Correlation Timing in Asset Allocation with Bayesian Learning

This paper assesses the relative economic value of volatility and correlation timing in the con-text of asset allocation strategies. Using exchange rate data, we model the dynamic covariance matrix of daily returns by implementing a set of multivariate models based on Dynamic Condi-tional Correlation (DCC) model of Engle (2002). Our analysis takes a Bayesian approach in both estimation and asset allocation. We develop a new MCMC estimation algorithm for the DCC model, which is key for evaluating the optimal portfolio decision of a risk-averse investor in a Bayesian asset allocation framework with CRRA utility. The allocation strategies are designed to account for parameter uncertainty, Bayesian learning as well as model risk by constructing combined forecasts across a large set of volatility and correlation speci\u85cations using Bayesian Model Averaging. We \u85nd that in foreign exchange markets there is substantial economic value in timing correlations in addition to the economic value of volatility timing; with daily rebalancing, correlation timing can add up to 350 basis points per annum to the 500 basis points of volatility timing. This result is robust to reasonably high transaction costs as well as parameter uncertainty, alternative volatility speci\u85cations, diagonal correlation structure and asymmetric correlations