The exploration of economic crises: parameter uncertainty and predictive ability
| Published date | 01 May 2019 |
| DOI | http://doi.org/10.1111/sjpe.12199 |
| Author | John Nkwoma Inekwe |
| Date | 01 May 2019 |
THE EXPLORATION OF ECONOMIC
CRISES: PARAMETER UNCERTAINTY
AND PREDICTIVE ABILITY
John Nkwoma Inekwe*
ABSTRACT
Over a long horizon, this paper examines joint economic crises and determines
the power of 49 variables in predicting such episodes. While incorporating dyna-
mism in the prediction, we generate the predictive power of various specifications
and model the uncertainty in the parameters of interest. The results reveal that
growth of real gross domestic product per capita, regulation, bank non-perform-
ing loans, interest rate and inflation rate are the most significant variables in
predicting the joint economic crises. These variables predict economic crises with
about 93% accuracy and can predict joint economic crises in developing coun-
tries and recent joint crises.
II
NTRODUCTION
The question of the nature of social, economic, political and financial dynam-
ics that makes a country less prone to certain types of crises has occupied a
central position in financial and economic stability studies. The quest to
understand the factors that generate turbulence in the entire economy has
been sustained by the plethora of financial crises across decades. Researchers
and policy-makers most commonly point to lax regulation and oversight,
excessive risk taken by financial institutions, risky financial innovations, inter-
dependencies in financial and real assets, credit and capital inflows as the
causes of financial crises (see for example Gorton (1988); Kaminsky and Rein-
hart (1999); Rajan (2010) and Laeven and Valencia (2013).
1
The extensive literature on financial crises has employed either signal
approach (see Kaminsky and Reinhart, 1999), a dummy dependent variable
(see Demirg€
uc
ß-Kunt and Detragiache, 1998) or a continuous proxy for crises
such as arrears or credit spreads in the case of sovereign default (see Chakra-
barti and Zeaiter, 2014). The signal approach aims to contrast the behaviour
of economic determinants before and after a crisis and to identify the best
*Macquarie University NSW
1
A micro level study of the relevance of financial and economic variables in firm default is
given in Bottazzi et al. (2011) while a survey of studies on financial crises is provided in
Bordo and Meissner (2016) and Jeffrey Frankel and Saravelos (2012).
Scottish Journal of Political Economy, DOI: 10.1111/sjpe.12199, Vol. 66, No. 2, May 2019
©2018 Scottish Economic Society.
290
indicator that signals an imminent turmoil based on the under or overshoot-
ing of the specific threshold values. Each approach aims to underpin the sig-
nificance of each variable in predicting banking or any form of crisis.
Adapting from these studies, we model joint economic crises across 68
countries and between the periods of 1960 and 2010. Using both the limited
dependent variable approach and a linear approach, we model economic crises
that are categorized to include currency crisis 1, currency crisis 2, banking cri-
sis, stock market crisis, inflation crisis, domestic and external debt (sovereign
default) episodes as documented by Reinhart and Rogoff (2011).
2
Using this
approach, we model economic crises by employing the occurrence of these
seven episodes.
We order the events such that an economy that experiences all the episodes
in a particular year takes the value of seven while the occurrence of no epi-
sode takes the value of zero. In this case, the dependent variable takes the
value of 0 to 7. In contrast to a 0–1 dummy that is mostly used in the litera-
ture, we model joint economic crises to include various and varying episodes
of crises. Flexibility in the analysis becomes a reality as both binary and con-
tinuous dependent models can be used in the prediction of such economic
crises. However, to adhere to conventional methods, we further identify such
economic crises by using a 0–1 dummy. Using the same dataset, a country
takes the value of 1 if any of these seven episodes occurs in a year and 0
otherwise. For confirmatory analysis, we define economic crises to include
sovereign debt crisis, sovereign debt restructuring, systemic banking crisis and
currency crisis. This is retrieved from the episodes of crises as documented in
Laeven and Valencia (2013). Similar approach as documented above is fol-
lowed such that the dependent variable takes the value of 0 to 4. Thus, a
country that witnesses all events in a particular year gets the maximum value
of 4 while the occurrence of no event takes the minimum value of 0.
The prediction of financial or economic crises has remained an important
issue for most regulatory authorities, policy-makers and risk managers. In this
paper, we construct a measure of economic crises over the periods of 1960 to
2010. We contribute to the literature by modelling crises in all its forms. First,
while previous studies concentrate on one or two events, we model about
seven different episodes. We contribute to the literature by examining the pre-
dictive ability of 49 variables and by modelling the determinants of joint crises
at the univariate and the multivariate level. Our contribution is strengthened
by the use of various estimation techniques. We employ the ordered logit
models in determining the predictive ability of each variable. The strength of
these variables is examined by using the area under the receiver operating
characteristic curve (AUC) and the R
2
.
To model uncertainty about parameters, the Bayesian model average
(BMA) is employed and lastly, we specify a recursive simultaneous
2
They define currency crisis to involve currency crashes (annual depreciations of 15 per-
cent per annum or more) and currency debasement (reduction in metallic content of coins in
circulation of 5 percent or more, and a reform where a depreciated currency in circulation is
replaced by a new one).
THE EXPLORATION OF ECONOMIC CRISES 291
Scottish Journal of Political Economy
©2018 Scottish Economic Society
To continue reading
Request your trial