Online estimation for a predictive analytics platform with a financial-stability-analysis application

Abstract

An online parameter estimation via filtering recursions is constructed to support a data-analytics scheme in the predictive domain. Multivariate financial market indices or signals revealed in real time are used in our numerical implementation. This work contributes to the analysis and forecasting of financial crises in an environment that evolves dynamically. In particular, we capture the regime-switching characteristics of the Financial Stress Index (FSI) and Industrial Production Index (IPI), designed to detect periods of financial crisis. We integrate two different stochastic models for FSI and IPI deemed to mirror the systemic financial stress levels in the financial and business cycles, respectively. The joint dynamics of the FSI and IPI, exhibit stochasticity, mean reversion, seasonality, and occasional jumps are identified in the most efficient way. All parameters are modulated by a discrete-time hidden Markov chain that switches between economic regimes reflecting various interacting economic forces. Through change of reference probability technique, adaptive multivariate filters are derived which in turn provides online optimal parameter estimates. Historical Canadian economic-based FSI and IPI are examined and an early-warning signal extraction method is put forward to generate alerts at the early stage of some crisis events. Our modelling approach captures the empirical characteristics of FSI and IPI as well as provides auspiciously early warnings for episodes of systemic financial crisis.