A MODEL OF STOCK MARKET CORRECTIONS USING STATISTICAL MACHINE LEARNING

Abstract

ABSTRACT Global real economic activity is closely linked through various financial mechanisms, including credit and equity markets, as illustrated by the 2008 U.S. financial crisis and the resulting widespread economic dislocations. We apply statistical machine learning to the problem of identifying the key risk factors associated with daily equity market corrections and to address whether changes in monetary policy influence aggregate stock returns. Monetary policy changes that persist from one trading day to the next increase the odds of an equity market correction, as back-to-back adjustments in the funds rate might signal an unexpected policy action. If market agents view the temporal evolution in monetary policy as being unpredictable, the likelihood of market corrections increases as the overall confidence in the Fed’s ability to provide long-term price stability diminishes. The likelihood of a stock market decline is reduced if monetary policy follows a predictable path. An implication of this finding is that monetary authorities tend to remain in a tightening or an easing pattern until they receive a strong signal that they may have tightened or eased excessively. Such a signal in a contractionary cycle could be an inversion in real economic growth accompanied by a sizable equity market correction. Another finding of this paper is that heightened variability over time increases the odds of market corrections. The market response may be driven by the waning confidence of market agents. The impact of monetary policy changes is not symmetric as declines in the interest rate may lead to larger changes in equity prices. This fits the notion that many market agents fail to consider not only the effect of asymmetric temporal adjustments across a range of risk factors but also the impact of persistence in monetary policy changes on equity prices. Model results indirectly incorporate momentum-driven investor psychology and bandwagon effects. An original multiple feature lag-length optimization algorithm and the associated generalized non-linear vector autoregressive (GENVAR) framework introduced in the paper merit further study. Keywords Monetary Policy, Equity Prices, Stock Market Corrections, Statistical Machine Learning, GENVAR, Multiple Feature Optimal Lag Algorithm, Interest Rates, Asymmetric