Do banking crises improve democracy?

The baseline results are also obtained without fixed effects or without controlling for the lagged value of the dependent variable. The results of those regressions are reported in the online appendix.

We test the robustness of our results when using another measure of democracy in the robustness check section. All dependent variables are described in Table 9 in the "Appendix". Descriptive statistics are reported in Table 10 in the “Appendix”.

We use the polity2 variable with a slight modification. The database assigns a zero score to the index in periods where politics cannot exercise effective authority over at least half of their territory due to foreign intervention. These are the so-called interregnum periods. For our analysis, we treat these observations as missing. The reason is that since we are interested in the evolution of democracy, we want to avoid misinterpretation of improvement or deterioration relative to preceding years. For instance, suppose a country has a negative score of democracy in year t − 1. Then, if that country experiences a foreign intervention in year t, we will interpret it as an improvement in democracy according to the initial database. The reverse holds for a country that has a positive score of democracy in year t − 1 and a foreign intervention in year t. We thus decide to treat observations related to interregnum periods as missing values. This results in the deletion of 70 out of 4680 observations in our sample.

We have to stress that considering a larger T requires further hypotheses about the realization of systemic banking crises over the period before 1970.

Figures B3 and B4 in the online appendix report the evolution of the Polity2 index for each country in the sample, respectively for countries that experienced a crisis and those that did not.

Table 11 in the “Appendix” reports the detailed classification of the countries that experienced a banking crisis.

To assess the real effect of banking crises, we estimated our baseline equation using successively the logarithm of real GDP then the Gini coefficient as dependent variables. The results of those estimations are reported in Table 12. The results show that banking crises are followed by an increase in inequality, in line with de Haan and Sturm (2017), and that the effect is statistically significant at the ten-percent level. Conversely, over the baseline ten-year study window, the coefficient of the banking crises dummy on GDP per capita is statistically insignificant. However, when estimating an impulse-response function we observe that the growth rate of GDP decreases in the 2 years after the onset of a crisis. The effect is statistically significant beyond the five-percent level. GDP growth becomes statistically indistinguishable from its pre-crisis level in the third year following the onset of the crisis. Those estimations and results are reported and discussed in the online appendix.

The result is obtained by applying the following formula of a geometric series: \(\beta \frac{{1 - \gamma^{j + 1} }}{1 - \gamma }\) where γ is the coefficient of the lagged dependent variable and β is the coefficient of banking crises.

On the history of Indonesia, one may refer to Ricklefs (2008).

On the transition in Mali, one may refer to Villalón and Idrissa (2005).

On Tanzania over the period, one may refer to Tripp (2000).

Out of the 97 banking crises in our sample, 31 took place in an autocracy, 29 in an anocracy, and 37 in a democracy. Table 13 in the “Appendix” reports the distribution by political regimes of the countries that experienced a banking crisis. Table 14 in the “Appendix” displays the democratic transitions by highlighting those preceded by severe banking crises.

We use a mix of geographic and economic linkages in the sense that for European countries, we distinguish the ones that are member of the Eurozone from the others. Likewise, for Sub Saharan African countries, we distinguish between West Africa, East Africa and the others.

In fact, Mendoza and Terrones (2008) show that though capital flow bonanzas are associated with lending boom, not all lending booms end up as financial crises.

We estimate the first-stage regression using the linear probability model using lagged values of both instruments.