Design of a data management system for value bet detection and soccer performance analysis

Investing in the stock market has become incredibly popular in recent times. More noticeably a huge number of young, risk averse and inexperienced traders are now impacting the market with over 25 millions of options contract traded daily. The instability that this trend brought to the table gave rise to increased demand for uncorrelated asset classes - i.e. assets that are not influenced by the current market situation – such as collectibles and sports betting. In a certain sense the betting and trading world have a lot in common, whoever is able to process the information at their disposal better and faster is usually going to make a profit. A classic example of how technology is impacting the stock market is HFT or High Frequency Trading, a financial trading approach that requires dedicated software to perform the acquisition and liquidation of positions on the very short-term, sometimes seconds, allowing investors to leverage small profits on a huge number of transactions. Since the first attempts to apply predictive models to horse racing in the 1970s, which failed due to the lack of data, technology has made a huge step forward. In recent years some companies have specialized in collecting and selling sports data that is now more available than ever before. With the aid of artificial intelligence we can now analyze each event in a match and evaluate its effects on the final outcome. This approach is exploited by sports organizations at all levels. Managers know which players are the most valuable and which generate the higher fan engagement. Bookmakers have a tool to correctly evaluate the performance of each team to offer profitable odds, but so do investors. The ability to identify a betting scenario where the odds don't represent correctly the possible outcome opens up the possibility for a systematic betting approach that in the long run will yield a profit. Even if betting has always been associated to gambling, with the aid of big data and artificial intelligence, one individual can now profit from mispriced wagers in the sports market in the same way that one could profit from trading mispriced options in the stock market. The necessity of extracting the most valuable information and metrics from the data that is at our disposal represents a valuable competitive advantage as it boasts our ability to find profitable investment opportunity. In this thesis I present a complete data management pipeline that can efficiently extract valuable information that are later going to be used to train AI models. The theoretical approaches to identify value betting scenarios have been already openly presented in the scientific literature, on the other side the technical details are mostly kept secret by companies who don't want to disclose their design choices. I hope that this work can represent a valuable starting point for future research in the betting and sports analytic world.