The Effect of Prior Probability Disclosure and Information Set Construction on Bankers' Ability to Predict Failure

Abstract

Recent papers in this Journal have debated the effect of prior probability disclosures on the ability of bank loan officers to predict business failure. The debate was stimulated by results of such studies which appear to be in conflict (notably Casey [1980] with Zimmer [1980] and more recently Casey [1983] with Houghton [1984]). In almost identical extensions of Libby's [1975] banker prediction work, Zimmer's [1980] bankers achieved an overall failure/nonfailure prediction accuracy of 77.1% (a result consistent with Libby [1975]), whereas Casey's [1980] participants could only attain 56.7% accuracy. Casey's very low level of performance was largely linked to bankers' inability to correctly categorize failure firms (an average of only 4 out of 15 failed firms compared to 13 out of 15 nonfailed firms). Libby [1981] and Zimmer [1980] attributed these inconsistent results to the fact that Zimmer advised his bankers of the 50/50 failure/nonfailure split in the firms to be examined, while Casey provided no such information. Both Casey [1983] and Houghton [1984] then tried to determine whether the specification or nonspecification of priors could account for these different results. Houghton [1984] concluded that the prior notification of the failure/nonfailure split in his experiment was associated with a change in the bankers' prediction accuracy. However,