An early warning system for emerging markets

Abstract

Financial markets of emerging economies are vulnerable to extreme and cascading information spillovers, surges, sudden stops and reversals. With this in mind, we develop a new online early warning system (EWS) to detect what is referred to as `concept drift' in machine learning, as a `regime shift' in economics and as a `change-point' in statistics. The system explores nonlinearities in financial information flows and remains robust to heavy tails and dependence of extremes. The key component is the use of conditional entropy, which captures shifts in various channels of information transmission, not only in conditional mean or variance. We design a baseline method, and adapt it to a modern high-dimensional setting through the use of random forests and copulas. We show the relevance of each system component to the analysis of emerging markets. The new approach detects significant shifts where conventional methods fail. We explore when this happens using simulations and we provide two illustrations when the methods generate meaningful warnings. The ability to detect changes early helps improve resilience in emerging markets against shocks and provides new economic and financial insights into their operation.