Market Regimes and Machine Learning
Overview
When constructing a trade strategy that includes a market regime, it is often necessary to construct an algorithm that identifies
the regime so that the strategy can be objectively back-tested.
In some cases, the trader just identifies the regime with certain thresholds of selected indicators. In the cases where the
trader has identified the regimes, but has not provided an algorithm for recognizing the regime, machine learning
can be used to try to learn the regime identifications from the data.
Supervised Learning
Once an expert has indentified the relevant market regimes in the past and assigned labels to the past market history,
machine learning can be used to construct a model that will identify those regimes.
The first step is to identify the features of the market that will be used to identify the regime. That is, what information will
be made available to the machine learning algorithm to train on. Typically, this will usually involve market stress indicators
such as the TED spread, or other indicators such as the momentum in various indices.
(see
identifying regimes).
Next, each market regime should be encoded on each historical record. For example, if the regime represents risk on/off, then
either a number 0 or a number 1 can be appended to each record depending on whether the regime is present or not.
Choosing an Algorithm
Lastly, an algorithm needs to be chosen to train with the data. The following represent typical choices.
