Comparing current Highstreet's assets model with advanced machine learning models

点击翻译按钮获取中文摘要

Highstreet Asset Management Inc. is an investment management firm based in London, Ontario. It provides discretionary money management services to individual investors and their families, foundations, pension plans, and institutional investors. As data becomes more readily accessible and grows exponentially, quant analysts at Highstreet are looking for efficient ways to analyze stock market data. Currently Highstreet analysts manually test and select factors to build multi-factor linear models. However, the model building process can be quite time-consuming and the final model cannot be completely free from personal bias, which is extremely dangerous in a fast-changing market. Highstreet spent 6 months to complete a Canadian model updating based on a universe of 280 stocks. In 2015, Highstreet will need to finish the US model review based on a universe of 800 stocks. It involves testing on hundreds of factors and selecting appropriate ones to build a model that fits different market dynamics. Thus, a more efficient and systematic process of mode building and testing is urgently needed. My former PhD student Robert Yan (who works in Highstreet now) and I have proposed several algorithms to analyze stock market data. To address the real-world investment problems at Highstreet, we plan to apply feature selection / feature extraction and predictive modelling techniques to the data from the 260 Canadian stocks and 800 US stocks at Highstreet. These machine learning techniques could potentially help Highstreet speed up model review process, as well as improve stock market forecasting. As we work with Highstreet and aim to provide better investing strategies for Canadian companies and individuals, we will generate indirectly economic and social benefits for Canadians.