Danger Zones for Banking Crises in Emerging Markets
Published date | 01 October 2016 |
DOI | http://doi.org/10.1002/ijfe.1550 |
Date | 01 October 2016 |
DANGER ZONES FOR BANKING CRISES IN EMERGING MARKETS
†
PAOLO MANASSE
a,
*, ROBERTO SAVONA
b
and MARIKA VEZZOLI
c
a
Economics Department, University of Bologna, Piazza Scaravilli, 40100 Bologna, Italy
b
Department of Economics and Management, University of Brescia, C/da S. Chiara 50, 25122 Brescia, Italy
c
Department of Molecular and Translational Medicine, University of Brescia, VialeEuropa 11, 25123 Brescia, Italy
ABSTRACT
This paper employs a recently developed statistical algorithm in order to build an early warning model for banking crises in
emerging markets. The procedure creates many ‘artificial’samples by iteratively perturbing the original data set and estimates
many models from these samples. The final model is constructed by aggregation, so that, by construction, it is flexible enough to
accommodate new data for out-of-sample prediction. Out of a large number (540) of candidate explanatory variables, ranging
from macroeconomic variables to balance sheet indicators, our procedure selects a handful of indicators (and their combina-
tions) that is sufficient to generate accurate out-of-sample predictions of banking crises. Using data covering emerging markets
from 1980 to 2010, the model identifies two banking crisis’‘danger-zones’, e.g. economic configurations that are conducive to
crises. The first occurs when high interest rates on bank deposits, possibly reflecting liquidity risks and solvency fears, interact
with credit-booms and capital flights; the second occurs when an investment boom is financed by a large rise in banks’net for-
eign exposure. We compare our model to models derived by standard econometric techniques, and find that our approach
delivers much better out-of-sample predictions. Copyright © 2016 John Wiley & Sons, Ltd.
Received 27 November 2014; Revised 02 November 2015; Accepted 18 February 2016
JEL CODE: E44; G01; G21
KEY WORDS: banking crises; early warnings; regression and classification trees; CRAGGING; stepwise logit
1. INTRODUCTION
The recent wave of banking crises stemming from developed countries, the sub-prime-Lehman crisis in the US and
private/public debt crises in Europe, has spurred a renewed interest in the quest for ‘early warning’indicators. In
the Europe, banks’exposure to collapsing real estate prices has threatened the joint solvency of the banking sector
as well as of the sovereign, as in the classical ‘twin’paradigm described by Kaminsky and Reinhart (1999) and
experienced in many episodes of the past three decades in emerging markets (the ‘Asian crisis’comes to mind).
As in the past, the new wave of crises took most international financial institutions, rating agencies and academics,
by surprise. The costs of the bailouts, their implication for government finances and sovereign solvency are so large
that the failure to anticipate such events and to respond adequately has proved the most serious threat to the very
existence of the Euro area.
The literature on early warnings for banking crisis, which we will review in the following section, presents a
limited degree of consensus. For example, the role of lending booms as predictors of crises is generally accepted
and, similarly, financial institutions’leverage and balance sheets interdepence are acknowledged as risk factors.
Yet the empirical literature is, with one exception, silent on four crucial issues. The first concerns the ranking of
indicators: which variable is the ‘most important’indicator, in the sense of being associated to the largest
*Correspondence to: Paolo Manasse, Economics Department, University of Bologna; Piazza Scaravilli, 40100 Bologna, Italy.
E-mail: paolo.manasse@unibo.it
†
This project has received funding from the European Union’s Seventh Framework Programme (FP7-SSH/2007-2013) for research, technolog-
ical development and demonstration under grant agreement no 320270 –SYRTO.
Copyright © 2016 John Wiley & Sons, Ltd.
International Journal of Finance & Economics
Int. J. Fin. Econ. 21: 360–381 (2016)
Published online 4 April 2016 in Wiley Online Library
(wileyonlinelibrary.com). DOI: 10.1002/ijfe.1550
conditional probability of a crisis in the following periods? The second concerns the interactions of vulnerabilities:
some indicator may be relatively ‘safe’in ‘normal times’, and yet become extremely dangerous when interacted
with other vulnerabilities in the economy. An example from the literature of currency crises (Krugman, 1979) is
that of a pegged exchange rate system in conjunction with a money-financed fiscal deficit. Thirdly, crises are a
non-linear phenomenon, and they typically show snow-ball dynamics. Think for example of bank-runs and spec-
ulative attacks, which take-off when some particular threshold is crossed. Can we identify trigger levels for early
warnings of banking crises? Finally, crisis episodes may be heterogeneous, in the sense that the combinations of
underlying vulnerabilities may differ across episodes. An investment boom financed by foreign borrowing or a
sharp recession leading to irrecoverable credits may both be early warning of naming crises. These issues are
obviously key when assessing a banking sector’s vulnerability in a country and when pre-emptive policy responses
need to be taken promptly.
The reason why most of the empirical literature on early warnings for banking crises is silent on these
features relates to the empirical methodology commonly adopted. Most contributions use simple regression
analysis. While this is a suitable technique for establishing partial correlations, it cannot address any of the
above issues. The aim of this paper is to identify the most relevant predictors of crisis and to identify their
possibly heterogeneous and non-linear effects, as well as their possibly complex interactions. To this end, we
employ a state-of-the-art technique borrowed from the statistical literature on ‘perturbation and combination’
algorithms, the CRAGGING (see Vezzoli and Stone, 2007 and more in general Breiman, 2001; Friedman and
Popescu, 2003). We develop an empirical model which summarizes the relevant information on future crises
by means of a small set of predictors, and identifies their key interactions and ‘threshold’values. Also, the
model shows great accuracy in predicting crises out-of-sample while sending few ‘false alarms’. In order to
ease the comparison with the previous literature, we apply the algorithm on the well-established data on sys-
temic banking crises in emerging markets put together by Laeven and Valencia (2013), covering the period
1980–2010.
The main findings of the paper are as follows. First, out of a large number (540) of candidate predictors, we
show that banking crises can be accurately predicted by just a handful of variables: the interest rate on deposits,
the net accumulation of foreign liabilities of the banking sector, the (change in the ratio of) domestic credit to
GDP, the ratio of investment to GDP. Second, not all banking crises in emerging markets are equal: our early
warning model identifies two main ‘danger zones’. The first, and most dangerous, derives from the combina-
tion of high interest rates on deposits together with capital flights (a rise in banks’investment in (net) foreign
assets) and with a credit boom (acceleration in bank lending). The conditional probability of a crisis following
this configuration is above 40%. This is the case of a few Latin American banking crises. A different type of
danger zone is found when a boom in real investment is financed by banks’foreign borrowing, when domestic
banks already have large foreign debt. This situation results in a crisis with a probability of 37%, and it is
exemplified by a few Southern Asian countries in crisis starting in 1997. Finally, when we ran a ‘horse race’
exercise with competitor models, our model clearly dominates the others in terms of out-of-sample predictive
power.
The plan of the paper is as follows. Section 2 reviews the empirical literature on early warnings for bank ing cri-
sis; Section 3 describes the data set, and Section 4 presents the methodology. Section 5 discusses the estimated
model and presents three alternative competitor models built on the same data set. Section 6 compares the proper-
ties of the different models and Section 7 concludes.
2. EARLY WARNINGS FOR BANKING CRISIS: A REVIEW OF THE LITERATURE
The empirical literature on Early Warnings has applied a variety of empirical methodologies to predict sov-
ereign, exchange rate and banking crises, mainly in emerging markets. These methodologies include paramet-
ric (e.g. Logit, Probit regression), and non-parametric models (e.g. mainly the Signal-to-Noise approach).
Among the papers employing the former methodology, Eichengreen and Rose (1998) use multivariate Probit
regression and find that macroeconomic variables such as the interest rates and GDP growth rates of OECD
countries have a significant and large effect on bank fragility in developing countries. In a similar vein,
DANGER ZONES FOR BANKING CRISES 361
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Fin. Econ. 21: 360–381 (2016)
DOI: 10.1002/ijfe
To continue reading
Request your trial