Fitting and Forecasting Sovereign Defaults using Multiple Risk Signals

Date01 February 2015
DOIhttp://doi.org/10.1111/obes.12052
Published date01 February 2015
66
©2013 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 77, 1 (2015) 0305–9049
doi: 10.1111/obes.12052
Fitting and Forecasting Sovereign Defaults using
Multiple Risk Signals*
Roberto Savona† and Marika Vezzoli
Department of Economics and Management, University of Brescia, c/da S. Chiara 50, 25122,
Brescia,Italy (e-mail: savona@eco.unibs.it)
Department of Molecular and Translational Medicine, University of Brescia, Viale Europa 11,
25123,Brescia, Italy (e-mail: marika.vezzoli@med.unibs.it)
Abstract
In this article, we try to realize the best compromise between in-sample goodness of fit and
out-of-sample predictability of sovereign defaults. To do this, we use a new regression-
tree based approach that signals impending sovereign debt crises whenever pre-selected
indicators exceed specific thresholds. Using data from emerging markets and Greece,
Ireland, Portugal and Spain (GIPS) over the period 1975–2010, we show that our model
significantly outperforms existing competing approaches (logit, stepwise logit, noise-
to-signal ratio and regression trees), while balancing in- and out-of-sample performance.
Our results indicate that illiquidity (high short-term debt to reserves) and default history,
together with real GDP growth and US interest rates, are the main determinants of both
emerging market country defaults and the recent European sovereign debt crisis.
I. Introduction
The recent sovereigndebt crisis in the Eurozone revived the debate on ‘forecasting vs. policy
dilemma’ introduced in Clements and Hendry (1998) and on the gap between models used
for forecasting and models used for policy-making. Abundant empirical evidence proves
that simple models are usually better than complexmodels in ter ms of forecast accuracy,but
the latter provide a better description of past data. How should we combine the in-sample
goodness of fit and out-of-sample predictability in the context of sovereign default? How
should we evaluate model performance when jointly considering in- and out-of sample
accuracy? Our objective is to givean answer to these questions by inspecting the sovereign
defaults in emerging markets occurring between 1975 and 2010, and the recent Eurozone
sovereign debt crisis.
The questions we face in this article have achieved new relevance given the recent
global financial crisis for different decision-maker categories. International investors, who
*The authors are grateful to the Editor, Christopher Bowdler and two anonymous referees for comments and
suggestions which substantially improvedthe article. The research leading to these results has received funding from
the European Union Seventh FrameworkProgramme (FP7/2007-2013) under grant agreement no. 320270 - SYRTO.
JEL Classification numbers: C53, F33, F34.
Fitting and forecasting sovereign defaults 67
are generally more focused on pure forecasting (i.e. expectation of risk/return profile),
are showing different risk tolerance levels (from low to high risk aversion) depending
on the increased sensitivity towards macroeconomic conditions after the Greek crisis.1
Policy makers are concerned with realizing optimal early warning systems (EWSes) to
provide risk signals with a sufficient lead time to implement adequate policy measures. In
this perspective, first, it is preliminarily essential that stylized facts on crisis occurrence
are well established based on past data; second, the EWSes should be conceived with
the main objective of minimizing false alarms (type-II errors) while maintaining a high
predictive ability of impending crises, rather than with the objective of controlling for
missing defaults (type-I errors). The costs associated with false alarms are in fact potentially
huge in terms of negative market sentiment, international reputation, contagion effects and
political interventions, which translates into a great concern towards type-II errors.
The literature on sovereign defaults is extensive in terms of early warning indicators
and model specification. On the selection of best crisis predictors, the empirical evidence
suggests that the probability of a debt crisis is positively correlated with higher levels of
total (McFadden et al., 1985) and short-term debt (Detragiache and Spilimbergo, 2001),
negativelycor related with GDP growth (Sturzenegger, 2004), and the level of international
reserves (Dooley,2000). Moreover, defaults are also related to more volatile and persistent
output fluctuations (Cat˜ao and Sutton, 2002), less trade openness (Cavallo and Frankel,
2008), political conditions (Manasse, Roubini and Schimmelpfennig, 2003), previous his-
tory of defaults (Reinhart, Rogoff and Savastano, 2003) and contagion (Eichengreen, Rose
and Wyplosz, 1996). Taken together, these articles contribute to our understanding of
potential predictors of debt crises, which in turn, can be classified as follows: (i) insol-
vency risk, which includes capital and current account variables (international reserves,
capital flows, short-term capital flows, foreign direct investment,real exchange rate, current
account balance and trade openness) and debt variables (public foreign debt, total foreign
debt, short-term foreign debt and foreign aid); (ii) illiquidity risk , proxied by liquidity
variables (short-term debt to reserves, debt service relative to reserves and/or exports,
M2 to reserves); (iii) macroeconomic risk, measured by macroeconomic variables (real
GDP growth, inflation rate, exchange rate overvaluation, and international interest rates);
(iv) political risk, measured by institutional/structural factors (international capital mar-
ket openness, financial liberalization, degree of political instability, political rights and
default history;2(v) systemic risk, namely the contagion variable usually proxied by the
number/proportion of other debt crises3while focusing on the geographical localization
of the countries.4
As for model specification, different approaches have been explored based on the
philosophical assumptions about the nature of sovereign default. One approach is based
1De Grauwe and Ji (2012) found evidence that a largepar t of the surgein the government bond spreads of Greece,
Ireland, Portugal and Spain (GIPS) during 2010–11 was a result of negative market sentiments that have become
very strong since the end of 2010.
2In this perspective, default history assumes a signalling role about the credibility of a sovereign to meet creditor
needs, and this is coherent with the debt intolerance view introduced in Reinhart et al. (2003).
3This definition is in line with Eichengreen et al. (1996) who define contagion as a case where knowing that
there is a crisis elsewhere increases the probability of a crisis at home, even after taking into account a country’s
fundamentals.
4The prevalent literature assumes that contagion is regionally-based (Kaminskyand Reinhar t, 2000).
©2013 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT