IMPROVING ON THE MARKOV-SWITCHING REGRESSION MODEL BY THE USE OF AN ADAPTIVE MOVING AVERAGE (original) (raw)

Regime detection is vital for the effective operation of trading and investment strategies. However, the most popular means of doing this, the two-state Markov-switching regression model (MSR), is not an optimal solution, as two volatility states do not fully capture the complexity of the market. Past attempts to extend this model to a multi-state MSR have proved unstable, potentially expensive in terms of trading costs, and can only divide the market into states with varying levels of volatility, which is not the only aspect of market dynamics relevant to trading. We demonstrate it is possible and valuable to instead segment the market into more than two states not on the basis of volatility alone, but on a combined basis of volatility and trend, by combining the two-state MSR with an adaptive moving average. A realistic trading framework is used to demonstrate that using two selected states from the four thus generated leads to better trading performance than traditional benchmarks, including the two-state MSR. In addition, the proposed model could serve as a label generator for machine learning tasks used in predicting financial regimes ex ante. Keywords Regime switching • Technical analysis • Markov model • Trading 1 Introduction Financial markets are characterised by periods of evolving, low-volatility growth separated with a disruptive, highvolatility contractions. These regimes can be distinguished by significant changes in asset returns, variances, and correlations, laying a groundwork for accurate detection techniques to be exploited by portfolio managers. One such detection technique is the Markov-switching regression model (MSR), introduced by Goldfeld & Quandt in 1973 [1], and later extended by Hamilton in 1989 [2] and Krolzig in 1997 [3], which has since become one of the most popular statistical methods to distinguish regime shifts in economics and finance. However, the original two-state MSR model is not ideal, as two volatility states are insufficient to describe the complexities of market dynamics, and attempts to use a three-or higher state MSR can suffer from instability due to overly frequent regime switches [4], which in addition adversely impact portfolio returns by increasing the costs of trading. Furthermore, volatility by itself is not an infallible indicator of up-or down-trending markets. This paper improves on the current state of the art by combining the two-state MSR with the use of Kaufman's adaptive moving average (KAMA) [5], a method that is both stable (avoiding spurious regime shift detections) and accurate in locating the times of onset of regime switches. This combination generates four detected regimes based not on volatility alone, allowing, ultimately, a focus on those two market states with the greatest trading value, namely low-variance bullish and high-variance bearish, with trading results that will demonstrate the utility of the method.