## _wp0549

## Source details

**Canonical URL:** [_wp0549](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0549.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0549.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0549.pdf.json)

---

### I. Introduction: purpose and approach
- Main purpose: empirically determine the main causes of the worldwide diversity of inflationary experiences.
- Two shortcomings in prior literature identified:
  - Empirical models generally fail to account for inflation inertia and the endogeneity of important economic and political variables.
  - Earlier political variables were relatively poorer measures of political instability than those available in newer datasets (e.g., DPI and CNTS).
- Methodology and data overview:
  - Use generalized method of moments system (GMM) estimation applied to dynamic panel data to address econometric limitations of OLS.
  - Dataset covers around 100 countries for the period 1960–99.
- Preview of key empirical findings:
  - Political instability leads to higher inflation.
  - Impact of political instability on inflation is stronger for high-inflation than for moderate- and low-inflation countries.
  - Impact is higher for developing countries than for industrial countries.
  - Institutions such as economic freedom and democracy are important determinants of inflation; higher economic freedom and democracy are associated with lower inflation.

### II. Data, variables, and empirical model
- Dataset composition:
  - Annual political, institutional, and economic variables for 178 countries for years 1960–99; estimation samples reduce to at most 97 countries for inflation and 66 for seigniorage because of missing values.
- Key data sources:
  - Political/institutional: Cross National Time Series Data Archive (CNTS); Database of Political Institutions (DPI 3.0); Polity IV; Gwartney and Lawson (2002); Freedom House.
  - Economic: World Bank WDI; Global Development Network Growth Database (GDN); IMF IFS; Penn World Tables (PWT 6.1); OECD Statistical Compendium.
- Dependent variable and transformations:
  - Dependent variable: logarithm of inflation (IFS); first-difference of Log(Inflation) used in reported estimations (D1 Log(Inflation)).
- Explanatory variables (selected):
  - Lagged logarithm of inflation (IFS) treated as endogenous.
  - Political instability: Government crises (CNTS), Cabinet changes (CNTS).
  - Institutions: Index of economic freedom (Gwartney and Lawson, 2002); Polity scale (Polity IV, range -10 to +10).
  - Structural and external variables: Agriculture (percent of GDP), Trade (percent of GDP), Growth of real GDP per capita (PWT 6.1), Real overvaluation (GDN), Growth of oil prices (OECD), U.S. treasury bill rate (IFS).
- Endogeneity treatment and estimator:
  - Lagged dependent variable and variables potentially affected by inflation treated as endogenous.
  - System-GMM (Blundell and Bond) estimation used; two-step results with Windmeijer (2000) robust standard-error correction; Sargan tests do not reject over-identifying restrictions.
- Significance notation used in tables: a=1 percent; b=5 percent; c=10 percent.

### III. Main empirical results (selected coefficients and samples)
- Estimation framework: D1 Log(Inflation) dependent variable; five reported specifications (Columns 1–5) differ by included variables and sample (Column 5 = developing countries only).
- Lagged first-difference of Log(Inflation) (LD1) — strong persistence:
  - Column 1: 0.644 (t-statistic (16.7)a)
  - Column 2: 0.653 (18.8)a
  - Column 3: 0.672 (20.9)a
  - Column 4: 0.711 (23.9)a
  - Column 5: 0.639 (16.4)a
- Political instability variables (first differences):
  - Government crises D1:
    - Column 1: 0.139 (1.85)c
    - Column 2: 0.247 (2.33)b
    - Column 3: 0.161 (2.46)b
    - Column 5: 0.202 (2.34)b
  - Cabinet changes D1:
    - Column 4: 0.091 (1.98)c
- Institutions:
  - Index of economic freedom D1:
    - Column 1: -0.249 (-6.18)a
  - Polity scale D1:
    - Column 1: 0.002 (-0.58)
    - Column 2: -0.008 (-2.96)a
    - Column 3: -0.007 (-2.36)b
    - Column 4: -0.005 (-2.27)b
    - Column 5: 0.005 (-1.34)
- Other covariates (D1) — selected:
  - Trade (in percent of GDP):
    - Column 1: -0.00004 (-0.07)
    - Column 2: -0.001 (-2.26)b
    - Column 3: -0.001 (-2.71)a
    - Column 4: -0.001 (-3.40)b
    - Column 5: -0.002 (-2.73)a
  - Growth of real GDP per capita:
    - Column 1: -0.022 (-3.22)a
    - Column 2: -0.037 (-4.78)a
    - Column 3: -0.033 (-4.92)a
    - Column 4: -0.02 (-3.80)a
    - Column 5: -0.032 (-4.61)a
  - Real overvaluation D1:
    - Column 1: -0.002 (-2.41)b
    - Column 2: -0.002 (-1.93)c
    - Column 3: -0.002 (-2.21)b
    - Column 4: -0.002 (-2.52)a
    - Column 5: -0.001 (-1.53)
  - Growth of oil prices D1:
    - Column 1: 0.004 (6.03)a
    - Column 2: 0.004 (6.05)a
    - Column 3: 0.004 (6.59)a
    - Column 4: 0.004 (6.64)a
    - Column 5: 0.003 (4.17)a
  - U.S. treasury bill rate D1:
    - Column 1: 0.016 (2.39)b
    - Column 2: 0.033 (4.78)a
    - Column 3: 0.034 (5.62)a
    - Column 4: 0.028 (4.80)a
    - Column 5: 0.038 (4.00)a
- Sample sizes:
  - Number of observations:
    - Column 1: 1,703
    - Column 2: 2,225
    - Column 3: 2,629
    - Column 4: 2,630
    - Column 5: 1,877
  - Number of countries:
    - Column 1: 89
    - Column 2: 95
    - Column 3: 97
    - Column 4: 97
    - Column 5: 75
- Quantified effect statements from text:
  - "An additional government crisis increases the inflation rate by 16.1 percent (Column 3)."
  - A cabinet change increases inflation by 9.1 percent (Column 4) — reported text: "A government crisis increases inflation by 9.1 percent (Column 4)."
  - When only developing countries are considered, a government crisis increases inflation by 20.2 percent (Column 5).
  - Example: if the inflation rate is at its sample mean of 51.98 percent, a government crisis will push it to 60.35 percent, i.e., inflation increases by 8.37 percentage points.

### IV. Interaction results: high-inflation episodes and country groups
- Inflation-regime interactions:
  - Interactions with inflation < 50 percent are not statistically significant.
  - Interactions with inflation ≥50 are highly statistically significant and positive.
- Magnitudes in high-inflation contexts:
  - When inflation is high or very high, an additional government crisis increases inflation by 84.5 percent (see Column 1).
  - An additional cabinet change increases inflation by 97.3 percent (see Column 3).
- Country-group interactions:
  - Interactions with industrial countries are not statistically significant.
  - Interactions with developing countries are statistically significant.
  - Conclusion: the positive relationship between political instability and inflation holds essentially for developing countries.

### V. Seigniorage: determinants and political instability
- Seigniorage measure: ratio of the change in reserve money (IFS, line 14) to total government revenues (IFS, line 81).
- Main coefficients (Table 3, fixed-effects estimator; lags indicated as (-1)):
  - Cabinet changes (-1): .048 (t-statistic (2.48)b)
  - Government crises (-1): .048 (t-statistic (1.87)c)
- Interaction effects (seigniorage, lagged):
  - [Cabinet changes (inflation ≥ 50 percent)] (-1): .375 (t-statistic (2.41)b)
  - [Government crises (inflation ≥ 50 percent)] (-1): .568 (t-statistic (2.47)b)
  - [Cabinet changes (developing countries)] (-1): .068 (t-statistic (2.53)b)
  - [Government crises (developing countries)] (-1): .094 (t-statistic (1.91)c)
  - Interactions for industrial countries not significant (example: [Cabinet changes (industrial countries)] (-1): -.003 (-.58))
- Other seigniorage correlates:
  - Agriculture (percent of GDP): coefficient .015 with significance (e.g., (3.32)a)
  - Real overvaluation (-1): coefficient .0002 with strong significance (e.g., (4.43)a)
  - Growth of oil prices: negative coefficients (lower oil prices associated with higher seigniorage in some specifications) with marginal significance.
  - Treasury bill rates: coefficients around .007 to .008 with modest significance (e.g., (1.93)c to (2.35)b)
- Model specifics and fit:
  - Fixed-effects preferred (Hausmann tests); number of observations across Table 3 specifications ranges from 1,520 to 1,534; number of countries equals 65–66; Adjusted R2 ranges from .23 to .28.

### VI. Mechanisms, robustness checks, and volatility
- Mechanisms:
  - Excessive money growth leading to high inflation is generally caused by attempts to extract large seigniorage revenues.
  - More unstable and polarized political systems have more inefficient tax structures and rely more on seigniorage.
  - Frequent cabinet changes/government crises shorten policymakers’ horizons, increasing emphasis on short-term objectives and reducing ability to maintain low inflation.
  - Inflation can increase political instability via its costs and the responsibility hypothesis (voters hold governments responsible).
- Robustness tests (inflation models):
  - Adding variables or replacing some variables by reasonable alternatives produced similar results.
  - Variables that, when included, lead to lower inflation: greater executive constraints, more political rights, more civil liberties, and higher real GDP growth (when used instead of growth of real GDP per capita).
  - Proxies not statistically significant in many specifications: ideological polarization (in some tests), urbanization, currency inside banks, GDP growth of main trading partners, exchange rate regime, and central bank independence.
  - Additional checks: inclusion of year/decade or region dummies; alternative samples excluding extreme inflation values (annual rates above 1000 percent) and excluding Latin America — results remained very similar.
- Volatility:
  - Inflation becomes more volatile at higher levels and with greater political instability, less economic freedom, greater ideological polarization, and greater fragmentation of parties’ shares in parliament.

### VII. Policy implications and conclusions
- Main conclusions:
  - Higher political instability (measured by cabinet changes, government crises, and related institutional indicators) generates higher inflation rates and higher seigniorage.
  - Effects are more pervasive and stronger in developing countries and in high-inflation (above 50 percent) countries than in developed and low-inflation countries.
- Policy recommendations:
  - Reforms aimed at reducing political instability and increasing economic freedom and democracy would help reduce inflation.
  - Inflation-stabilization efforts may be only temporarily effective if they do not include serious fiscal and political reforms.
  - Policymakers in developing countries should reform institutions and create viable mechanisms conducive to long-run price stability.

*Source: System-GMM estimations for dynamic panel-data models (using Stata 8.2); results and notes as presented in the source content.*

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

### _wp0549 - References

### I. INTRODUCTION
- Main purpose: empirically determine the main causes of the worldwide diversity of inflationary experiences.
- Two identified shortcomings in prior literature:
  - Empirical models generally fail to account for inflation inertia and the endogeneity of important economic and political variables.
  - Earlier political variables were relatively poorer measures of political instability than those available in newer datasets (e.g., DPI and CNTS).
- Methodology and data overview:
  - Use generalized method of moments system (GMM) estimation applied to dynamic panel data to address econometric limitations of OLS.
  - Dataset covers around 100 countries for the period 1960–99.
- Key empirical findings previewed:
  - Political instability leads to higher inflation.
  - Impact of political instability on inflation is stronger for high-inflation than for moderate- and low-inflation countries.
  - Impact is higher for developing countries than for industrial countries.
  - Institutions such as economic freedom and democracy are important determinants of inflation; higher economic freedom and democracy are associated with lower inflation.

### II. POLITICAL INSTABILITY, INSTITUTIONS, AND INFLATION
- Theoretical and empirical background:
  - Structural features of an economy determine government tax-collection capacity; poorer countries may rely more on the inflation tax (seigniorage).
  - Theory of Optimal Taxation (Phelps, 1973; Végh, 1989; Aizenman, 1992) posits governments equate marginal cost of inflation tax with output taxes; evidence for developing countries is mixed.
  - Cukierman, Edwards, and Tabellini (1992) model: political instability and polarization determine tax system efficiency and seigniorage; higher instability/polarization → higher seigniorage.
- Empirical measures and improvements:
  - This paper employs direct measures of political instability: counts of government crises and cabinet changes in a year (CNTS), rather than probit-derived likelihood measures.
  - Democracy measured with the Polity Scale (range -10 to +10) rather than a dummy for democratic regimes.
- Mechanisms linking political instability and inflation:
  - Frequent cabinet changes/government crises shorten policymakers’ horizons, increasing emphasis on short-term objectives and reducing ability to maintain low inflation.
  - Inflation can increase political instability via its costs and the responsibility hypothesis (voters hold governments responsible for economic outcomes).
- Institutions and inflation:
  - Weak institutions (e.g., inadequate contract enforcement, weak property rights) are associated with poor macroeconomic performance and may lead to higher inflation due to reliance on seigniorage.

### III. DATA AND EMPIRICAL MODEL
- Dataset composition:
  - Annual political, institutional, and economic variables for 178 countries for years 1960–99.
  - Due to missing values, estimation samples reduce to at most 97 countries for inflation and 66 for seigniorage.
- Data sources:
  - Political/institutional: Cross National Time Series Data Archive (CNTS); Database of Political Institutions (DPI 3.0); Polity IV; Gwartney and Lawson (2002); Freedom House.
  - Economic: World Bank’s World Development Indicators (WDI); Global Development Network Growth Database (GDN); IMF’s International Financial Statistics (IFS); Penn World Tables (PWT 6.1); OECD Statistical Compendium.
- Dependent variable and transformations:
  - Dependent variable: logarithm of inflation (IFS); high variability → log used.
  - Empirical models estimate dynamic panel data for annual inflation levels; first-difference of Log(Inflation) used in reported estimations.
- Explanatory variables (hypothesized determinants):
  - Lagged logarithm of inflation (IFS).
  - Political instability and institutions:
    - Government crises (CNTS): counts number of rapidly developing situations in a year that threaten downfall of the regime, excluding revolt aimed at overthrow.
    - Cabinet changes (CNTS): counts times a new premier is named and/or 50 percent of cabinet posts occupied by new ministers.
    - Index of economic freedom (Gwartney and Lawson, 2002).
    - Polity scale (Polity IV): -10 (strongly autocratic) to +10 (strongly democratic).
  - Economic structural variables:
    - Agriculture (percent of GDP): share of value added of agriculture in GDP (WDI).
    - Trade (percent of GDP): openness to trade (WDI).
  - Economic performance and external shocks:
    - Growth of real GDP per capita (PWT 6.1).
    - Real overvaluation (GDN).
    - Growth of oil prices (OECD): percentage annual change in oil prices.
    - U.S. treasury bill rate (IFS): proxy for international interest rates.
- Treatment of endogenous variables:
  - Lagged dependent variable and variables potentially affected by inflation (cabinet changes, government crises, growth of real GDP per capita, real overvaluation) treated as endogenous.
  - System-GMM (Blundell and Bond) estimation used to exploit additional moment conditions and address persistence.

### IV. EMPIRICAL RESULTS
- Estimation approach:
  - System-GMM for linear dynamic panel-data models; dependent variable is D1 Log(Inflation) and explanatory variables in first differences.
  - Two-step results using robust standard errors corrected with Windmeijer’s (2000) correction.
  - Sargan tests never reject validity of over-identifying restrictions.
  - Significance notation: a=1 percent; b=5 percent; c=10 percent.
- Model specifications reported (Columns 1–5):
  - Column 1: All explanatory variables included.
  - Column 2: Index of economic freedom excluded due to high correlation with polity scale, agriculture, and trade.
  - Column 3: Agriculture excluded because not statistically significant.
  - Column 4: Cabinet changes used instead of government crises.
  - Column 5: Column 3 specification applied only to developing countries.
- Key coefficient estimates and statistical significance (selected):
  - Lagged first-difference of Log(Inflation) (LD1):
    - Column 1: 0.644 (t-statistic (16.7)a)
    - Column 2: 0.653 (18.8)a
    - Column 3: 0.672 (20.9)a
    - Column 4: 0.711 (23.9)a
    - Column 5: 0.639 (16.4)a
  - Government crises D1:
    - Column 1: 0.139 (1.85)c
    - Column 2: 0.247 (2.33)b
    - Column 3: 0.161 (2.46)b
    - Column 5: 0.202 (2.34)b
  - Cabinet changes D1:
    - Column 4: 0.091 (1.98)c
  - Index of economic freedom D1:
    - Column 1: -0.249 (-6.18)a
  - Polity scale D1:
    - Column 1: 0.002 (-0.58)
    - Column 2: -0.008 (-2.96)a
    - Column 3: -0.007 (-2.36)b
    - Column 4: -0.005 (-2.27)b
    - Column 5: 0.005 (-1.34)
  - Agriculture (in percent of GDP) D1:
    - Column 1: -0.003 (-0.94)
    - Column 2: 0.001 (-0.58)
  - Trade (in percent of GDP) D1:
    - Column 1: -0.00004 (-0.07)
    - Column 2: -0.001 (-2.26)b
    - Column 3: -0.001 (-2.71)a
    - Column 4: -0.001 (-3.40)b
    - Column 5: -0.002 (-2.73)a
  - Growth of real GDP per capita D1:
    - Column 1: -0.022 (-3.22)a
    - Column 2: -0.037 (-4.78)a
    - Column 3: -0.033 (-4.92)a
    - Column 4: -0.02 (-3.80)a
    - Column 5: -0.032 (-4.61)a
  - Real overvaluation D1:
    - Column 1: -0.002 (-2.41)b
    - Column 2: -0.002 (-1.93)c
    - Column 3: -0.002 (-2.21)b
    - Column 4: -0.002 (-2.52)a
    - Column 5: -0.001 (-1.53)
  - Growth of oil prices D1:
    - Column 1: 0.004 (6.03)a
    - Column 2: 0.004 (6.05)a
    - Column 3: 0.004 (6.59)a
    - Column 4: 0.004 (6.64)a
    - Column 5: 0.003 (4.17)a
  - U.S. treasury bill rate D1:
    - Column 1: 0.016 (2.39)b
    - Column 2: 0.033 (4.78)a
    - Column 3: 0.034 (5.62)a
    - Column 4: 0.028 (4.80)a
    - Column 5: 0.038 (4.00)a
- Sample sizes and coverage:
  - Number of observations:
    - Column 1: 1,703
    - Column 2: 2,225
    - Column 3: 2,629
    - Column 4: 2,630
    - Column 5: 1,877
  - Number of countries:
    - Column 1: 89
    - Column 2: 95
    - Column 3: 97
    - Column 4: 97
    - Column 5: 75
- Quantified effect statements reported in text:
  - "An additional government crisis increases the inflation rate by 16.1 percent (Column 3)."
  - A cabinet change leads to an increase of (text cut off in provided content at that point).

*Source: System-GMM estimations for dynamic panel-data models (using Stata 8.2); results and notes as presented in the source content.*

### 9.1 percent (Column 4). The effect of a government crisis is even higher when only

### _wp0549 - 9.1 percent (Column 4). The effect of a government crisis is even higher when only

### Major empirical findings on inflation
- A government crisis increases inflation by 9.1 percent (Column 4).
- When only developing countries are considered, a government crisis increases inflation by 20.2 percent (Column 5).
- A move of one point up the economic freedom scale (towards greater freedom) reduces the inflation rate by roughly 25 percent (Column 1).
- Democracy (an additional point in the polity scale) reduces the inflation rate by only 0.5 percent to 0.7 percent (Columns 3 and 4); this variable is not statistically significant when only developing countries are considered (Column 5).
- Trade (in percent of GDP), real overvaluation, and the growth of oil prices have relatively small impacts on inflation but with expected signs:
  - Greater openness to trade decreases inflation.
  - Real overvaluation of the currency decreases inflation.
  - Higher oil prices increase inflation.
- Marginal effects of other economic variables:
  - When the U.S. Treasury Bill rate goes up by one percentage point, the inflation rate increases by roughly 3 percent.
  - When the growth rate of real GDP per capita is one point higher, inflation decreases by at least 2 percent.
- Example quantification: if the inflation rate is at its sample mean of 51.98 percent, a government crisis will push it to 60.35 percent, i.e., inflation increases by 8.37 percentage points.

### Interaction results: high-inflation episodes and country groups
- Interactions of political instability variables with inflation regimes:
  - Interactions with inflation < 50 percent are not statistically significant.
  - Interactions with inflation ≥50 are highly statistically significant and positive.
- Magnitudes in high-inflation contexts:
  - When inflation is high or very high, an additional government crisis increases inflation by 84.5 percent (see Column 1).
  - An additional cabinet change increases inflation by 97.3 percent (see Column 3).
- Country-group interactions:
  - Interactions with industrial countries are not statistically significant.
  - Interactions with developing countries are statistically significant.
  - Conclusion: the positive relationship between political instability and inflation holds essentially for developing countries.

### Seigniorage: determinants and political instability
- Seigniorage definition used: ratio of the change in reserve money (IFS, line 14) to total government revenues (IFS, line 81).
- Main results (Table 3):
  - Cabinet changes (-1) coefficient: .048 (t-statistic (2.48)b) — cabinet changes lead to higher seigniorage.
  - Government crises (-1) coefficient: .048 (t-statistic (1.87)c) — government crises lead to higher seigniorage.
  - Interaction effects:
    - [Cabinet changes (inflation ≥ 50 percent)] (-1): .375 (t-statistic (2.41)b) — effect significant when inflation is high.
    - [Government crises (inflation ≥ 50 percent)] (-1): .568 (t-statistic (2.47)b) — effect significant when inflation is high.
    - [Cabinet changes (developing countries)] (-1): .068 (t-statistic (2.53)b).
    - [Government crises (developing countries)] (-1): .094 (t-statistic (1.91)c).
    - Interactions for industrial countries are not significant (e.g., [Cabinet changes (industrial countries)] (-1): -.003 (-.58)).
  - Democracy (polity scale) is not statistically significant for seigniorage in these specifications.
- Other economic correlates of higher seigniorage:
  - A larger agricultural sector (Agriculture (in percent of GDP)) — coefficient .015 with strong significance (e.g., (3.32)a).
  - Real overvaluation (-1) — coefficient .0002 with strong significance (e.g., (4.43)a).
  - Lower oil prices (growth of oil prices negative coefficients, marginal significance).
  - Higher treasury bill rates — coefficients around .007 to .008 with modest significance (e.g., (1.93)c to (2.35)b).
- Robustness and model specifics:
  - Fixed-effects (within-groups) estimator used for seigniorage models; Hausmann tests indicate fixed effects preferable to random effects or OLS.
  - Number of observations across Table 3 specifications ranges from 1,520 to 1,534; number of countries equals 65–66; Adjusted R2 ranges from .23 to .28.

### Mechanisms, robustness, and additional tests
- Mechanism: excessive money growth leading to high inflation is generally caused by attempts to extract large seigniorage revenues; more unstable and polarized political systems have more inefficient tax structures and rely more on seigniorage.
- Robustness tests (for inflation models) mentioned:
  - Adding variables or replacing some variables by reasonable alternatives produced similar results.
  - Variables that, when included, lead to lower inflation: greater executive constraints, more political rights, more civil liberties, and higher real GDP growth (when used instead of growth of real GDP per capita).
  - Proxies not statistically significant: ideological polarization (in some tests), urbanization, currency inside banks, GDP growth of main trading partners, exchange rate regime, and central bank independence.
  - Additional robustness checks: inclusion of year/decade or region dummies; alternative samples excluding extreme inflation values (annual rates above 1000 percent) and excluding Latin America — results remained very similar.
- Additional finding on volatility (not shown in tables but reported):
  - Inflation becomes more volatile at higher levels and greater political instability, less economic freedom, greater ideological polarization, and greater fragmentation of parties’ shares in parliament lead to higher inflation volatility.

### Policy implications and conclusions
- Main conclusion: higher political instability (measured by cabinet changes, government crises, and related institutional indicators) generates higher inflation rates and higher seigniorage.
- The effects are more pervasive and stronger in developing countries and in high-inflation (above 50 percent) countries than in developed and low-inflation countries.
- Policy recommendations:
  - Reforms aimed at reducing political instability and increasing economic freedom and democracy would help reduce inflation.
  - Inflation-stabilization efforts may be only temporarily effective if they do not include serious fiscal and political reforms.
  - Policymakers in developing countries should reform institutions and create viable mechanisms conducive to long-run price stability.

*Source: _wp0549 - 9.1 percent (Column 4). The effect of a government crisis is even higher when only (PDF chapter/section).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2005/_wp0549.pdf_