Forecasting the volatility of Financial Time Series by Tree Ensembles

The use of new tools for economic data analysis in the last decade has led to significant improvements in forecasting. This is due to the relevance of the question, and the development of technologies that allow implementation of more complex models without resorting to the use of significant computing power. The constant volatility of the world indices forces all financial market players to improve risk management models and, at the same time, to revise the policy of capital investment. More stringent liquidity and transparency standards in relation to the financial sector also encourage participants to experiment with protective mechanisms and to create predictive algorithms that can not only reduce the losses from the volatility of financial instruments but also benefit from short-term investment manipulations. The article discusses the possibility of improving the efficiency of calculations in predicting the volatility by the models of tree ensembles using various methods of data analysis. As the key points of efficiency growth, the author studied the possibility of aggregation of financial time series data using several methods of calculation and prediction of variance: Standard, EWMA, ARCH, GARCH, and also analyzed the possibility of simplifying the calculations while reducing the correlation between the series. The author demonstrated the application of calculation methods on the basis of an array of historical price data (Open, High, Low, Close) and volume indicators (Volumes) of futures trading on the RTS index with a five-minute time interval and an annual set of historical data. The proposed method allows to reduce the cost of computing power and time for data processing in the analysis of short-term positions in the financial markets and to identify risks with a certain level of confidence probability.

1. Quinlan J.R. Induction of Decision Trees. Machine Learning. 1986;(1):81–106.

2. Sollich P., Krogh, A. Learning with ensembles: How overfitting can be useful. Advances in Neural Information Processing Systems. 1996;8:190–196.

3. Yoo W., Ference B.A., Cote M.L., & Schwartz A.A. Comparison of Logistic Regression, Logic Regression, Classifcation Tree, and Random Forests to Identify Effective Gene-Gene and Gene-Environmental Interactions. International Journal of Applied Science and Technology. 2012;2(7):268.

4. Freund Y., Shapire R. A decision-theoretic generalization of on-line learning and an application to boosting. Proceedings of the Second European Conference on Computational Learning Theory; 1995:23–37.

5. Friedman J. Greedy Function Approximation: A Gradient Boosting Machine. The Annals of Statistics. 2001;29(5):1–39.

6. Breiman L. Random Forests. Machine Learning. 2001;45:5–32.

7. Geurts P., Erns D., Wehenkel L. Extremely randomized trees. Machine Learning. 2006;63:3–42.

8. Engle R.F. Estimating Time Varying Risk Premia in the Term Structure: The Arch-M Model. Econometrica. 1987;55(2):391–407.

9. Raileanu L.E., Stoffel K. Theoretical Comparison between the Gini Index and Information Gain Criteria. Annals of Mathematics and Artifcial Intelligence. 2004;41(1):77–93.

10. Bollerslev T. Generalized Autoregressive Conditional Heteroskedasticity. Journal of Econometrics. 1986;31:307–327.

11. Sundsoy P., Bjelland J., MIqba A. Big Data-Driven Marketing: How machine learning outperforms marketers’ gutfeeling. Massachusetts Institute of Technology. Lecture Notes in Computer Science. 2014;8393:367–374.