Regime Switching Models for Markets

Search for structure in the financial markets has recently become an active field of research by scientists from different areas of expertise. The Efficient-Market Hypothesis states that the markets reflect all publically available information within their price. This also means that no profit higher than the average market returns allow can be gained using only the price information. However, reasearch has shown that this is not the case. This thesis explores the methods of searching for structure using a single asset price information. The experiments are done with multiple profit data, generated by a basket of trading rules, selected by looking at the field of technical analysis. Methods of dimensionality reduction and market classification into regimes are used. Both linear (PCA) and non-linear (LTSA) dimensionality reduction is attempted. The results show structure within the low-dimensional representations of the highdimensional profit data and potential for risk analysis. The structure differets between real data and artificially generated data, showing importance of certain features of the real financial data. The market classification algorithm is shown to assign different regimes to visually different data, creating the potential for profitable exploitation.