Dimension reduction in mean-variance portfolio optimization

Dimension reduction methods are useful pre-processing tools for efficient quantitative analysis with the aim to preserve the main features of the multidimensional data. However, negative values resulting from the transformation may obscure the interpretation of the analysis. This novel study aims to investigate the effects of non-negative dimension reduction methods on the mean-variance portfolio optimization model. Backtesting results for major stock market indices show that reducing dimensionality of asset prices may improve the overall efficiency of the mean-variance portfolio optimization output.