A statistical analysis of reliability of audit opinions as bankruptcy predictors

Research measures the reliability of audit firms in predicting bankruptcy for
US-listed financial institutions. Object of the analysis is the Going Concern
Opinion (GCO), widely considered a bankruptcy warning signal to
stakeholders. The sample is composed of 42 US-listed financial companies that
filed Chapter 11 between 1998 and 2011. To highlight differences between
bankrupting and healthy firms, a matching sample composed by 42 randomly
picked healthy US-listed financial companies is collected. We concentrate on
financial institutions, whereas the existing literature pays considerably heavier
attention to the industrial sector. This research imbalance is remarkable and
particularly unexpected in the wake of recent financial scandals. Literature
points out two main approaches on bankruptcy prediction: 1) purely
mathematical; 2) approaches based on a combination of auditor knowledge,
expertise and experience. The use of data mining techniques, allow us to
benefit from the best features of both approaches. Statistical tools used in the
analysis are: Logit regression, Support Vector Machines and an Adaboost
Meta-algorithm. Findings show a quite low reliability of GCOs in predicting
bankruptcy. It is likely that auditors consider further information in supporting
their audit opinions, aside from financial-economic ratios. The scant predictive
ability of auditors might be due to critical relationships with distressed clients,
as suggested by recent literature.