 0.140; 25% = 0.181; 50% = 0.259; 75% = 0.375. - Sample sizes: Observations for OLS and various IVs range from 1,921 to 3,309 depending on specification; Number of id up to 157 (Tables 1–6). - Summary statistics (Table A2): Fiscal balance mean = -1.99; Std. Dev. = 6.05; Real GDP growth mean = 4.02; Trade openness mean = 84.02; Contiguity IV mean = 1.04. *Source: wpiea2019049 - introduction of fiscal rules, a dummy for periods when the countries are under Fund programs (IMF working paper content provided).* --- _Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019049.pdf_