An Analysis of Database Challenges in Financial Misconduct Research

• In Event: New Measure, New Data, and New Question

Abstract

Many published studies use single event-type databases that are not designed to examine financial misconduct cases. This practice can lead to erroneous conclusions, when researchers fail to recognize the inadequacy of a single event type to capture the impact of a misconduct case lasting years and involving a wide variety of disparate disclosures. We explain and document five features common to four commonly used databases. First, the dates provided by these databases occur, on average, 150 to 1,017 calendar days after the initial revelation of the financial misconduct. Second, because these databases were assembled for a single type of event that is sometimes associated with financial misconduct, they capture, on average, only 6% to 36% of the value-relevant announcements associated with the cases of misconduct they capture. Third, these databases identify, on average, only 9% to 54% of the material financial misconduct cases that occur during their collection interval. Fourth, these databases include related announcements that the researcher must identify. Fifth, on average, 31% to 98% of the events in each database involve no financial misconduct and must be extricated from those involving misconduct. Using late revelation dates could lead analyses of financial misconduct to understate the average loss in share value between 56% t0 73%. Researchers who use the information from these single event-type databases without supplementing their sample with information from other sources will miss, on average, from 64% to 94% of the value relevant information associated with case of financial misconduct. The large numbers of omitted cases and possible inclusion of non-misconduct events affects the power and bias of empirical tests. We conclude with suggestions for how future researchers who rely on these databases to collect samples of financial misconduct can effectively manage these five database features.

Authors

• Jonathan M. Karpoff, University of Washington

• Allison Koester, Georgetown University

• D. Scott Lee, Texas A&M University

• Gerald S. Martin, American University