The PCAOB adopted critical audit matters (CAMs) to meet public demand for informative audit disclosure, but stakeholders are concerned this goal has not been achieved. We explore this disconnect via interviews with 30 highly experienced auditors. We find that audit firms expended considerable resources to implement CAM best practices. However, overwhelming institutional pressure gave rise to informal rules of thumb that prioritize symbolic comfort over substantive change. The first is don’t be an outlier , so auditors defer to the national office to ensure conformity and avoid PCAOB scrutiny. The second is report the “right” number of CAMs by never reporting zero and reporting at least one recurring CAM. The third is avoid surprises by communicating with the client to ensure that CAMs do not contain original information and allowing management to preempt auditor disclosures. Collectively, these rules yield CAMs that comply with PCAOB standards but do not provide new information and instead maintain the status quo.

Machine learning (ML) could strengthen banks’ resilience through improved credit risk screening and ultimately benefit financial stability. Yet, ML adoption in banking remains limited, with simpler linear models still predominating. We argue that the emergence of highly flexible ML models has created a new challenge for forecasting tasks: ‘model multiplicity’—where equally accurate ML models at the aggregate level produce divergent individual-level predictions (‘predictive multiplicity’) or differ in their decision surfaces (‘procedural multiplicity’). These issues raise fundamental questions: Why should an individual or firm be subject to an adverse credit risk model outcome when there is an equally accurate model that treats them more favorably? Using the world’s largest loss database of corporate defaults, we examine these two phenomena in recovery rate ( RR ) modeling and propose heterogeneous ML ensembles as a natural solution. By combining predictions and decision surfaces from multiple well-performing ML models, ensembles mitigate risks associated with predictive multiplicity by ensuring that borrowers are not subject to the fluctuations of a single model, and reduce procedural multiplicity by providing a robust measure of features that ultimately improve out-of-sample RR predictions. By addressing the ‘multiplicity of good models’ problem, our study emphasizes the importance of model stability and provides new insights for the future development of ML models.

Alex M. Thomas of School of Arts and Sciences, Azim Premji University reviews “Relocating Development Economics: The First Generation of Modern Indian Economists” by Maria Bach. The Econlit abstract of this book begins: “Explores the work of the Early Nationalists, the first generation of modern economists in India in the mid- to late-nineteenth century, focusing on how they produced forward-thinking knowledge on economic development.”

The effect of health insurance on consumption risk depends in part on how it interacts with other risks beyond health care cost risk, such as income risk. Using a variety of approaches, I find that for U.S. households, the interaction with other risks transforms the risk protection from health insurance. Standard contracts amplify the impact of other risks, due to both subsidizing normal goods and undoing the protection against other risks from discounts, charity care, and bad debt. Alternative contracts that account for other risks, such as contracts that limit out-of-pocket spending relative to income, can provide better risk protection.