Browsing by Author "Asikainen, Juha"

The thesis handles application of principal component analysis (PCA) in momentum based investment strategies. Principal component analysis is a dimension reduction method for multidimensional datasets that seeks to find new mutually uncorrelated variables called principal components so that the explanatory power over the original dataset is maximized. Momentum strategies are long-short, zero investment portfolios that within an asset class buy instruments that have performed relatively well and sell short instruments that had weak performance. The performance is measured in by directly applying total returns or using some derivative of it. The evaluation horizons typically range for a few months to a year. Earlier studies have assessed total returns eg in relation to fundamental factor models. Main data source of the study is total return data for US equity spanning ca 30 years of recent history. The strategies are defined using monthly data. The principal component models are estimated using daily return series. These models are then utilised in two types of momentum strategies. First group uses residuals from PCA, i.e. the portion of return not explained by the principal components. Second type of strategies allocates money based on the returns of principal components. The strategies result in time return series of return that would have been achieved by investing according to the rule based strategies. These are then analysed by using statistical and econometric methods. Furthermore, the returns are compared to results obtained in prior studies that have utilized similar methods. This includes studying the effect of autocorrelations in total returns, residuals and principal components for the success of respective strategies. The results in indicate that both sets of strategies seem to generate absolute returns that are in line with those obtained using the raw total return signal. The volatility of these returns as measured by standard deviation is significantly lower than that of strategies based on total return. Earlier studies using residuals defined using different types of models have shown similar results. The results of residuals based strategies are not explained by the autocorrelations of the underlying instruments. In fact these autocorrelations would seem to distract from the returns. The principal components, on the other hand, seem to have positive autocorrelations which in large part explain the success of related strategies. The key finding of the thesis is the attribution of decomposition of momentum profits into distinct sources by applying the relatively simple and well-known method of principal component analysis. These results complement the earlier research regarding the split of momentum profits between systematic and non-systematic sources of variation in finding that both are significant. The residuals based strategies have a higher economic significance due their low correlations against conventional strategies. The analysis of autocorrelations points to differing econometric drivers between the two sets of strategies.