ABSTRACT
What type of crisis is generated when debt increases? We extend the literature by framework by introducing currency and stock market crises in the analysis. We apply our proposal to the case of Spain, since this is a country that has experienced a very important amount of financial crises from the nineteenth century onwards. We find the same results as the previous literature for the determinants of banking and debt crises but substituting external and public debt with perpetual debt and where perpetual debt has a less important role than crises in the private sector. Moreover, we find evidence in favour of the hypothesis that currency crises depend strongly and positively on financial centre crises and negatively and mildly on perpetual debt. We justify the negative relalionship due to an inflation tax. We also find evidence in favour of the hypothesis that stock market crises depend only positively and strongly on financial centre crises.
Acknowledgement
We wish to thank two referees and the co-editor for very helpful comments.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 See, for example, how Calomiris (Citation2010) examines the long-term record of banking crises in and outside the U.S., in a historical context.
2 Note also that recently in the literature (see, for example, Bordo, Redish, and Rockoff Citation2015), we have learned that the financial system is path-dependent and if we want to study for example banking crises, we need to take a look at data sets for a long period of years.
3 See Wooldridge (Citation2009) and Wang, Iglesias, and Wooldridge (Citation2013) as examples of the complexity that is added when response probabilities that are suitable for binary variables are introduced into different types of models.
4 See also Sudrià (Citation2014).
5 Note that the notion of twin-crises is well established in the literature (see also how even Bertrán et al. (Citation2012) mention the existence of triple crises occurring jointly in Spain), while we look at each type of crisis separately. This is also an important limitation of our empirical analysis, given the complexity that we would have to introduce in the econometric methodology in case we would allow for the existence of crises occurring jointly.
6 See also Comin (Citation2012), where it reads that sovereign defaults in Spain are during the years 1809, 1820, 1831, 1834, 1851, 1867, 1872, 1882, 1936–1939.
7 In order to create our dummy variables, 1 is related to the existence of a crisis. As an example of summary statistics of our data set, in the period 1850–1995 we record a mean of 0.079 for banking crises, 0.064 for currency crises, 0.093 for stock market crises and 0.021 for debt crises. 0.086 is the mean for crises of financial centres. Therefore, stock market crises are the ones that happen more frequently. We also record a 5% to be the probability of observing more than one crisis type in a given year. Moreover, if we obtain correlations in the data, it is similar between banking and currency and banking and debt crises (of 0.14) while it is a bit higher (0.27) between banking and stock market crises. Currency and debt and stock market and debt crises have a similar correlation of 0.16 and 0.12, respectively. Finally, the highest correlation is between currency and stock market crises (0.42).
8 See also Reinhart and Sbrancia (Citation2011) for financial repression in U.S. and the U.K. following WW2, and Reinhart (Citation2012) debt reduction in Italy and Australia from the financial repression ‘tax’.
9 See page 309, section 2.4 and the results of their Table 2 in page 315.
10 See Reinhart and Rogoff (Citation2009): 76, 101 in relation to these types of policies at the international level. See also Juselius and Toro (Citation2005), Sabaté, Gadea, and Escario (Citation2006) and Escario, Sabaté, and Gadea (Citation2011) for Spain.
11 All univariate models have been estimated and tested in STATA (Citation2013).
12 Therefore, we have showed evidence that among the different responses that the government can have on debt (using perpetual debt, treasury debt, etc.), the one that affects the existence of crises is perpetual debt.
13 Recent developments of econometric packages such as the estimation of bivariate Logit models with the R program ZeligChoice by Lau, Imai, and King (Citation2015) and multivariate Probit models estimated by maximum likelihood with the R program mvProbit of Henningsen (Citation2015) have allowed us to carry out our empirical results. Bivariate probit models have also been estimated and tested in STATA (Citation2013).
14 For a thorough introduction to BMA, consult Hoeting et al. (Citation1999). We have used the BMS package for R of Zeugner and Feldkircher (Citation2015) to obtain our BMA results.