Early Warning Indicator Model of Financial Developments Using an Ordered Logit – Conclusions

Early Warning Indicator Model of Financial Developments Using an Ordered Logit - ConclusionsFor central banks it is important to use early warning indicators to assess the possible implications of large asset price movements and the building up of financial imbalances in the economy. In this respect, several studies have shown that the analysis of credit developments may be very useful in this respect. This paper contributes to this literature by splitting up the development of a financial indicator into bust, normal and boom phases to describe the whole boom-bust-cycle. For these cycles it is investigated whether credit indicators can play an important role in detecting the stylized development of asset prices by looking at the evidence stemming from a sample of 17 OECD industrialised countries and the euro area over the period 1969 Q1 – 2011 Q2.
By using an asset price composite indicator (which incorporates developments in both the stock price and house price markets) and following the methodology illustrated in Gerdesmeier, Reimers and Roffia (2010, 2011), an empirical analysis is carried out based on a pooled ordered logit-type approach, which considers several economic variables. According to statistical tests, credit aggregates (growth gap), and house price changes (and growth gap) jointly with developments in stock prices and the interest rate spread turn out to be the best indicators which help to forecast composite indicator development. These results demand, for example, central banks to observe the development of credit aggregates as well as stock and house markets to get signals of mispricings at these markets. Moreover, the term spread shows a channel to affect the probability of boom and bust phases. Hence, it indicates that an monetary authority can lend against the wind to avoid devastating effects on the real economy caused by boom-busts-cycles.
The model is cross-checked vis-a-vis the estimation methods, forecasting horizon and probability thresholds and it turns out to be quite robust. These results reflect the good performance to estimate the normal situation. To capture the extreme cases, which are turning points of the development, is much more difficult. These phases are relatively short and show a low persistence. To be more successful for the boom and bust phases an extension of the data base could be promising. In this sense for further research higher frequency or mixed frequency approaches might be fruitful venues.