532 – Bayesian Modeling for Business and Economics
Establishing the Foundation of Hedge Fund Asset Allocation Decisions Using Bayesian Modeling
Weiren Chang
JP Morgan
This paper attempts to estimate the diversified fund-of-hedge-funds (FoHF) industry's aggregate hedge fund (HF) strategy allocations. Unlike long-only equity and fixed income indices that have published constituents and composition weights, such a benchmark does not exist for hedge fund investors and asset allocation decision makers. As a result, it's desirable yet difficult for a FoHF manager to asses whether the portfolio has significant strategy/style biases so performance attributions can be conducted. The author proposed several classic and Bayesian regression models to address this need. Hedge fund strategy allocations are model parameters; dependent variables are Diversified FoHF index and individual FoHF performance data; independent variables are major HF strategy index performance data. Investment industry experience provided guidance for setting Bayesian prior (ex-ante) parameter values; Markov Chain Monte Carlo simulations generated posterior (ex-post) allocation estimates. The author believes a Bayesian hierarchical model provides good balance between these objectives: (1) results that are consistent with industry experience and could be easily interpreted; (2) model parsimony and good fit to data. Future research opportunities such as capturing dynamic parameter behaviors are also discussed.