Foreign booms, domestic busts: The global dimension of banking crises

Abstract

This paper provides novel empirical evidence showing that foreign financial developments are a powerful predictor of domestic banking crises. Using a new data set for 38 advanced and emerging economies over 1970–2011, we show that credit growth in the rest of the world has a large positive effect on the probability of banking crises taking place at home, even when controlling for domestic credit growth. Our results suggest that this effect is larger for financially open economies, and is consistent with transmission via cross-border capital flows and market sentiment. Direct contagion from foreign crises plays an important role, but does not account for the whole effect.

Introduction

It is well established that financial crises are often “credit booms gone bust” (Eichengreen, Mitchener, 2003, Schularick, Taylor, 2012). But this is not always the case. Why, for example, did Finland and Germany suffer financial crises, in the early 1990s and 2008 respectively, when credit growth had been subdued in both places? Why did Ireland suffer a crisis after a credit boom in 2008, but not in 2000 when credit had been growing even more quickly?

This paper documents the crucial role of global financial conditions in determining the risk of a domestic banking crisis. While it is well established that credit growth and banking crises are synchronized across countries (Laeven, Valencia, 2013, Reinhart, Rogoff, 2009, Claessens, Kose, Terrones, 2011, Mendoza, Terrones, 2014), the literature has typically focused on domestic determinants of banking crises and has singled out high domestic credit growth as the single best predictor (Schularick, Taylor, 2012, Jorda, Schularick, Taylor, 2011).

In this paper we depart from this domestic focus. Using a new data set linking credit growth and financial crises for 38 advanced and emerging economies over 1970–2011, we study the role of foreign credit growth (that is, domestic credit growth in the rest of the world) in affecting the probability of experiencing domestic banking crises. Our results provide novel empirical evidence demonstrating a systematic link between global financial conditions, as summarized by foreign credit growth, and the subsequent occurrence of domestic banking crises, conditional on domestic credit. This link improves dramatically the predictive ability of banking crises models that only rely on domestic indicators.

Our main findings are as follows. First, we provide evidence that both credit growth and the occurrence of crises are synchronized across countries. We start by showing that the empirical distribution of the number of banking crises at any one time has fatter tails than a binomial distribution, i.e. the distribution they would follow if crises were independently distributed across countries. We formally test for this correlation with a ‘stable correlation binomial model’ (Witt, 2014), a generalization of the binomial distribution that allows for a positive correlation between any two pairs of trials. Our estimates show that such correlation is positive and statistically different from zero in two different data sets, formalizing the informal notion that banking crises display a positive degree of cross-sectional dependence.

We also show that real domestic credit growth is correlated across countries, and that this synchronization has increased over time. We compute different metrics to assess the degree of international comovement of real domestic credit growth in our data set, which includes a larger set of countries relative to what has typically been considered in the previous literature. We find that a single factor (extracted with a simple principal component analysis) can explain up to 50% of the variance of countries’ real domestic credit growth in recent times; and that the average correlation between country-specific credit growth and world credit growth has increased over time.

Second, we show that global financial conditions, as summarized by foreign credit growth, can substantially increase the predictive power of models that only rely on domestic credit as an explanatory variable for the occurrence of banking crises.² Specifically, we find that foreign credit growth is a significant predictor of domestic banking crises, even when controlling for domestic credit growth. This is shown to be true for our new data set as well as for the longer, narrower panel in Schularick and Taylor (2012), which covers 14 advanced countries over the 1870–2008 period.

Third, and finally, we explore the role played by openness to international trade and financial transactions with non-residents, as well as by a number of other covariates suggested by the literature, to help distinguish between the potential economic mechanisms that drive our findings. We find that the role played by foreign credit growth is more important for financially open countries, but not for countries more open to trade. This suggests that the channel of transmission behind our findings is itself financial, rather than going through the effect of global conditions on foreign real activity and hence demand for domestic goods and services via trade.

To shed further light on the channels that mediate this effect, we explore how the inclusion of additional covariates affects our results. We demonstrate a statistically significant association between cross-border portfolio inflows and subsequent domestic banking crises, but cross-border bank lending (to either domestic banks or non-banks) does not have a significant effect. We also find that a reduction in US short-term interest rates and in global risk aversion, as proxied by a fall in the VIX index (as emphasized inter alia by Rey, Bekaert, Hoerova, 2014), an increase in the leverage of US broker-dealers (Bruno and Shin, 2015), and a compression in the level of US corporate bond spreads (Lopez-Salido et al., 2016), all portend an increased risk of a domestic banking crisis further down the line. This suggests that it is global financial conditions, of which foreign credit is a reflection, that affect domestic financial stability. Finally, we also find that the occurrence of crises abroad raises the probability of a crisis at home, but that foreign credit growth remains a robust predictor over and above this, suggesting that while contagion may play a role, it cannot be the whole story.

We interpret our evidence as suggesting that domestic financial stability is at the mercy of exogenous push shocks and broader swings in global sentiment, which can affect the probability of domestic banking crises over and above their relationship with both domestic credit growth and the realization of banking crises abroad. Global risk sentiment can be captured with variety of price- and quantity-based proxies, of which foreign credit growth is a prominent example.

Related literature. This paper is related to three broad strands of literature. First, it relates to a growing literature on the time series and cross-sectional properties of financial crises and their determinants. As such, this paper is first and foremost related to the literature on the classification and description of financial crises. See, among others, Caprio, Klingebiel, 1996, Caprio, Klingebiel, 2002; Laeven and Valencia (2013); Bordo et al. (2001); Reinhart and Rogoff (2009); Qian et al. (2011) and Gourinchas et al. (2001).

A second strand relates to the determinants of financial crises. More specifically, our work is closely related to papers that investigate whether there is systematic evidence of credit growth–induced financial instability, as motivated by theoretical work on debt-driven booms and busts, such as Fisher (1933); Minsky (1986), and Kindleberger (1978).³ In a series of recent papers, Schularick and Taylor (2012) (ST hereafter) and Jorda et al. (2011) have revived this literature using a long-run data set for advanced economies, and relying on tools from the theory of binary classification and signal detection (see Jorda and Taylor, 2011). We are most directly related to these latter papers. Relative to them we consider a shorter, but wider panel data set and, most importantly, we consider the role of foreign credit as an explanatory variable for domestic banking crises.

The third strand has investigated the link between global variables and domestic financial stability (see, for example, Frankel and Rose, 1996). Mian et al. (2015) uncover a global cycle in household debt, and show that countries with domestic cycles more closely correlated with the global one see stronger declines in output growth following a rise in the domestic household debt-GDP ratio. In two related studies, Alessi and Detken (2011) and Duca and Peltonen (2013) use global variables as early warning indicators for costly asset price boom/bust cycles and periods of financial stress (as measured by a synthetic index computed using financial markets data). In our paper we use a similar insight but relate it to the synchronicity of banking crises and apply it to the literature of binary classification and prediction of crises started with the seminal paper by ST. Moreover, relative to those studies, we expand the number of countries under consideration and/or consider a longer sample period, and we explore additional dimensions not considered by them (such as financial conditions in centre countries and, importantly, the role of countries’ financial and trade openness in influencing the effect of global variables on domestic financial stability).

The paper is structured as follows. In Section 2 we present stylized facts on the international synchronization of banking crises and domestic credit growth. In Section 3, we set out to quantify the links between domestic banking crises and foreign credit growth, exploring the relevance of financial and trade openness. In Section 4 we inspect the mechanisms behind our main results. Section 5 contains extensive robustness checks. Finally, Section 6 concludes.