Predicting crisis : this time is different (?)

This thesis aims to predict the currency, banking and debt crises and more specifically investigates the effect of the housing sector on the crisis prediction. This study is not only constructing a crisis prediction method, which uses the previous literatures data set, but also proposing a new one including the housing market data, and comparing the performances of the two in order to measure the impact of the housing market on the prediction power. The data are taken from World Bank, IMF, OECD and Eurostat and cover the years between 1999 and 2010. Multinomial logistic regression is used for crisis prediction. As an advantage, it prevents the ‘post-crisis bias’ problem; by this way the robustness of the analysis is also improved. The multinomial logistic regression is run for two different time windows ‘t-1, t, t+1’ and ‘t, t+1, t+2’ as t denoting the current year. By inclusion of the housing market data, the prediction power is increased from 60% to 100% in the case of ‘t-1, t, t+1’. The model sends no false alarms in this case. For the case of ‘t, t+1, t+2’, the in-sample prediction power is improved from 68% to 95% and the false alarm ratio is reduced from 6.6% to 3%. The-out-of-sample predictive performance of the system in ‘t, t+1, t+2’ is improved from 33% to 60% by the inclusion of the housing market data. Due to the restrictions of the data set, out-of-sample analysis could not be performed for the case of ‘t-1, t, t+1’. The proposed crisis prediction method succeeds in predicting the crises of 2000s by using housing sector data. The impact of the housing sector in predicting crisis is clearly shown. Also, it is shown that chosen time window for the multinomial logistic regression in predicting crisis can lead to variations on the predicted results.