In 2005, the PCAOB began inspecting foreign auditors in their home countries if their clients are public companies in the U.S., marking the first U.S. securities regulation to systematically conduct on-site oversight beyond domestic borders. 

We examine whether this policy, intended to enhance audit quality, unintentionally discouraged foreign firms from listing in the U.S. We find that the heightened oversight redirected firms from the U.S. to alternative markets. The deterrent effect (estimated on the right) is concentrated among firms facing high compliance costs or proprietary information concerns. However, we find no evidence that these firms are of lower quality.

The SEC allows firms to redact filings to protect proprietary information, contingent upon SEC review. Introduced in the SEC’s first Staff Legal Bulletin in 1997, this procedure has served as the primary mechanism for protecting proprietary information and has been used by roughly one-third of public firms every year for over 25 years. 

The 2019 FAST Act relaxed this procedure by allowing firms to redact without SEC review, aiming to reduce the regulatory burden for proprietary information protection. We show that this change transformed the redaction market into a lemons market: when oversight was reduced, low-proprietary firms exploited the system to conceal bad news, while legitimate redactions by high-proprietary firms were crowded out of the market. The figure on the right illustrates these contrasting effects.

Look familiar? Figures on the right are all pseudo treatment effects driven by pre-existing differences between the treatment and control groups (aka pre-trends), yet they still manage to pass the standard pre-trends test!

Pre-treatment trends threaten inferences in difference-in-differences (DiD) regressions, which have become ubiquitous in accounting research. We survey recent studies in accounting employing DiD, finding that many use short pre-treatment windows and test few parameters for evidence of pre-treatment trends. Using both simulated data and actual “shocks” employed in the literature, we show that this approach has low power and significantly increases the risk of reporting biased treatment effects. We evaluate various strategies to both identify and control for pre-treatment trends, which we find significantly affects inferences in common accounting settings. 

Large technology companies (BigTechs) have expanded beyond their core businesses into the lending market. As a distinctive segment within the Financial Technology (FinTech) sector, BigTechs had surpassed all other FinTechs in global lending volume by 2019, becoming three times larger than the rest of the sector combined.

To understand this trend, we leverage a randomized setting and highly granular transaction-level data to examine the impact of BigTech lending business on the BigTech’s core business. We find that credit provision significantly enhances the core business by increasing user engagement, transaction volume, and retention. Further analysis reveals that credit improves the user well-being by relaxing liquidity constraints and facilitating lending-core business synergies without increasing delinquency.

We develop a novel continuous-time Deep Neural Network to model digital footprints for dynamic credit risk assessment in BigTech lending. Our method handles irregular and sporadic consumer activities, enabling real-time prediction of default risk. By jointly estimating Probability of Default and Exposure at Default through an end-to-end loss function, the model offers a more comprehensive view of credit risk. Using longitudinal data from a leading BigTech platform, we demonstrate that our approach significantly outperforms traditional credit scoring models, particularly among high-risk users.