Pitfalls In linear modeSs for style analysis

摘要

We discuss the statistical properties of return-based OLS style analysis intfoduced by Sharpe (1992). The ahn of style analysis is to inl'er a fund manager's Investment decisions using only publicly available data on the fund Performance and on the time evolution of market indexes. We show that the niodel proposed by Sharpe suffers of relevant drawbacks, most notably that it fails to yield correct results even in the simple case of a buy-and-hold strategy that only invests in the market indexes. Under this hypothesis we show that a model linear in index levels, as opposed to index returns, e.stimated via a Kaiman filter avoids Sharpe's model drawbacks. We further extend our analysis to strategies where the fund manager policy changes with time and the asset classes in which the fund manager invests are not known exactly. In this last case we show that a style analysis is possible only conditional to either an orthogonality hypothesis on the "active" investment strategy, or by the introduction of suitable instrumental variables.