Matthew Selove presented his paper, entitled "The Importance of Model Accuracy in Conjoint-Based Product Design" at the 2010 Sawtooth Software Conference while a student at the MIT Sloan School of Management. His paper derives game theoretic results demonstrating that, even if two conjoint models yield the same relative partworth estimates, the more accurate model (with lower error variance in customers’ product utility in the Hierarchical Bayes logit) gives different strategic recommendations.
In particular, a model with high accuracy implies that differentiation from competitors is important, whereas a model with low accuracy implies that each firm should simply choose feature levels that provide customers the greatest utility (adjusting for marginal cost). To illustrate these results, Matthew fielded a Choice-Based Conjoint exercise on student apparel.
You can read a copy of Matthew's paper at http://www.mit.edu/~hauser/Papers/Selove_Hauser_Price_Competition_Conjoint_Sawtooth.pdf or in our 2010 Conference Proceedings