This paper illustrates that least squares learning may lead to suboptimal outcomes even when the estimated function perfectly fits the observations used in the regression. We consider the Salop model with three firms and two types of consumers that face different transportation costs. Firms do...