Investment Strategies Using Orthogonal Portfolios

This paper evaluates the usefulness of the orthogonal portfolio approach proposed by MacKinlay and Pastor (2000), for the estimation of the expected returns of Swedish industrial portfolios from 1980 to 1997. In this approach the expected returns are linked to the residual covariance matrix of a given factor model. The analysis consists of two related but distinct parts. We first examine the significance of the orthogonal approach as an asset-pricing model. The second part judges the ability of the orthogonal portfolio approach in forming investment strategies. The estimation of the parameters of the applied models requires solution of a difficult and nonlinear likelihood function. We use simulated annealing as our procedure to find the global optimum of the likelihood function. Thus, in contrast to MacKinlay and Pastor (2000) we can let the model determine the importance of the unobserved factor, the orthogonal portfolio, and allow for heteroskedasticity in the idiosyncratic risk of the test assets. We first compare the estimated expected returns and the characteristics of the factor portfolios from different models. Then, all models are evaluated from a portfolio perspective: We calculate the weights of the portfolio with the maximum Sharpe-ratio, the tangency portfolio, and the out of sample Sharpe-ratios of these portfolios are used as the evaluation metric.