Misspecifications of Covariance Matrix of Random Coefficients in the BLP Model: Simulation Studies and Empirical Applications ∗

[tentative] This paper discusses misspecification problems of covariance matrix of random coefficients in the BLP model, finding that when the correlated random coefficients are assumed to be independent of each other, which is commonly encountered in empirical literature when implementing the BLP procedure, their standard deviations may have negative estimates. This case occurs even if some random coefficients with very small correlations are incorrectly specified to be uncorrelated. We also conduct the demand estimation with BLP and Nevo’s datasets using the BLP model with a full specification of covariance matrix of random coefficients, obtaining a reasonable estimate for the demand parameters especially for standard deviations of random coefficients.