Much of the information available to participants in speculative markets is in the nature of expert opinion, analysis, professional advice, and so on. Markets discount widely held factual information very well; this paper studies market efficiency with respect to subjective information. We examine the "market" for bets on thoroughbred horse races to determine whether the published forecasts of professional handicappers are completely discounted. A multinomial logit probability model is used to measure the information content of the forecasts, and we find that they do contain considerable information but that the track odds generated by betting discount almost all of it. Within the population of bettors, those betting at the track appear to discount the handicapper information fully, but those betting through New York's off-track betting system do not.