Analysis of Indicators of Crisis in USA

In this paper we seek to find both the best model and the best crisis variable which can reflect incidence of US crisis. Our sample consists of quarterly data from 1970Q1 to 2017Q4. We consider individual time variation of potential leading indicators and search for optimal lags for each indicator independently of each other based on information criterions. Once decided with optimal lags for each potential indicator across six crisis variables we apply Bayesian Model Averaging in order to estimate the best model and at the same time account for model uncertainty. We also use dummy crisis variables to account for the crisis events in US biggest trading partners. Finally, we compare out-of-sample performance of our models and conclude their ability to predict future crisis. The novelty of this work lays in the simultaneous search for the best model and crisis variable, which is usually subjectively assumed by other researchers. The results suggest that IRC tend to be the most precise index to catch crisis event in the US if compared to other dependent variable candidates used in this paper. Moreover, optimal model found for the IRC results in the lowest mean absolute percentage errors. Such variables as loan-to-deposit ratio, growth in average residential prices, government debt as percentage of GDP, M1, growth in fixed capital formation and money market rate occurred to be the best predictors of index of real costs. We also estimate predicted value of IRC for 2019Q3 and resulting value does not breach our 1.5σ crisis threshold.