Regime shift models for systematic trading
I got many questions about regime shift models based on the rationale of the strategy I posted here https://wrightresearch.smallcase.com/. This topic has been of immense interest since the beginning of 2018 given the extreme volatility in the markets.
Regime shifts are large, abrupt, persistent changes in the structure and function of a system. A popular interpretation of this in the stock markets is modeling of variance regimes or dividing the market behavior into high volatility and low volatility segments. One has to adapt their trading methodology based on these regimes to get consistent returns.
Economists model recessionary and expansionary regimes using GDP and other macroeconomic numbers using similar regime shift models.
There are many approaches that one can follow for such a model. A very simple model would be looking at convergence and divergence of long and short term moving averages. Sophisticated models would use machine learning models to predict next period risk based on relevant technical or fundamental factors.
A famous model for this is the Markov Switching Auto-regressive model. I wrote a paper in 2018 titled Evaluating the Building Blocks of a Dynamically Adaptive Systematic Trading Strategy where I evaluate the usage of this model to create a dynamically adaptive trading strategy. For the regime model, I took inspiration from Wyckoff Price Cycles.
With this model, I demonstrated that there is a very clear indication that certain asset classes and segments work better in a particular regime consistently while others work better in another regime. Also while trend following works in the uptrend and downtrend regimes, rangebound strategies work in the accumulation and distribution phases. The stop losses and other strategy specific parameters vary based on regimes.
With a predictive model that can indicate the regime in the next period one can adapt their trading style and asset allocation and get a consistent alpha in all market scenarios. I demonstrate such an example in the paper.
Links for papers on regime shift models:
Evaluating the Building Blocks of a Dynamically Adaptive Systematic Trading Strategy
Regime Detection based Risk Modelling of Asset Classes
For people interested in modeling markov regimes, this is a statsmodels notebook with examples
