Optimal Hedging with Views: A Bayesian Approach by

Readers may make verbatim copies of this document for non-commercial purposes by any means, provided that this copyright notice appears on all such copies. Practitioners Abstract The optimal hedging model has become the standard theoretical model of normative hedging behavior due to its intuitive tradeoff of expected return with risk, its efficient use of information and its easy implementation. In practice, the model can be easily implemented with the Parameter Certainty Equivalent procedure, which substitutes sample estimates for the true but unknown model parameters. But subjective views, which refer to opinions concerning the directions of market returns of the assets involved in hedging decisions, are either completely ignored or handled in an ad hoc and unsatisfactory manner within the optimal hedging model. Given the widespread use of subjective views in hedging practice and the potential economic benefit of selective hedging, the lack of accommodation of subjective views in the optimal hedging model is a serious problem and could hamper the model's application in risk management practice. With an empirical Bayesian approach adopted, this study proposes an alternative Bayesian optimal hedging model, in which a hedger can adjust his/her optimal hedging position (ratio) according to his/her view(s) on the expected returns of assets under consideration. Like Lence and Hayes' Bayesian optimal hedging model (1994a, 1994b), the optimal hedging position is also determined by mean-variance optimization conditioned on the predictive expectation vector and predictive covariance matrix of asset returns, but unlike their model, the number and type of subjective views that can be expressed is quite flexible because of the adoption of an empirical Bayesian approach. The empirical Bayesian optimal hedging model provides practitioners with a theoretically intuitive yet quantitatively rigorous framework to blend subjective views and the market consensus estimated from sample data according to their relative confidence levels.