Abstract:
This paper applies machine learning methods and anomaly detection to sudden stop analysis of portfolio flows. Using the isolation forest methodology, univariate as well as bivariate sudden stops of equity and bond fund flows to emerging markets are generated. An anomaly score and an anomaly classification are provided. The results point to an increase in anomalous portfolio flows to emerging markets in recent years. In addition, the isolation forest methodology appears to yield better results than the traditional approach to sudden stop analysis in classifying anomalies connected with the recent capital flow volatility related to the outbreak of the COVID-19 pandemic as well as the interest rate reversal in advanced economies in recent years. The bivariate approach to anomaly detection is better able to identify anomalous episodes of financial stress, where both equity and bond markets are simultaneously affected. Most of the classified anomalies are related to fund flow stops (i.e. simultaneous stops to both equity and bond flows) or surges (i.e. surges in both equity and bond flows). In general, univariate and bivariate anomaly detection using machine learning techniques can play an important part and lead to a better understanding of sudden stops and surges.
