Statistical arbitrage trading with implementation of machine learning : an empirical analysis of pairs trading on the Norwegian stock market

Abstract

The main objective of this thesis is to analyze whether there are arbitrage opportunities on the Norwegian stock market. Moreover, this thesis examines statistical arbitrage through cointegration pairs trading. We embed an analytic framework of an algorithmic trading model which includes principal component analysis and density-based clustering in order to extract and cluster common underlying risk factors of stock returns. From the results obtained we statistically prove that pairs trading on the Oslo Stock Exchange Benchmark Index does not provide excess return nor favorable Sharpe ratio. Predictions from our trading model are also compared with an unrestricted model to determine appropriate stock filtering tools, where we find that unsupervised machine learning techniques have properties which are beneficial for pairs trading.