M.S. (Master of Science)
Department of Electrical Engineering
The interest in applying machine learning to financial trading in the hedge fund industry has exploded in the last five years due to the massive success of a handful of ‘quantitative’ investment firms like Renaissance Technologies who has pioneered the use of machine learning techniques in investment since the 1980s. The failure rate of such firms attempting to deploy financial machine learning strategies is very high. This thesis reviews many of the causes for failure such as harmful correlations between examples in the dataset, redundant observations, improper data sampling paradigm, and multiple testing bias. Of these, multiple testing bias is the most serious and pervasive. This thesis evaluates the efficacy of Combinatorial Purged Cross Validation, a technique for combatting multiple testing bias, when used with SOA neural network classifiers. CPCV is experimented with on a dataset containing no relationship between features and outcomes as well as on a synthetic dataset with a known subtle relationship between the features and outcomes. It is concluded that CPCV is likely highly effective at eliminating false positives but much future work is needed to explore CPCV’s merit when using various financial datasets of varying sizes and characteristics.
Fritz, Colin, "A Review of Reasons For Failure in Applying Machine Learning to Financial Trading and An Experiment investigating Combinatorial Purged Cross Validation’s Merit in Preventing The Most Prominent of These Reasons, Multiple Testing Bias" (2019). Graduate Research Theses & Dissertations. 7049.
Northern Illinois University
Rights Statement 2
NIU theses are protected by copyright. They may be viewed from Huskie Commons for any purpose, but reproduction or distribution in any format is prohibited without the written permission of the authors.