Auditors' Going Concern Opinions (GCOs) Are Noisy and Biased – How Can We Improve Them?

This paper draws on psychological theories and uses archival data to study the error, noise, and bias of auditors' Going Concern Opinion (GCO), which requires auditors' professional judgment. We find that for firms that go bankrupt the following year, GCO is inaccurate 45 percent of the time, about half of which comes from noise, a source of error without a systematic pattern, and the other half from bias, a type of error with a systematic pattern. We then explore two ways to reduce GCO's error, noise, and bias. The first approach is to assist auditor judgment with algorithms. The second approach aggregates opinions from artificial auditors. We find that advanced machine learning (ML) models perform best in reducing costs generated from GCO's error, noise, and bias (up to 44.5, 24.5, and 80.1 percent, respectively). Our study adds to the accounting literature by examining the quality of GCO and exploring ways to improve GCO from the perspective of error, bias, and noise