Seminar: Forecasting Trends in U.S Equities Markets Using Machine Learning and Full-Depth Market Da
Supervisor: Dr. Wolfgang Banzhaf
Forecasting Trends in U.S Equities Markets Using Machine Learning and Full-Depth Market Data
Department of Computer Science
Tuesday, April 21, 2015, 11:30a.m., Room EN 2022
Traditional forecasting in the financial markets uses static mathematical formulas called technical indicators to predict the future behaviour of an equity. These indicators rely on the most shallow level of data, namely level 1 market data, which contains only a small portion of the data available for any given equity. Moreover, these indicators have no recourse when the market does not adhere to their static formulas.
In this research, we utilize full depth-of-book market data, which provides us with a finer-grained look at market activity – the individual orders that have been executed or are waiting to be executed – which cannot be seen in level 1 market data. By using the power of machine learning, we are able to evolve dynamic, stock-specific solutions, where technical indicators are not appropriate because of the complexity at this level of data. We employ a multi-objective optimization for profit and Sharpe-ratio which is an important metric of risk adjusted return. By doing this, we are not only evolving a solution that favours profit, but also one that takes on the least amount of risk.