Textual/Financial Prediction is the linking together of financial news articles and machine learning to create a new class of HFT stock prediction engine that can recognize and adapt to immediate market changes from breaking news articles. Within this area we have created two systems; AZFinText and CentralFinance.
The AZFinText Project examines the relationship between financial news articles and their impact on stock prices. This project utilizes various textual representation schemes, price prediction models and machine learning techniques to accomplish profitability in extreme near-term trading. Based on the premise that unexpected news events can shape the price of a stock, we capitalize on automatically identifying the relevant keywords in the news article and then execute a trade well-before human traders have a chance to read the news article.
CentralFinance is the 2.0 version of textual/financial prediction called Press Release Engineering. Knowing that HFT systems are now using textual/financial information, can we manipulate the news articles to drive stock price. This system looks at three aspects, terms, timing and outlet. Terms deals with carefully crafting news articles using specific terms known to move stock price. Timing looks at optimizing article release for maximized results. Outlet refers to which media outlet to release the article. Taken together, Press Release Engineering may be a new avenue for companies to manage stock price.
Sports Analytics is the mining of relevant data from Sports-related databases and producing accurate predictions. This knowledge can provide an edge to sports organizations and gamblers alike.
Greyhound racing is an exciting sport that works well with Sports Analytics and machine learning in particular. Leveraging advanced web mining and machine learning techniques, the AZGreyhound system was able to identify winners 45.35% of the time and furthermore would correctly identify Superfecta Box combination winners 6.35% of the time as compared to random chance at 2.79%.
Following upon the success of AZGreyhound, the program was expanded to include other racing sports such as Harness racing. This series of research papers investigate portability of greyhound machine learning techniques to harness racing, compares them against track experts, crowd wisdom and other successful predictive techniques such as the Dr. Z System and then investigates the proper amount of race history to optimize the machine learning algorithms. Further extensions of this work include investigating different machine learning techniques and adapting this to thoroughbred and Nascar prediction.
This recent research project focuses on using Twitter as a prediction vehicle for match outcomes. Using the ideas of crowd wisdom, we explore the role that crowd sentiment can play in evaluating game results. Will crowds that are more confident or nervous lead to better predictability? We plan to explore these questions within the domain of English Premier League Soccer. Results from this exploratory analysis have the potential application to similar sporting domains.