This article develops a Bayesian framework for estimating multivariate treatment effect models in the presence of sample selection. The methodology is applied to a banking study that evaluates the effectiveness of lender of last resort (LOLR) policies and their ability to resuscitate the financial system. This paper employs a novel bank-level dataset from the Reconstruction Finance Corporation, and jointly models a bank's decision to apply for a loan, the LOLR's decision to approve the loan, and the bank's performance a few years after the disbursements. This framework offers practical estimation tools to unveil new answers to important regulatory questions.

We examine how the threat of disclosure (stigma) changes the quality of banks that approach emergency lending facilities. We study a financial crisis where two confidential facilities were available to banks. Unexpectedly, a partial list of bank names from one facility was published, suddenly stigmatizing that facility. We find that the composition of banks that approached each facility changed, where the newly stigmatized facility attracted weaker banks that maintained smaller liquidity buffers, while the alternative confidential facility attracted both weaker and stronger banks. Our results shed light on how stigma prevents regulators from reaching many banks to inject critical liquidity into the banking sector during a crisis.

A nationwide panic forced President Roosevelt to declare a banking holiday in March 1933. The government reopened banks sequentially using a process that sent noisy signals about banks’ health. New microdata reveals that the public responded to these signals. Deposits at rapidly reopened banks rebounded quicker than at comparable or stronger banks that reopened even a few days later. The stigma of late reopening shifted funds from stigmatized to lauded banks and among communities that they served. Despite persisting over a decade, the shift had no measurable impact on the rate at which localities recovered from the Great Depression.