Stochastic Correlation in Commodity Spread Options

I examine a dataset of short-dated gasoline crack spread options traded on NYMEX using a two-asset version of Black (1976) model and find that the changes in implied correlations display economically significant variations not explained by changes in futures prices or changes in implied volatilities of single-asset options. I show that unspanned stochastic correlation can be generated in Wishart-based stochastic volatility models proposed by Gourieroux (2006) and da Fonseca et al. (2007) and confirm that temporal changes in conditional correlation are absolutely necessary to explain the variations in spread options prices across time: average relative pricing error increases from 5\% to 32\% when the correlation state variable is fixed for the whole sample