Benjamin J. Gillen

A Taxonomy of Utility Functions

Modern financial theory began with the publication of two articles in 1952: Markowitz (1952a) and Roy (1952). The principal difference between the Roy article and the Markowitz article is that Roy recommended a specific portfolio from the mean-variance frontier, namely, the one which maximizes

A Pari-Mutuel-Like Mechanism for Information Aggregation: A Field Test inside Intel

A new information aggregation mechanism (IAM), developed via laboratory experimental methods, is implemented inside Intel Corporation in a long-running field test. The IAM, incorporating features of pari-mutuel betting, is uniquely designed to collect and quantize as probability distributions dispersed, subjectively held information. IAM participants’ incentives support timely information revelation and the emergence of consensus beliefs over future outcomes. Empirical tests demonstrate the robustness of experimental results and the IAM’s practical usefulness in addressing real-world problems. The IAM’s predictive distributions forecasting sales are very accurate, especially for short horizons and direct sales channels, often proving more accurate than Intel’s internal forecast.

Demand Estimation with High-Dimensional Product Characteristics

Structural models of demand founded on the classic work of Berry, Levinsohn, and Pakes (1995) link variation in aggregate market shares for a product to the influence of product attributes on heterogeneous consumer tastes. We consider implementing these models in settings with complicated products where consumer preferences for product attributes are sparse, that is, where a small proportion of a high-dimensional product characteristics influence consumer tastes. We propose a multistep estimator to efficiently perform uniform inference. Our estimator employs a penalized pre-estimation model specification stage to consistently estimate nonlinear features of the BLP model. We then perform selection via a Triple-LASSO for explanatory controls, treatment selection controls, and instrument selection. After selecting variables, we use an unpenalized GMM estimator for inference. Monte Carlo simulations verify the performance of these estimators.