For many years, the Random Walk Hypothesis (RWH) and Efficient Market Hypothesis (EMH) have been the cornerstones to economists understanding financial markets. RWH and EMH hypothesise that financial markets are perfectly efficient, and profits cannot be derived from price information alone. Even RWH was initially disproved in 1988 by MacKinlay and their results have subsequently been confirmed by others, the underlying idea remained in the collective mindset of a wide number of economists.
Recently the focus of attention has turned towards the EMH, in which extensive analysis of financial data is empirically leading the way to the breakdown of the EMH in its weak-form. Findings such as these suggest that recurring patterns and underlying structures exist within financial time-series. Particularly, studies into the randomness of financial time-series hypothesise that fractal repetitions could occur, and the degree of predictability can be estimated. This suggests the idea that inefficiencies exist in financial markets which can be exploited for financial gain. The ability to detect and predict such structures and patterns would not only produce profitability in trading, but would also lead to a deeper understanding of the way financial time-series work.
While much research effort has been put into finding structures between different financial time-series to optimise portfolios of stocks, little effort has been put into the identification of structures within a single financial time-series. My hypothesis aims to identify such structures through the application of unsupervised learning techniques on an information rich feature space. Once identified, these structures are to be used in a variety of autonomous trading systems to identify whether such structures point to inefficiencies that can be exploited for financial gain. This provides fascinating reasons to investigate the possibility of finding structures and patterns using unsupervised learning techniques within a single financial time-series not sufficiently explored in the financial space.
The irruption of the IoT and the massive usage of telecommunications networks have push the research community to look for more innovative and efficient ways to find structures in the apparent chaos. I hope the lessons learnt in other industries can be successfully applied to the problem of predicting financial time series.