Interval Validity Testing and Optimization in Quantitative Trading
| Autor: | Huarun Dai, Jiangrui Chen |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Feature engineering
Rate of return Mathematical optimization Sequence 050208 finance Index (economics) Discretization Computer science 05 social sciences 02 engineering and technology Interval (mathematics) Function (mathematics) Test (assessment) 0502 economics and business 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing |
| Zdroj: | 2020 The 11th International Conference on E-business, Management and Economics. |
| DOI: | 10.1145/3414752.3414807 |
| Popis: | Machine learning algorithm has been widely used in quantitative investment, but there are some problems in factor testing, factor construction and objective of the loss function. First of all, this paper constructs the style index of the CSI 300 index, including institutional index, QFII index and shareholder concentration index. From the CSI 300 index and these three indexes, four factors are extracted respectively to form 16 factors we will test and use in this paper. Secondly, traditional factor testing method tests the linear validity or sequence validity. In this paper, we propose a method to test the interval validity because factors may validate internally instead of linearly. Then, we can discretize the factors and complete the feature engineering. Third, current machine learning applications in the quantitative trading are through the way of discretization of the rate of return and minimizing the loss function. But it ignores a lot of information of the rate of return. Therefore, this paper reconstructs the objective function of machine learning, modifies it to maximize the cumulative rate of return. Finally we use the heuristic algorithm Particle Swaps Optimization (PSO) to solve the equation, and obtains the corresponding parameters. Finally, we use the dataset from February 9th, 2009, to October 24th, 2019, to back testing, with 80% training data. The annualized return of the strategy is 24.96%, with the maximum withdrawal of 17.2%. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|