Non-Stationarity, is defined as a quality of a process in which the statistical parameters (probability distributions) of the process change with time.
One of the consequences is that it might not make much sense to consider and model financial markets as one big distribution over time. And this is where Trading Regimes come in.
The concept behind regimes is that non-stationary time series, such as market data, shift through different “modes” during which market behavior and dynamics are different.
Being able to recognise these trading regimes would a allow a trader to adapt his strategies to better respond to these regime shifts.
Readers interested in the more formal definition of regime switching in econometrics might want to investigate Markov-based systems.
In this post, we will make a looser interpretation of the term: Trading regimes can be thought as providing a context to the trading strategy. This could be any meta-indicator or external variable susceptible to affect or “predict” the performance of a trading strategy. Let’s scratch the surface by making a (non-exhaustive) list of potential ideas.
OK, we start with a wide departure from Trading Regime in the formal sense of the term.
The directional trend filter however is a basic and popular filter. It is based on the idea that “the trend is your friend”: it considers the trend on a higher timeframe and only allow trades in the direction of the main trend. The trading regimes might be thought of 2 or 3 modes: bullish/uptrend, bearish/downtrend or neutral.
Another possible filter could be a switch indicating when the market makes an extreme divergence from history. The idea being that a strategy designed and tested on historical norms might degrade during these abnormal periods.
Michael Stokes from MarketSCI discusses his Abnormal Market Filter implementing a smart “shades of grey” concept, with a “dimmer” switch rather than a binary ON/OFF, normal/abnormal switch.
This filter is based on extreme volatilities but I have also seen traders using abnormal correlation filters (when many markets in the portfolio exhibit sudden spurious correlations such as during credit events).
There are many causes to instabilities of financial markets, for example business cycles or monetary policies.
I alluded in a previous post how business cycles could be used in an interesting meta-strategy, trading regime filter:
it would be interesting to measure and test the impact of a macro filter on a Trend Following system (ie something in the vein of: “favour long trades in the system when the macro indicators indicate a period of expansion and short trades during declines”). Maybe exploring a mechanized version of Schumpeter Business Cycles or Kondratiev Waves as a very long-term filter would be a worthy approach.
Similarly, macro indicators such as yield curve, GDP, Monetary supply might provide some input into a Trading Regime model.
For a Trend Following system (or a Mean Reversion system) it could be a great advantage to be able to identify those markets that are more likely to trend (or not). The Hurst exponent can be estimated on time series to derive their tendency to revert to the mean or cluster in the same direction.
Each market could be classified as trending or mean-reverting based on the value of its Hurst exponent.
This is a concept implicitely used by traders trading the equity curve of their systems. A simple approach is to monitor the equity curve of the system and apply a moving average to it: you can allocate more or less funds based on the current relative position of the equity curve to its moving average. An anti-martingale-type strategy would reduce funds when the system equity curve is below its MA.
This can be taken further by having a collection of “dummy” trading strategies, which performance is monitored to derive the current market regime (identified by which group of strategies is over/under performing). This is a concept explored in the Hack the Market post, linked above.
This approach is what triggered this post. One of my current working assumption is how Trend Following represents a trading strategy positioned to benefit from positive kurtosis and positive serial correlation.
One hypothesis I am investigating regarding these characteristics is whether they could be used for Trading Regime identification:
Understanding these market characteristics is a first step towards being able to identify and measure them. This, in turn should be a step to linking Trend Following performance to the state of these market characteristics. Finally, this might be a step towards devising a way for a Trend Following strategy to adapt to these changing market characteristics (this last point makes a very big assumption: market characteristic changes can be predicted with some degree of accuracy).
Hopefully more on that coming soon, with an attempt at calculating serial correlation “in the tails”.
There are of course other price behaviors that might be used for Trading Regime identification (mean, variance, volatility, covariance, cointegration, skew, etc.)
An important point to consider, for a trader planning on using trading regimes, is the fact that the approach would result in trading a collection of “more specialized” strategies (read: less robust), switching between them as the market switches between its various regimes.
Thinking that such an approach can be more succesful than a “one-size-fits-all”, more robust trading strategy makes a strong assumption about regime persistence, namely that the regime persists longer than the time it takes to identify the shift to the new regime.
In essence, if the regime switch identification lag is too long, you would probably end up applying the wrong specialized trading strategy to the current regime and suffer from under-performance.