Core Matters I A Machine Learning approach to equity quantitative models
- This paper presents a real-life application of Machine Learning (ML) through Genetic Algorithm (GA): a step forward from traditional econometric modelling.
- Efficient search-based analytics, implemented with the use of GA, results in a considerable reduction of time needed to develop models, and an increased predictive ability. This is done by maximising results from the out-of-sample (OOS) back testing, which is the theoretical profit derived from investment positions based on the signals coming from the models.
- Prior to 2020, we had developed “barometer models” for 16 European equity sectors and 12 styles, using a classical regression-based approach, to support equity investment decisions. By construction, the models are set up to exploit large deviations from the fair value in trading over/under-valued sectors against the market (MSCI Europe), thus justifying the description given to them – barometer models.
- The performance is measured by the profits under a simple stylized trading rule: Sell (buy) the over-valued (under-valued) sectors. In the traditional approach, finding a well-specified model has proved to be rather time consuming. Furthermore, there is also no guarantee that a good fair value model would bring a satisfactory (not to speak of the best) investment performance. On the other hand, GA are used for solving optimisation problems in machine learning, helping to find the best possible model in terms of out-of-sample results. In our initial application of the ML procedure, we obtained models for 31 European equity sectors, 12 European equity styles and 12 equity emerging markets.
- Our investment decisions are not solely based on relative over/under-valuation coming from our models; we also use additional complementary quantitative tools as well qualitative analysis.
- Next steps: increase the number of asset classes to cover with the described approach.