As the term implies, an early warning system consists of economic indicators that signal in advance the onset of a crisis. This idea, however, is not without its difficulties. The development of leading indicators presumes that an economy exhibits consistent and regular patterns of behaviour prior to a crisis. It assumes that the same economic variables will get out of line ahead of an impending crisis, crisis after crisis. However, not all crises are alike, and consequently one would not expect that the same indicator would be a good signal for each type of crisis. Ideally, one would like to have several observations for the same type of crisis and test which indicators would signal each of them. This ideal has to be tempered, however, by empirical considerations. It is highly unlikely that enough observations can be gathered on crises of a similar origin or nature¹⁰ in recent times to make statistical analysis feasible. If they could, it would certainly reflect badly on the ability of policy makers to learn from mistakes.

Most of the research studies on early warning systems analyse data on crises for a cross-section of countries to identify the indicators that have historically foreshadowed crises. One could argue, though, that each country (or economy) is unique and indicators that work for some countries might not work for others. For example, in developed countries a significant proportion of the population invests in stock markets, making stock market indexes an important indicator. On the other hand, in many developing countries the stock market may be too thin and insignificant to have major implications for the economy. Again, this issue reflects an empirical constraint that instances of crises for a single country may not afford enough observations for statistical testing.

The models used for predicting crises can be classified into three types:

• A signals approach - a set of economic indicators exceeding a certain threshold constitutes a signal of an approaching crisis. Policy makers can keep track of a single indicator deemed representative of the sector being monitored (an example is the New Zealand Reserve Bank focus on the inflation rate). Instead of relying on a single catch-all indicator, a group of indicators considered reflective of various aspects of the economy may be chosen. This form is more common as the use of a number of indicators is less likely to be thrown off by "noise" than that of a single indicator. In both cases, often one cannot predict the exact timing of a crisis but can indicate stress or vulnerabilities in the system
• Probit or logit models that compute the probability of a crisis occurring based on indicators
• A regression model approach

Whichever approach is adopted, the leading indicators should ideally possess certain properties:

• Timeliness - a signal of an impending crisis, even if correct, is of limited use if it arrives after the onset of the crisis or with insufficient time left to avert it
• Low noise - the number of false signals is minimized, and a leading indicator should not fail to signal a crisis that does occur

Next

Footnotes:

¹⁰ In practice, not all crises may be put neatly into boxes that is, the delineation between one type of crisis and another may not be black and white.