NBER Publications by James W. Roberts

A bidding process can be organized so that offers are submitted simultaneously or sequentially. In the latter case, potential buyers can condition their behavior on previous entrants' decisions. The relative performance of these mechanisms is investigated when entry is costly and selective, meaning that potential buyers with higher values are more likely to participate. A simple sequential mechanism can give both buyers and sellers significantly higher payoffs than the commonly used simultaneous bid auction. The findings are illustrated with parameters estimated from simultaneous entry USFS timber auctions where our estimates predict that the sequential mechanism would increase revenue and efficiency.

We document and study the strategy and performance of two distinct types of investors in Los Angeles between 1988-2009 who purchased homes with the intention of quickly re-selling: middlemen buy at below-market prices from motivated sellers and re-sell quickly; speculators target periods and areas of rapid market appreciation. Unlike that of middlemen, speculative activity increased sharply in the housing boom and was strongly associated with subsequent price bubbles at both the metropolitan and neighborhood levels. We present evidence suggesting that speculation fueled excess short-term appreciation rather than reflected the ability of speculators to predict when and where such appreciation would occur.

We develop and estimate an entry model for second price and open outcry independent private value auctions where potential bidders receive an imperfectly informative signal about their value prior to deciding whether to pay a sunk entry cost. In this way the model flexibly allows for selection on values, which will affect an entrant's subsequent competitiveness, at the entry stage. As signals become more informative, the entry process exhibits greater selection as firms with higher values are more likely to enter. We allow for asymmetries across bidders and unobserved heterogeneity across auctions, which are important features of most data sets. We show how incorrectly assuming the extremes of either no selection (no signal) or perfect selection (prior knowledge of one's value) - the commo...