Advances in Financial Machine Learning

CHAPTER 17 Structural Breaks

17.1 Motivation

In developing an ML-based investment strategy, we typically wish to bet when there is a confluence of factors whose predicted outcome offers a favorable risk-adjusted return. Structural breaks, like the transition from one market regime to another, is one example of such a confluence that is of particular interest. For instance, a mean-reverting pattern may give way to a momentum pattern. As this transition takes place, most market participants are caught off guard, and they will make costly mistakes. This sort of errors is the basis for many profitable strategies, because the actors on the losing side will typically become aware of their mistake once it is too late. Before they accept their losses, they will act irrationally, try to hold the position, and hope for a comeback. Sometimes they will even increase a losing position, in desperation. Eventually they will be forced to stop loss or stop out. Structural breaks offer some of the best risk/rewards. In this chapter, we will review some methods that measure the likelihood of structural breaks, so that informative features can be built upon them.

17.2 Types of Structural Break Tests

We can classify structural break tests in two general categories:

• CUSUM tests: These test whether the cumulative forecasting errors significantly deviate from white noise.
• Explosiveness tests: Beyond deviation from white noise, these test whether the process exhibits exponential growth or collapse, ...