Were Financial Crises Predictable?
Canova, Fabio, Journal of Money, Credit & Banking
This paper empirically investigates the nature of financial crises in the United States in the pre-World War I era in an attempt to shed light on their informational characteristics and on their generating mechanisms. In particular, the paper addresses the questions of (1) whether there are variables that reveal conditions conducive to crises, (2) whether financial crises were forecastable on the basis of the information set available to agents, and (3) whether they were alike, in the sense that a set of statistical relationships was common to all episodes.
There are several reasons to address these questions at an empirical level. First, early literature on the subject has suggested that movements in interest rates, in stock returns and in changes in the deposit to currency ratio are key factors in understanding the occurrence of crises (see, for example, Friedman and Schwartz 1963, Kindleberger 1978, and Minsky 1977). However, surprisingly little statistical work has been expended to document the significance of these variables in revealing the conditions conducive to crises. Two exceptions are Bordo (1985) and Gorton (1989). Second, although before 1914, banking panics occurred almost simultaneously with financial crises and stock market crashes, the current literature on banking panics (see, for example, Diamond and Dybvig 1983, Jacklin 1987, Waldo 1985, Chari and Jagannathan 1988, and Williamson 1989) has failed to spell out their implications for financial markets. An empirical analysis of the generation mechanism of financial crises may shed some light on these links and on the causal relationship between banking panics and financial crises [see Wilson, Sylla, and
Jones (1990) for an attempt in this direction]. Third, although crises have been described as "unsettling but not especially surprising, they were expected to happen every so often; the real question was when" (Wilson, Sylla, and Jones 1990), the question of the predictability of financial crises has never been explicitly investigated. One reason for the lack of empirical work in this area is that in most recent theories panics are generated through sequential liquidity constraints and self-fulfilling expectations. Therefore they offer little guidance on the choice of leading indicators for crises and impose very weak restrictions on the joint behavior of macroeconomic and financial variables both before and during crisis periods.
The questions of the significance of certain variables in revealing conditions conducive to crises and of the predictability of financial crises are important because past experience may help us understand the 1987 stock market crash and the recent savings-and-loans crisis and indicate whether agents could have avoided wealth losses using selected variables and simple forecasting rules.
Since the earlier literature on financial crises pointed out their tendency to occur in seasons when the money market was tight (see, for example, Jevons 1884, Palgrave Inglis 1910, Sprague 1910) or at the peak of the business cycle (see, for example, Mitchell 1913, Fisher 1933, or Minsky 1977), one forecasting device is a model including simple seasonal or cyclical indicators. This model can also provide formal evidence on the long-standing, but never formally tested belief that financial crises had a recurrent and/or quasi-periodic nature (see, for example, Miron 1986).
In addition to models including seasonal and cyclical indicators the paper examines the performance of several other theory-based forecasting specifications. Contrary to existing work in the area (see, for example, Calomiris and Gorton 1991), the paper uses out-of-sample criteria to compare the forecasting ability of various specifications and to judge the relevance of results. The conclusions of the paper are as follows: there are variables that are significant in explaining the economic conditions conducive to crises (the money supply, the volume of excess cash reserves of New York banks, stock returns, and the volatilities of stock returns and of the domestic interest rate spread). …