Title: A robust knock-out strategy for an high-dimensional portfolio choice problem
Authors: Marco Gambacciani - University of Zurich and Swiss Financial Institute (Switzerland) [presenting]
Abstract: A stylized fact of financial returns is that extreme values are part of their historical behavior, while sometimes mistakenly referred to as outlying observations or ``outliers''. In opposition to other scientific fields, the outliers in financial applications should not be discarded as measurement error, but considered as useful information. The use of robust statistics still provide a very useful and viable tool for application with financial data, as such as optimal portfolio applications, although in a completely different fashion than how is implemented in other fields. We intend to exploit the difference between classical and robust estimators as a provider of useful information about market conditions, which is then converted in proposed trading strategies. With an empirical application to weekly returns of the constituents of the MSCI World Developed Single Stocks Index, we show that the portfolio performance can benefit by incorporating standard portfolio allocation methods with the information derived from comparing robust and classical estimators. In particular, the methods introduced are shown to be viable as knock-out strategies, i.e. starting with a large number of stocks, the strategy is able to do a pre-selection of a group with the most performing stocks, to which the final portfolio allocation method will be applied.