Managing Stigma during a Financial Crisis

Sriya Anbil∗†

Abstract

How should regulators design effective emergency lending facilities to mitigatestigma during a financial crisis? I explore this question using data from an unex-pected disclosure of a partial list of banks that secretly borrowed from the lender oflast resort during the Great Depression. I find evidence of stigma in that depositorswithdrew more deposits-to-assets from banks included on the list in comparison tobanks left off the list. However, when all banks were simultaneously revealed to haveborrowed, there was no stigma. Overall, the results suggest that an emergency lendingfacility that reveals bank identities simultaneously will mitigate stigma. (JEL Codes:G01, G21, G28)

∗Board of Governors of the Federal Reserve, Division of Monetary Affairs, 20th and Constitution Ave.NW, Washington, DC 20551, USA. sriya.l.anbil@frb.gov†This paper is part of my dissertation at Yale University. I thank Heather Tookes (co-chair), Gary Gorton

(co-chair), Andrew Metrick, and Justin Murfin for valuable advice and comments. I would also like to thankseminar participants at the Yale School of Management, Board of Governors of the Federal Reserve, OlinBusiness School, Federal Reserve Bank of Cleveland, Anderson School of Management, Kelley School ofBusiness, Columbia Business School, Office of Financial Research, Federal Reserve Bank of New York, TheWharton School, Leeds Business School, McCombs Business School, and the Scheller College of Businessfor their useful comments. This paper has also benefited from discussions with Claire Brenneke, JuliaGarlick, Sean Hundofte, Wenxi Jiang, Steve Karolyi, Toomas Laarits, Margaret Leighton, Steve Malliaris,Corina Mommaerts, Marina Niessner, Andrew Sinclair, and Jacqueline Yen. I also thank Nikhil Seval, ChenWang, and Michaela Pagel for crucial data access, the Yale International Center for Finance and WhiteboxAdvisors for research funding, and Haruna Fujita and William Longinotti for excellent research assistance.All remaining errors are my own. The views expressed in this paper are solely the responsibility of the authorand should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System,its staff or any other person associated with the Federal Reserve System.

1

1 Introduction

During the recent financial crisis, the Federal Reserve (Fed) was frustrated in its efforts

to inject liquidity into the financial system through its main emergency lending facility, the

discount window (Armantier et al. (2014)).1 At the time, the Fed Chairman Ben Bernanke

explained that “the banks’ concern was that their recourse to the discount window, if it

somehow became known, would lead market participants to infer weakness–the so-called

stigma problem (Bernanke (2009)).” The consequences of approaching the discount window

could be costly and might even lead to bank closures (Madigan (2009)). As a result, banks

are reluctant to borrow from the lender of last resort (LOLR) because they fear the public

reaction, which prevents the Fed from inserting liquidity into critical parts of the banking

sector during a financial crisis.

An important challenge for all central banks is the design of emergency lending facilities

that alleviates the “stigma problem.”2 The stigma problem is not new—arguably, it goes

back to the beginning of the 1920s. At the time, the Fed emphasized that lending from the

discount window should only be temporary, implying that a bank that borrowed from the

discount window must be in trouble (Gorton and Metrick (2013)), which has complicated

LOLR policy ever since. Although central bankers are concerned with the stigma problem,

direct evidence of its existence is limited. In this paper, I ask two related questions. First, are

banks stigmatized after news is revealed that they received emergency loans from the LOLR?

Second, if they are, how can the LOLR effectively control the information environment to1In the U.S., the Fed’s traditional means of providing emergency credit to depository institutions is

through its discount window. Lending from the discount window is in the form of promissory notes securedby adequate collateral (Furfine (2003)).

2Stigma was a central topic discussed during Federal Open Market Committee (FOMC) meetings. BenBernanke announced “...we have been looking at... how to address the stigma of the discount window. Arethere ways to provide liquidity that would help normalize money markets... but would avoid the stigma andcreate a more efficient system? (FOMC (2007))”

2

manage public reactions during a crisis? There are two requirements that must be met

in order to answer these questions. First, the researcher needs to be able to compare the

outcomes of banks that borrowed from the LOLR whose identities were revealed to similar

banks that also borrowed but whose identities were not revealed. Second, the researcher

needs to know specifically when the banks’ identities were revealed. No setting exists during

the recent crisis where both requirements are met.

I address these two requirements by using an unexpected event from the Great Depression.

On August 22, 1932, the Clerk of the House of Representatives took it upon himself to

publish a partial list of banks that had secretly borrowed from the Reconstruction Finance

Corporation (RFC), the LOLR during the Great Depression. Using a unique hand-collected

dataset of balance sheet information for 2,674 banks, I implement a difference-in-differences

approach to estimate the effect of the publication of the list on the deposits-to-assets ratio

and other bank outcomes. To do so, I compare the performance of those banks included on

the list (“the treatment group” or “revealed banks”) to banks not included on the list (“the

control group” or “non-revealed banks”). Banks were included on the list according to the

date of their loan authorization; non-revealed banks had a loan authorized by the RFC before

revealed banks. This framework meets both requirements. First, all banks in the analysis

borrowed from the RFC which eases any sampling problems by allowing me to compare

revealed and non-revealed banks that borrowed from the RFC. Second, the public learned

the identities of borrowers on August 22, 1932, via the publication of the list. Consequently,

the mechanism and timing of how the public discovered news of emergency loans is clear.

I first ask whether depositors stigmatized banks after their emergency loans were revealed

in an era before deposit insurance. I find results consistent with depositors interpreting news

of a LOLR emergency loan as financial weakness; the deposits-to-assets ratio dropped by

3

5-7% at revealed banks compared to non-revealed banks. Moreover, revealed banks were

61.4% more likely to fail after the list was published and, conditional on failing, the time-to-

failure for revealed banks was nearly five weeks sooner than non-revealed banks. Altogether,

the results provide evidence that deposits imposed a liquidity risk on revealed banks and

increased the speed and likelihood of their failure.

To the best of my knowledge, this paper is the first to identify and quantify the magni-

tude of stigma that banks experienced after their emergency loans were revealed to market

participants. Specifically, in 1932, banks on average lost 81% of their deposits before failing

(Wicker (1996)) while, at the same time, I find that the presence of stigma imposed a 5-7%

loss in the deposits-to-assets ratio at revealed banks. Accordingly, the marginal contribu-

tion of stigma towards the likelihood of failure was considerable and is significant enough to

warrant the attention of policy makers.

Having found evidence of the stigma problem, I next ask how policy makers can use this

information to improve the design of emergency lending facilities and manage the informa-

tion environment during a crisis to alleviate the problem. One possibility is to resolve a

coordination problem between banks that results from banks reluctance to borrow from the

discount window (Armantier et al. (2014)). Because of stigma, a bank’s decision to borrow

from the LOLR involves predicting the actions of other banks. If a bank believes that other

banks are reluctant to borrow, its own belief in stigma is reinforced, and, as a result, the

bank decides not to borrow from the LOLR. To identify the optimal policy design to address

this coordination problem, I must determine the conditions under which market partici-

pants did not withdraw more deposits from revealed banks than non-revealed banks. This

requires a setting that can compare market participant reactions in defined environments

where researchers can randomize how news is released from the LOLR to the public.

4

No such setting is available from the recent crisis, but it is available during the Great

Depression. Because banking markets during the 1930s were defined by city’s boundaries

(Heitield et al. (2013)), I can compare depositor reactions between two types of defined

banking markets, i.e. cities, where the important variation between each type of city was

how the information about RFC borrowers was released. The first type of banking mar-

ket includes cities where nearly all banks were publicly revealed to have borrowed from the

LOLR, while the second type includes cities where only few banks were publicly revealed. I

find that the presence of stigma was larger when information about bank borrowers was re-

vealed asymmetrically. In cities where few banks were publicly revealed, depositors withdrew

significantly more funds at revealed banks than non-revealed banks. In these cities, infor-

mation was revealed asymmetrically and depositors were more likely to interpret the news

as reflective of each bank’s financial weakness rather than reflecting an aggregate adverse

shock. However, in cities where nearly all banks were publicly revealed, depositors withdrew

their funds from all banks; indeed, they then stored their funds in the US postal savings

system (O’Hara and Easley (1979)).3 In these cities, information about bank borrowers was

revealed simultaneously.4 I find that in cities where nearly all banks were revealed (informa-

tion was revealed simultaneously), the presence of stigma was mitigated. Deposit outflows

in these cities were lower than cities where few banks were revealed. Because the presence of

stigma was larger in cities where few banks were publicly revealed (information was revealed

asymmetrically), this explains why banks are reluctant to borrow from the discount window.3The US postal savings system was the only bank to have its deposits fully insured by the federal

government, was intended to attract rural depositors, and was designed to limit its competition with privatebanks.

4In cities where more than 66% of banks were revealed, the deposits-to-assets ratio did not drop atrevealed banks compared to non-revealed banks. However, in cities where fewer than 33% of banks wererevealed, the deposits-to-assets ratio dropped by 6.89 percentage points at revealed banks. These results arerobust to using different thresholds.

5

Therefore, designing emergency lending facilities that reveal information about bank bor-

rowers simultaneously may be associated with less stigma. This can be achieved by providing

a coordination mechanism at the lending facility for banks to request loans simultaneously.5

In the case that identities are revealed, it is likely that more than one bank identity would

be simultaneously revealed, thus preventing an individual bank from being targeted when

approaching the LOLR for a loan. The design of the facility implies that revealing infor-

mation simultaneously (which can take the form of all banks being revealed or no banks

being revealed) dominates revealing information asymmetrically (where few banks are re-

vealed).6 Moreover, such lending facilities will solve the coordination problem that arises

because of stigma. Market participants would find banks receiving loans from the LOLR to

be acceptable simply because many other banks were receiving loans as well. Finally, these

conclusions also imply one rather extreme solution to the stigma problem. Central banks

could force all banks to receive capital injections, leading the public to infer that receiving

LOLR funds was acceptable since all other banks would be receiving funds as well.7

After considering depositor reactions to the August 22, 1932 publication of the list of bank

identities, I examine the behavior of those banks whose identities were revealed. An aim of

the central bank is to ease liquidity needs for solvent yet illiquid banks during a financial crisis

where, in turn, these banks use the funding to maintain loans to their consumers (Freixas

et al. (1999)). After the revelation, central banks might be concerned that revealed banks

would exhibit precautionary behavior by hoarding cash, consistent with banks taking self-

insurance (Winters (2012)). Did banks hoard cash after their identities were revealed? I find5New lending programs created by the Fed during the recent crisis provided such a coordination mecha-

nism.6Although I do not explicitly show that keeping all bank identities private dominated revealing all bank

identities at the same time, Armantier et al. (2014) infers that this was the case because banks believed instigma.

7The feasibility of this policy implication may be unlikely.

6

that revealed banks increased their bonds-and-securities-to-assets ratio between 10-12.5%

compared to non-revealed banks. Moreover, revealed banks dropped their total loans-and-

discounts-to-assets ratio by 4.5-6% in comparison to non-revealed banks. My findings show

that central bank goals were circumvented when a bank’s identity was publicly revealed

because they exhibited precautionary behavior.

Because researchers face major challenges measuring stigma, the limited size of the lit-

erature reflects this. Furfine (2003) finds preliminary empirical evidence from the federal

funds market that banks were reluctant to borrow from the discount window. Armantier

et al. (2014) extend this result by showing that banks are willing to pay a premium in ex-

cess of 44 basis points on average to avoid borrowing from the discount window. Finally,

Kleymenova (2013) finds little evidence of stigma after the disclosure of discount window

borrowers on March 31, 2011. She finds that the disclosure decreased bid-ask spreads and

revealed banks’ cost of capital.8 A possible reason for the lack of stigma in 2011 is because

the disclosure occurred well after the financial crisis had ended. Asymmetric information

released in non-crisis times may have less impact because financial fragility is low. However,

the same asymmetric information released during a crisis can cause a decline in output and

consumption (Gorton and Ordonez (2014)), shedding light on why stigma may exist only

during times of economic stress.

This paper also contributes to the literature on the effect of information disclosure from

the LOLR on bank runs and on the need for banks to keep secrets. Hautcoeur et al. (2013)8Kleymenova (2013) is the first empirical paper to study the effect of the Federal Reserve’s publication

of information about banks from the recent crisis. Bloomberg News filed a lawsuit against the Fed onNovember 7, 2008 to gain access to the list of discount window borrowers. The Fed argued that doing sowould stigmatize those borrowers. The Fed lost the three-year, well-publicized lawsuit, and on March 31,2011, it released the names of the discount window borrowers. Unfortunately, it is difficult to measure thecost of stigma using this instance of disclosure because of methodological limitations due to the lack ofappropriate empirical setting during the recent crisis.

7

show that the French LOLR public bailout of the Comptoir d’Escompte, a large bank in

France, in 1889 did not turn into a general panic. Because all major banks were involved

in designing the bailout, the public interpreted banks receiving emergency loans as accept-

able. Furthermore, during the National Banking Era (1863-1913) in the U.S., private clear-

ing houses responded to severe panics by suppressing bank-specific member information in

newspapers, and published only aggregate information about the clearing house (Gorton and

Talman (2014)). My paper offers an explanation for these historical examples by showing

that the differential effects of partial and full disclosure of banks that accept emergency

credit can have strong adverse impacts on the health of the financial system.

Finally, this paper provides insights into the success of several new emergency lending

facilities introduced by the Fed during the recent crisis. The new facilities—The Term

Auction Facility (TAF), the Term Securities Lending Facility (TSLF) and the Primary Dealer

Credit Facility (PDCF)—all mitigated stigma by providing coordination mechanisms for

banks to submit bids for LOLR loans simultaneously through auctions. If leaks were to

occur, all borrowers’ names would be divulged at the same time. My results show that

such coordination and simultaneous disclosure help explain why these new facilities were

associated with less stigma than the discount window (Benmelech (2012)). By quantifying

stigma, central bankers can prioritize whether solving the stigma problem is necessary. I find

that the design of an emergency lending facility that allows banks to coordinate requests for

LOLR loans can mitigate stigma.

The remainder of the paper is organized as follows. Section 2 reviews the relevance of

stigma during the recent crisis. Section 3 presents the publication of the list on August

22, 1932, which is the main source of identification used in this paper. Section 4 describes

the data. Section 5 presents the results of testing for stigma, and Section 6 presents the

8

conclusions of how depositor reactions changed between cities in response to how information

was released about RFC borrowers. Section 7 describes how revealed banks hoarded cash

following the publication of the list, and Section 8 concludes.

2 Stigma

In the following section, I discuss how central banks attempted to manage the stigma

problem before 2008 but failed to adopt a systematic method of releasing information about

banks. This lack of policy has caused differences in how the public perceives news of gov-

ernment assistance.

2.1 Relevance during the recent crisis

I define stigma as the cost imposed on a bank after its emergency LOLR loan was revealed to

the public. The belief in stigma can arise in a rational equilibrium. Because private trades

in the interbank market are not observable by the public, information from the interbank

market, which would otherwise remain private, can flow to the public through reported

activity at the discount window (Ennis and Weinberg (2013)). In equilibrium, banks holding

bad assets cannot obtain loans in the interbank market and are forced to borrow from the

discount window. Consequently, adverse selection arises at the discount window making

stigma an equilibrium phenomenon.

During the recent crisis, central bankers believed the primary reason banks refused to

borrow from the discount window was because of stigma (Geithner (2014)).9 The Fed’s

priority of solving the stigma problem was also illustrated by its extreme reluctance to

release discount window borrowers after losing the Freedom of Information Act lawsuit filed9Geithner (2014) explains how banks’ belief in stigma hindered the effectiveness of the Fed in 2007.

9

by Bloomberg News. The Fed had argued that releasing the identities of discount window

borrowers would stigmatize banks and impede the Fed’s ability to respond to future crises

(Berry (2012)). Moreover, the Fed introduced new lending programs designed to eliminate

the perception of stigma (Armantier et al. (2014)). The Term Auction Facility (TAF) was the

first new lending program created on December 12, 2007. It seemed that banks associated

less stigma with the TAF than the discount window because total borrowings at the TAF

were much higher (Benmelech (2012)). However, the precise reason that the TAF seemed

to carry less stigma (or be preferable to banks) was unknown. The TAF and the discount

window were identical lending facilities with respect to collateral requirements and borrower

eligibility, and both also protected borrower identities. However, the borrowing rate at the

TAF was set through competitive bidding at auction whereas the discount window borrowing

rate was determined by the Fed. Under the TAF, banks were required to approach the

Fed at the same time; bids were submitted to the Fed every Monday, multiple winners

were announced on Wednesday, and the funds were received on Thursday.10 It provided

a coordination mechanism for banks to submit bids for loans simultaneously decreasing

the probability that an individual bank’s identity would be leaked. On the other hand,

the discount window required that banks approached for a loan individually providing no

coordination mechanism across banks that might mask individual identities. The relevance

of stigma in the recent crisis was significant.

How have market participants perceived news of government assistance before 2008? Cen-

tral banks have chosen different approaches to managing information about bank borrowers

with varying consequences. The first approach involved a bank identity being accidentally

leaked. On September 13, 2007, the BBC News revealed that Northern Rock bank had10Both the TSLF and PDCF also determined the borrowing rate through competitive bidding at auction

where bids were submitted by banks simultaneously.

10

secretly received assistance from the Bank of England resulting in depositors queuing to

demand their funds. A parliamentary inquiry into the leak concluded that depositors had

interpreted emergency assistance to Northern Rock as a sign of financial weakness and ran

on the bank (Treasury (2008)). The second approach involved requiring all banks to accept

LOLR capital injections while revealing their identities and loan amounts simultaneously. In

October 2008, the US Treasury simultaneously announced the identities and loan amounts

of nine banks it was investing $125 billion of TARP funding.11 This action prompted an

11% increase in the Dow Jones Industrial Average and the nine TARP recipient banks were

not stigmatized for receiving LOLR funding. From these examples, it appears that central

banks follow both asymmetric and symmetric methods of releasing information. It would be

useful to adopt a policy that best manages public reactions to news during a crisis.

3 Identification Strategy

3.1 The Reconstruction Finance Corporation

The Reconstruction Finance Corporation (RFC) was the lender of last resort to the US

banking system during the Great Depression. It was created by President Hoover in response

to a series of banking failures in 1931 after Britain left the gold standard (Friedman and

Schwartz (1963)).12 Figure 1 illustrates the steady drop in deposit levels after the stock11See the New York Times “US Investing $250 Billion in Banks” from October 13, 2008. The Troubled

Asset Relief Program (TARP) was signed into law on October 3, 2008 and was designed to purchase assetsand equity from banks to strengthen the financial sector.

1261% of banks failed between 1930 and 1933 (Wicker (1996)) while available customer deposits dropped by34% (FDIC (1920-1936)). Bank failures mostly ended after President Roosevelt declared a national bankingholiday during March 6-13, 1933. While some banks were allowed to reopen on March 13, 25% were notlicensed to reopen after the holiday. Over half of the surviving banks after the national banking holiday were

11

market crash in October 1929 and accelerations of bank suspensions in 1931. President

Hoover submitted the RFC Act to Congress on December 7, 1931 and it was passed into law

on January 22, 1932.

Figure 1: Trends of Deposits and Banks from 1921-1933

Source: Wicker (1996) p.2

This figure describes the banking sector in the US from 1921-1933. The line illustrates total deposits(in millions) across all banks in the US over time. The solid bars depict the total number of banksin the US while the dashed bars reflect the total number of suspended banks in the US. The dataare from Wicker (1996).

The Reconstruction Finance Corporation began secretly authorizing loans on February

2, 1932 to several types of institutions including commercial banks, insurance companies,

receivers of closed banks, and a variety of nonfinancial firms including state and local govern-

ments, railroads, and mortgage institutions. Loan interest rates and collateral requirements

were identical to the discount window and did not change until 1933 (Ebersole (1933); Olson

estimated to have borrowed from the RFC at least once (Todd (1992)).

12

Figure 2: Reconstruction Finance Corporation Loans Outstanding

Source: Gorton and Metrick (2013)

This figure illustrates the scale of loans from the Reconstruction Finance Corporation to banks aswell as to other institutions like state and local governments, railroads, and mortgage institutions.The vertical line depicts when borrower names were first disclosed on August 22, 1932. The dataare from Gorton and Metrick (2013).

(1977)).13 Figure 2 illustrates the scale of loans outstanding authorized by the RFC to banks

and nonfinancial firms. Prior to the August 22, 1932 publication of the list of bank identities

that had secretly borrowed from the RFC, total RFC borrowing had reached approximately

$1 billion, with about half of this total going to banks. Following the publication of the list,

net bank borrowing flattened even though nonfinancial borrowing, where stigma seems far

less of an issue, rose to more than $2 billion (Gorton and Metrick (2013)).

Most banks borrowing from the RFC were ineligible to borrow from the discount window

because they were not Federal Reserve member banks. Large reserve banks in cities tended

to be member banks (Mitchener and Richardson (2013)) and nearly 20% of my sample were

eligible to borrow from the discount window. Figure A.1 in the Online Appendix shows that

total borrowings from the Federal Reserve decreased in 1932 after the introduction of the13For more detailed information about the RFC see Olson (1977), Butkiewicz (1995); Mason (2001), Mason

and Schiffman (2003), and Calomiris et al. (2013).

13

RFC. Loans outstanding from the Fed totaled $817 million in January of 1932 and decreased

to $254 million by December.

By the end of 1932, 6,865 eligible institutions (banks and nonfinancial firms) had been

authorized over $1.6 billion in loans by the RFC (Corporation (1932)). The Reconstruc-

tion Finance Corporation’s function as a lender of last resort was large and economically

significant.

3.2 The publication of the list on August 22, 1932

The main source of identification in this paper is the unexpected publication of a list of

banks on August 22, 1932 that had secretly borrowed from the Reconstruction Finance

Corporation. I analyze the response of depositors and banks to the publication of the list

in a difference-in-differences (DID) setting. In this framework, I avoid the sample selection

problem of comparing the performance of banks that chose to borrow from the LOLR to the

performance of banks that chose not to borrow, because all banks in the estimation borrowed

from the RFC.

Initially, the identities of all RFC borrowers (banks and non-banks) were kept secret from

the public. Since its establishment, the RFC had used elaborate secret codes to transmit

messages to its loan agencies and individual banks (Olson (1977)). However, in July 1932,

an amendment was introduced to the RFC Act that extended Reconstruction Finance Cor-

poration authority into state and local relief, public works construction, slum clearance, etc.

In this amendment, a last minute clause was added requiring that all loans authorized by

the RFC be made public to Congress. President Hoover initially planned to veto the bill

due to the addition of the last-minute clause but was assured by the Senate Majority Leader

14

that RFC loans would not be revealed to the public without congressional approval. It was

decided that monthly reports of RFC loans would be of confidential nature and held by the

Clerks of the Senate and the House of Representatives until Congress resumed session in

December (Chronicle (1932b)).

Despite this, on August 22, 1932, South Trimble, the Clerk of the House of Representa-

tives, took it upon himself to release a partial list of bank identities that accepted loans from

the RFC because he felt it was his duty to inform the US public. The list was first published

in the New York Times and the Commercial & Financial Chronicle (CFC) and coverage of

this list was widespread.14 It was highly likely that the publication of the list was unexpected

given the last-minute nature of the clause and the assurances that no borrower list would

be released without congressional approval.15 The loan authorization date for a bank deter-

mined whether or not the bank identity was revealed. Mr. Trimble decided to reveal banks

that had a loan authorized between July 21 and July 31, 1932 on August 22, 1932 to coincide

with the passing of the amendment to the RFC Act on July 20, 1932. He regarded the details

of any loan authorized after July 20, 1932 to be subject to the last-minute clause (which was

then law), and required congressional approval for release (Chronicle (1932a)). Banks not

revealed on August 22 had a loan authorized on or before July 20, 1932 and therefore were

not subject to the last-minute clause. Therefore, non-revealed banks had a loan authorized

between February 2 and July 20, 1932. Figure 3 displays a timeline illustrating how banks

are assigned to the treatment and control groups. The treatment group includes 351 banks

revealed on August 22, 1932 while the control bank includes 713 banks not revealed, yielding

1,064 banks in total.14I confirm this by checking several local newspapers such as the Hartford Courant and the Ames Tribune

to determine when the list was published.15Newspaper articles do not discuss the revelation of bank borrower identities to the public until Mr.

Trimble forewarns of his intended announcement on August 18, 1932.

15

Figure 3: Timeline by Loans Authorized

This figure illustrates how the treatment group (revealed banks) and control group (non-revealedbanks) were assigned. Banks with loans authorized between July 21 and July 31, 1932 were revealedon August 22, 1932 while banks with loans authorized on or before July 20, 1932 remained notrevealed.

Since Congress was not in session, Mr. Trimble followed the August 22, 1932 list publica-

tion with five additional monthly lists finishing on January 26, 1933. In this paper, I focus on

the publication of the first list on August 22, 1932 because after this date, a bank choosing to

borrow after August 22 may have anticipated that their identity would be revealed following

their loan authorization. For the remaining discussion, I refer to banks published on the

August 22, 1932 list as “the treatment group or revealed banks.” Banks not revealed as of

this date are referred to as “the control group or non-revealed banks.” All results presented

will only include banks with loans authorized before August 22, 1932.

It is important for my identification that the list of revealed banks published on August

22, 1932 be chosen by Mr. Trimble in a way that is uncorrelated with the outcome variables

used in the estimation. There should be nothing systematically important about the dates of

loan authorizations that he chose to publish implying that the decision of when a bank chose

16

to borrow from the RFC also needs to be uncorrelated with all outcome variables. Otherwise,

the treatment and control groups may differ along a number of observable dimensions and

this will bias the results of the estimation.

I perform four robustness checks to test the validity of my identification strategy. First,

I restrict the control group to only include banks with loans authorized between July 10 and

July 20, 1932 and compare the performance of these banks to the treatment group. In this

test, I compare short symmetric windows of authorized loans. Second, I exploit differences

in the timing of loan authorization and loan application dates. Mr. Trimble revealed banks

with loans authorized between July 21 and July 31, 1932 which does not necessarily correlate

with the actual date the bank applied for the loan. Historical evidence suggests that the time

period between loan application and authorization varied across banks. For example, Olson

(1977) discusses how RFC board examiners would visit banks to discuss a loan application

and assess the bank’s financial condition. He explains that a loan authorization may be

delayed if there was not a prior relationship between the bank and the RFC.16 As long as

the period between loan application and authorization varied across banks, the treatment

and control groups were assigned quasi-randomly. Third, I restrict the treatment group to

banks with a loan authorized during the same time period as the control group, i.e. between

February 2 and July 20, 1932. I then compare the performance of these banks to the control

group. Fourth, a crucial assumption for the DID estimation to be valid is that revealed banks

and non-revealed banks should have parallel trends were it not for the publication of the list.16Olson (1977) discusses the extreme caution displayed by the RFC when distributing funds to banks.

This resulted in a delay to when a bank received a loan. The delay would sometimes result in the governor ofstate of the bank’s location intervening to request the RFC to send funds expeditiously, e.g. the governor ofTennessee intervening on behalf of the Bank of America and Trust Company in Memphis. He also describeshow the RFC would dispatch its board examiners to evaluate the conditions of banks, e.g. the UnionGuardian Trust Company in Detroit; Fidelity National Bank in Kansas City; and Hibernia Bank and TrustCompany in New Orleans. This process would obviously delay the process of when a bank received RFCfunds.

17

In the absence of the event, the observed DID estimates should be zero. In Figure 4, I plot

the average change in the deposits-to-assets ratio at revealed and non-revealed banks from

June 30, 1930 to June 30, 1933, along with 5% confidence intervals. The graph shows that

both groups follow parallel trends providing more evidence that the treatment and control

groups do not differ with respect to the deposits-to-assets ratio before the publication of

the list. The deposits-to-assets ratio then dropped sharply for revealed banks compared to

non-revealed banks following the publication of the list.

Figure 4: Trends in the deposits-to-assets ratio

This figure displays the average change in the deposits-to-assets ratio for the treatment group (351revealed banks) and control group (713 non-revealed banks), normalized by deposit levels at June30, 1930. The dashed lines denote 5% confidence intervals. The darker shaded area between June30, 1932 and December 31, 1932 denotes the period the treatment group was revealed (August 22,1932). The second lighter shaded area denotes the period the control group was revealed (January26, 1933).

18

3.3 The publication of the retroactive list on January 26, 1933

On January 26, 1933, Mr. Trimble published a final unexpected retroactive list in the New

York Times that included names of RFC borrowers with loans authorized on or before July

20, 1932. This list contained the names of the 713 banks not revealed as of August 22,

1932. Therefore, all banks in the sample were eventually revealed. Consequently, all of my

comparisons of revealed banks to non-revealed banks focus on outcomes between August 22,

1932 and January 26, 1933, before the control group was revealed. After January 26, 1933,

no obvious control group to compare the performance of banks revealed on August 22, 1932

exists. Figure 5 displays a timeline of the lists published on August 22, 1932 and January

26, 1933. See Figure A.2 in the Online Appendix for a timeline of all list publications.

Figure 5: Timeline of the August 22, 1932 List Publication

This figure displays the timeline of publications of bank borrower lists after the RFC began givingloans on February 2, 1932. 351 banks were revealed on the list published on August 22, 1932(“treatment group” or “revealed banks”) while 713 banks were not revealed (“control group” or“non-revealed banks”). However, the 713 non-revealed banks as of August 22, 1932 were retroac-tively revealed to the public on January 26, 1933. All banks in the sample were eventually revealedto the public.

19

4 Data

4.1 Reconstruction Finance Corporation loan data and bank bal-

ance sheet information

The RFC loan authorization data and bank identities are from the New York Times (Times

(1932)).17 In the original publication, bank identities were organized by state because bank

names were common across states. Loan amounts and the interest rate were also provided.

The date of loan authorization was provided for loans authorized between February 2 and

July 20, 1932. Following July 20, only loan authorization intervals were provided. The data

are hand-collected and then verified in the Commercial & Financial Chronicle.

Bank balance sheet data are from Rand McNally Bankers’ Directory which was published

every six months. I collect the amounts of paid-up capital, surplus and profits, deposits, other

liabilities, loans and discounts, bonds and securities, miscellaneous, and cash due from other

banks, and the name of the president for each bank. The coverage of the Directory seems to

be comprehensive across the U.S. as only 18 banks that had loans authorized from the RFC

were not matched in the Directory. The data are hand-collected from eight books beginning

December 31, 1929 to June 30, 1933 resulting in eight observations per bank. For roughly

5% of the data, bank balance sheet observations are repeated. I exclude this repeated data,

but in a robustness check I ensure this exclusion does not affect any of the results. I also

hand-collect the number of banks in a city as of June 30, 1932.

The three bank-specific outcomes of interest in this paper are the deposits-to-assets,

loans-and-discounts-to-assets, and bonds-and-securities-to-assets ratios. I do not observe

the detailed composition of deposits, loans and discounts, or bonds and securities. The17The list also includes identities of non-banks such as railroads, state relief, and mortgage companies.

20

total book value of deposits provided in Rand McNally Bankers Directory equals demand

deposits and time deposits. The total book value of loans-and-discounts equals the book

value of loans to banks and trust companies, loans on securities (to brokers and dealers),

real estate loans and mortgages, and all other loans. Finally, the total value of bonds-and-

securities in the Directory equals the book value of US government securities, state, county,

municipal bonds, railroad bonds, all other bonds, stock of the Federal Reserve Bank, stock of

other corporations, foreign government bonds, other foreign securities, the real estate value

of the banking house, other real estate owned, and the value of furniture and fixtures within

the banking house.

In addition, I collect the date a bank failed, was suspended (in conservatorship), was

merged, or was bought from Rand McNally Bankers’ Directory. Bank failures during this

time period must be recorded carefully because suspended banks may have reemerged under

new leadership and should not be considered failed banks (Calomiris and Mason (2003)).

I supplement the information from the Directory by using Moody’s volumes from this pe-

riod which generally provided information on Federal Reserve member banks. In the data,

receiverships with a confirmed failure date and voluntary liquidations are treated as bank

failures, identical to the methodology used by Calomiris and Mason (2003). Because Rand

McNally Bankers’ Directory only reports the date a bank failed (suspended) and not its bal-

ance sheet information directly before its failure (suspension), I record these balance sheet

values as zero. However, as a robustness check, I proxy for the deposits-to-assets ratio,

the main variable of interest used in this paper, at failed banks with the total amount of

suspended deposits divided by total deposits in the state of the bank’s location.

For all lists published between August 22, 1932 and January 26, 1933, my data are 4,983

loans to 3,353 unique banks. I match 3,335 banks to balance sheet information in Rand

21

McNally Bankers’ Directory. I exclude banks which failed before August 22, 1932 (the RFC

was allowed to authorize loans to banks under conservatorship) and banks newly revealed on

January 26, 1933 that had a loan authorized in December.18 The final sample includes 2,674

unique banks. Table A.1 in the Online Appendix provides more detail about the number of

banks revealed on each subsequent list.

4.2 Additional covariates

To account for differing macroeconomic trends across each state, I include several additional

control variables in the estimation. These variables also capture the broader health of the

banking system. I use the dollar amount of total deposits and the total number of banks in

each state to account for the size, organization, and resources of the banking system. Next,

I use the dollar amount of suspended deposits and the total number of suspended banks

in each state to account for the health of the banking system. Suspended banks include

banks that closed their door to depositors for at least one business day and later resumed

operations, or banks that ceased operations, surrendered their charters, and repaid creditors

under a court-appointed receiver (Heitield et al. (2013)). The data are from the FDIC Bank

Deposit Data, 1920-1936 (Inter-university Consortium for Political and Social Research) and

are available yearly. I also use yearly state level per capita income data from the U.S. Bureau

of Economic Analysis, Survey of Current Business, May 2002. Per capita income captures

the potential cross-sectional variation in depositor income across states.

18Banks newly revealed on January 26, 1933 have no obvious comparison group since the control group isalso revealed on January 26, 1933.

22

5 Testing for stigma

5.1 Empirical specification

To compare the performance of revealed and non-revealed banks after the publication of the

list on August 22, 1932, I run the following bank-level OLS regression from June 30, 1930 to

June 30, 1933.

Yi,t = α+β1Revealedi× I{t = List}+β2Revealedi× I{t = List+ 1}+γXi,t−1 +ηt+ δi+ εi,t (1)

where Yi,t is the outcome of interest measured every six months t for bank i. I{t = List} is

a dummy equal to one following the publication of the list on August 22, 1932. December

31, 1932 is the first balance sheet date I observe for banks after the publication of the list.

I{t = List+ 1} is a dummy equal to one following the publication of the retroactive list on

January 26, 1933 revealing the control group.19 Revealed is a dummy equal to 1 if the bank

was publicly revealed on August 22, 1932. The coefficient of interest is β1, which measures the

change in Yi following the publication of the list for revealed banks compared to non-revealed

banks relative to their controls.20 The coefficient β2 measures the change in Yi following the

publication of the retroactive list for banks revealed on August 22, 1932 relative to banks

revealed on January 26, 1933 (non-revealed banks). The main outcome variable used is the

deposits-to-assets ratio. Other outcome variables are the loans-and-discounts-to-assets ratio

and the bonds-and-securities-to-assets ratio. For failed banks, I proxy for the deposits-to-19I do not use the traditional Revealed × Post specification because the control group was revealed on

January 26, 1933. Thus, the effect of stigma can only be measured between August 22, 1932 and January26, 1933.

20Note that I do not include a I{t = List} dummy, a I{t = List + 1} dummy, nor a Revealed dummysince they are not identified once I include half-year and bank fixed effects.

23

assets ratio with the total amount of suspended deposits divided by total deposits in the

state of the bank’s location so I can include failed banks in the estimation. The results are

also robust to recording zero for the deposit balance at failed banks.

A key issue that prevents the above specification from identifying the effect of the rev-

elation on Yi,t is that Yi,t may be correlated with unexplained macroeconomic conditions

and/or bank borrower characteristics in the error term εi,t. Therefore, I include controls,

Xi,t−1, to mitigate this bias and lag the covariates to ensure that Xi,t−1 does not confound

Yi,t. Xi,t−1 is a vector of controls measured six months prior to Yi,t and includes the log of

bank assets, the log of state per capita income, the log of state level deposits, the log of state

level deposits at suspended banks, the log of the number of banks at the state level, and

the log of the number of suspended banks at the state level. These covariates are intended

to capture observable proxies for macroeconomic conditions and bank characteristics that

might explain Yi,t. However, the specification may still be biased if some bank characteristics

are unobservable. Therefore, I rely on bank fixed effects, δi, to exclude biases that could

result from time-invariant bank characteristics and capture the extent to which each bank

affects Yi,t. Additionally, I include half-year fixed effects, ηt, to account for time trends in

Yi,t eliminating the concern that aggregate changes in Yi,t and the publication of the list

occurred together.

Finally, standard errors are clustered at the bank level according to Bertrand et al. (2004).

Furthermore, all continuous variables are winsorized at the 1% level to avoid outliers driving

the estimation results.

24

5.2 Restricted control group

One concern with the main specification is that some banks in the control group were revealed

before I observe any outcomes. Due to data limitations, I only observe bank-specific balance

sheet data at six-month intervals. After the first list was published on August 22, 1932, the

nearest balance sheet data are from December 31, 1932. If a bank in the control group had

another loan authorized between August 22 and November 30 and was subsequently revealed,

its December 31 balance sheet data would reflect this revelation. This characteristic of the

data makes it difficult to detect treatment effects because it biases the results towards zero

(Imbens and Angrist (1994)).21 Hence, I interpret my results as a lower bound identified

estimate of the effect of the revelation (Robins (1989)).

Figure 6 illustrates how 216 of the 713 non-revealed banks had additional loans autho-

rized. For clarity, I will refer to non-revealed banks who were not revealed as of August 22,

1932 and remained non-revealed until January 26, 1933 as the “restricted control group.”

Because I have to impose a constraint on the restricted control group of no loans authorized

between August 22 and November 30, 1932, I may be introducing a possible sample selection

bias into the results. Nevertheless, it is also worth emphasizing that banks in the restricted

control group are not limited to banks that never borrowed again from the RFC after Au-

gust 22, 1932. If a bank in the restricted control group had a loan authorized anytime in

December, its identity would not be revealed until January 26, 1933. Hence, its balance

sheet data as of December 31, 1932 would not reflect any early revelation. In the Online

Appendix, I perform a robustness check to potentially satisfy concerns about the possible21The common example used in the public health literature is a randomized drug trial. The treatment

group will receive the testable drug while the control group will receive a placebo. This assumes the con-trol group will remain compliant and not receive the testable drug elsewhere. Assuming compliance of allparticipants can be a strong assumption.

25

sample selection bias of the restricted control group.

Figure 6: Timeline Including Additional List Publications

This figure illustrates the four additional bank lists published between August 22, 1932 and January26, 1933. 216 of the 713 non-revealed banks had a loan authorized between August 22 and November30, 1932 and were revealed on a list prior to January 26, 1933. The remaining 497 banks were newlyrevealed on January 26, 1933.

5.3 Summary statistics

Table 1 displays descriptive statistics that compare revealed and non-revealed banks. Panel

A presents the summary statistics on bank-level balance sheet information on June 30, 1932,

the last balance sheet date I observe prior to the first list being published on August 22,

1932. Although the summary statistics show level differences between the treatment and

control groups, I find strong evidence that the parallel trend assumption holds with respect

to the deposits-to-assets, loans-and-discounts-to-assets, and bonds-and-securities-to-assets

ratios before the publication of the list in all the estimations and Figure 4. Non-revealed

banks were slightly larger in size and had more bonds and securities scaled by total assets.

This was offset by a smaller loan portfolio scaled by total assets. Panel B presents summary

statistics describing the loans authorized to revealed banks and non-revealed banks by the

26

RFC during 1932. Non-revealed banks received more loans from the RFC because these

banks were more likely to approach the RFC for a loan after the publication of the list

compared to revealed banks.

Table 1: Summary Statistics on Treatment and Control GroupsThis table presents summary statistics for the revealed and non-revealed banks prior to the pub-lication of the bank borrower list on August 22, 1932. There are 351 banks in the treatmentgroup (banks revealed on August 22, 1932) and 713 banks in the control group. Panel A presentsbank-level statistics as of June 30, 1932 while Panel B presents loan-level statistics for all of 1932.The loans-and-discounts-to-assets ratio is the book value of the bank’s loans outstanding scaled bytotal assets. The bonds-and-securities-to-assets ratio is the book value of the bank’s bond portfolioscaled by total assets.

Panel A (at June 30, 1932)Revealed Banks (351) Non-revealed Banks (713)

Obs. Mean Std. Dev Obs Mean Std. Dev Diff. in Meanstotal assets 351 2597.3 5539.0 713 4664.2 6348.3 ***

log(total assets) 351 6.74 1.43 713 7.95 0.92 ***DepositsAssets

351 0.70 0.12 713 0.69 0.12LoansAssets

351 0.69 0.17 713 0.65 0.15 ***BondAssets

351 0.31 0.16 713 0.34 0.14 ***

Panel B (all of 1932)Revealed Banks (351) Non-revealed Banks (713)

Obs. Mean Std. Dev Obs Mean Std. Dev Diff. in MeansLoan Amount 351 370.3 1233.4 713 530.04 1064.5 **(in thous.)

Number of Loans 351 1.7 1.21 713 2.09 1.35 ***

Table 2 displays summary statistics describing the number of bank failures amongst

the treatment and control groups. Because the control group was eventually revealed, the

period of interest to study bank failures was between August 22, 1932 (when the first list

was published) and January 26, 1933 (when the retroactive list was published). Therefore,

Table 2 illustrates that 9.2% of revealed banks failed during this period while only 6% of

non-revealed banks failed. This provides preliminary evidence that revealed banks were

27

more likely to fail after the publication of the list. Overall, most bank failures occurred after

January 26, 1933 because of the national banking holiday between March 6 and March 13,

1933 when President Roosevelt closed all banks across the US.

Table 2: Summary Statistics on Failed BanksThis table presents summary statistics for the number of bank failures amongst the 1,064 banks inthe treatment and control groups. Failed banks are banks that did not emerge from conservatorshipor were liquidated. Percentages of bank failures are provided in parentheses.

1064 banks 351 banks 713 banksTotal Revealed Banks Non-Revealed Banks

Failed 315 (29.6%) 105 (29.9%) 210 (29.5%)Failed before August 22, 1932 0 0 0Failed btw. August 22, 1932 75 (7.0%) 32 (9.2%) 43 (6.0%)and January 26, 1933Failed after January 26, 1933 240 73 (20.8%) 167 (23.4%)

5.4 The effect of the August 22, 1932 list on deposits

I start with the DID estimation of the effect of the August 22, 1932 publication of the

list of RFC borrowers on the deposits-to-assets ratio. Table 3 presents the main results.

Revealed banks experienced a highly statistically significant drop in their deposits-to-assets

ratios by 3.3 to 4.8 percentage points compared to non-revealed banks. This translates to a

5-7% relative drop in the deposits-to-assets ratios with respect to the pre-publication level,

and a $2.1 to 3.6 million loss in deposits in 2014 dollars relative to deposit levels before

the disclosure.22 Column (2) adds bank-level and state-level controls which slightly lowers

the coefficient yet remains highly economically and statistically significant. Columns (1)

and (2) also include one additional interaction term with Revealed. I find no differential22$134,000-$223,000 loss in deposits in 1932 dollars.

28

effect with respect to the deposits-to-assets ratio between treated and control firms prior

to the publication of the list. This provides evidence that the parallel trend assumption,

which is crucial for the validity of the DID estimate, holds in the data. The drop in the

deposits-to-assets ratio starts immediately after the list was published on August 22, 1932,

but disappears after the control group was revealed on January 26, 1933. The coefficient on

Revealedi × I{t = List + 1} indicates no differential effect in deposit withdrawals between

treated and control groups after all RFC borrowers were revealed. Indeed, after all borrowers

were revealed, the deposits-to-assets ratio for non-revealed banks caught up to the revealed

banks (see Figure 4), suggesting that depositors withdrew from all RFC borrowers equally

after all banks were revealed. Finally, Table A.2 in the Online Appendix provides evidence

that this drop in the deposits-to-assets ratio due to the August 22, 1932 list being published

was not driven by the amount borrowed from the RFC. Depositors do not withdraw more

from revealed banks that borrowed larger amounts.

29

Table 3: Effect of the Published List on LOLR Borrower DepositsThis table presents the DID estimates of the effect of the list published on August 22, 1932 on thedeposits-to-assets ratio. Revealed is a dummy that equals 1 if the bank was revealed on August 22,1932 (revealed banks). There are 351 revealed banks in the treatment group and 713 non-revealedbanks in the control group. Revealedi × I{t = List} is a dummy equal to one for revealed banksafter the August 22 list was published, and zero otherwise. Revealedi× I{t = List+1} is a dummyequal to one for banks revealed after the retroactive January 26, 1933 list was published, and zerootherwise. Revealedi × I{t = List − 1} is a dummy equal to one for revealed banks before theAugust 22 list was published, and zero otherwise. Xi,t−1 are controls measured six months prior tothe deposits-to-assets ratio and include the log of bank assets, the log of state per capita income,the log of state level deposits, the log of state level deposits at suspended banks, the log of thenumber of banks at the state level, and the log of the number of suspended banks at the statelevel. Standard errors are clustered at the bank level and presented in parentheses. All continuousvariables are winsorized at the 1% level. ***, **, and * indicate significance at the 1%, 5%, and10% respectively.

deposits-to-assets(1) (2)

Revealedi × I{t = List− 1} -0.0049 -0.0002(0.0072) (0.0070)

Revealedi × I{t = List} -0.0482*** -0.03295**(0.0163) (0.0157)

Revealedi × I{t = List+ 1} 0.0007 0.0166(0.0264) (0.0236)

Xi,t−1 No YesBank Fixed Effects δi Yes Yes

Half-Year Fixed Effects ηt Yes Yes

Obs. 6303 6303R2 0.6255 0.6626

The main concern of the identification strategy is that the treatment and control groups

may differ along a number of observable dimensions which might be correlated with the

outcome variables and bias the estimation. I address this concern through two robustness

checks (mentioned in the identification strategy section). First, I restrict the loan autho-

rization window for the control group to a short window of time by only including banks

with loans authorized between July 10 and July 20, 1932. Table 4 presents the results of

30

this test that uses a narrow time window of authorized loans. Revealed banks experienced

a 5.5 percentage point drop in the deposits-to-assets ratio compared to non-revealed banks.

This amounts to an 8% relative drop in deposits-to-assets with respect to the pre-publication

level. The drop in deposits at revealed banks continues to be economically and statistically

significant. In addition, Table 4 provides evidence that the parallel trend assumption still

holds for banks that had loans authorized in July.

Table 4: Effect of the Published List on LOLR Borrower Deposits for LoansAuthorized in JulyThis table presents the DID estimates of the effect of the list published on August 22, 1932 onthe deposits-to-assets ratio at banks with loans authorized in July. Revealed is a dummy thatequals 1 if the bank was revealed on August 22, 1932 (revealed banks). These revealed banks hada loan authorized from the RFC between July 21 and 31, 1932 while non-revealed banks had aloan authorized from the RFC between July 10 and July 20, 1932. Revealedi × I{t = List} is adummy equal to one for revealed banks after the August 22 list was published, and zero otherwise.Revealedi × I{t = List + 1} is a dummy equal to one for banks revealed after the retroactiveJanuary 26, 1933 list was published, and zero otherwise. Revealedi × I{t = List− 1} is a dummyequal to one for revealed banks before the August 22 list was published, and zero otherwise. Xi,t−1is a vector of controls identical to those used in Table 3. Standard errors are clustered at the banklevel and presented in parentheses. All continuous variables are winsorized at the 1% level. ***,**, and * indicate significance at the 1%, 5%, and 10% respectively.

deposits-to-assetsRevealedi × I{t = List− 1} -0.0139

(0.0124)Revealedi × I{t = List} -0.0546**

(0.0250)Revealedi × I{t = List+ 1} 0.0231

(0.0514)Xi,t−1 Yes

Bank Fixed Effects δi YesHalf-Year Fixed Effects ηt Yes

Obs. 2400R2 0.6213

Second, I restrict the treatment group to only include banks that had a loan authorized

31

Table 5: Effect of the List Publication on Deposits for Revealed Banks withEarlier Authorized LoansThis table presents the DID estimates of the effect of the list published on August 22, 1932 on thedeposits-to-assets ratio at banks with earlier authorized loans. Revealed is a dummy that equals1 if the bank was revealed on August 22, 1932 (revealed banks). These revealed banks had a loanauthorized between July 21 and 31, 1932 and between February 2 and July 20, 1932. Revealedi ×I{t = List} is a dummy equal to one for revealed banks after the August 22 list was published, andzero otherwise. Revealedi × I{t = List+ 1} is a dummy equal to one for banks revealed after theretroactive January 26, 1933 list was published, and zero otherwise. Revealedi× I{t = List− 1} isa dummy equal to one for revealed banks prior to the August 22 list, and zero otherwise. Xi,t−1 isa vector of control variables and is identical to those used in Table 3. Standard errors are clusteredat the bank level and presented in parentheses. All continuous variables are winsorized at the 1%level. ***, **, and * indicate significance at the 1%, 5%, and 10% respectively.

deposits-to-assetsRevealedi × I{t = List− 1} -0.0119

(0.0118)Revealedi × I{t = List} -0.0716**

(0.0354)Revealedi × I{t = List+ 1} -0.0362

(0.0547)

Xi,t−1 YesBank Fixed Effects δi Yes

Half-Year Fixed Effects ηt Yes

Obs. 4771R2 0.6588

during the same time period as the control group, i.e. between February 2 and July 20,

1932. Table 5 presents the results of this test. The deposits-to-assets ratio dropped by 7.2

percentage points at revealed banks compared to non-revealed banks. This translates to an

11.5% relative drop in the deposits-to-assets ratio with respect to the pre-publication level.

I then estimate the effect of the publication of the list on August 22, 1932 on the deposits-

to-assets ratio after constraining the control group to banks who remained non-revealed until

January 26, 1933, i.e. the restricted control group. Because some of the banks in the control

32

group receive treatment, i.e. revealed at a later date but before I observe its balance sheet,

the DID estimates are biased towards zero. Comparing revealed banks to banks in the

restricted control group in the following test will resolve this bias. The results are presented

in Table 6. The deposits-to-assets ratio significantly dropped by 5.5-7.2 percentage points

at revealed banks compared to non-revealed banks. This translates to an 8-10.5% relative

drop in the deposits-to-assets ratio with respect to the pre-publication level. I show that the

magnitude of the drop in deposits is larger after restricting the control group to banks who

remained non-revealed until January 26, 1933. If a bank in the control group had another

loan authorized after August 22, 1932 and was subsequently revealed before January 26,

1933, its December 31, 1932 balance sheet data would reflect this revelation, biasing the

estimation result towards zero.

33

Table 6: Effect of the Published List on LOLR Borrower Deposits using the“Restricted Control Group”This table presents the DID estimates of the effect of the list published on August 22, 1932 on thedeposits-to-assets ratio using the “restricted control group”. Revealed is a dummy that equals 1if the bank was revealed on August 22, 1932 (revealed banks). There are 351 revealed banks inthe Revealed group and 497 non-revealed banks in the restricted control group. Bank identitiesbelonging to the “restricted control group” remained secret until January 26, 1933. Revealedi ×I{t = List} is a dummy equal to one for revealed banks after the August 22 list was published, andzero otherwise. Revealedi × I{t = List+ 1} is a dummy equal to one for banks revealed after theretroactive January 26, 1933 list was published, and zero otherwise. Revealedi × I{t = List − 1}is a dummy equal to one for revealed banks before the publication of the August 22 list, and zerootherwise. Xi,t−1 is a vector of controls identical to those used in Table 3. Standard errors areclustered at the bank level and presented in parentheses. All continuous variables are winsorizedat the 1% level. ***, **, and * indicate significance at the 1%, 5%, and 10% respectively.

deposits-to-assets(1) (2)

Revealedi × I{t = List− 1} -0.0106 -0.0063(0.0075) (0.0070)

Revealedi × I{t = List} -0.0719*** -0.0547***(0.0166) (0.0157)

Revealedi × I{t = List+ 1} -0.035 -0.0070(0.0280) (0.0249)

Xi,t−1 No YesBank Fixed Effects δi Yes Yes

Half-Year Fixed Effects ηt Yes Yes

Obs. 4967 4967R2 0.6161 0.6582

One potential concern regarding comparisons of the performance of the treatment group

to the restricted control group is that I am imposing a restriction onto control group banks

thereby introducing a possible sample selection bias. A bank in the restricted control group

had no loans authorized between August 22 and November 30, 1932 whereas banks in the

treatment group could have had additional loans authorized during this period. I address this

concern in a robustness check by restricting the treatment group to only include banks that

had no additional loans authorized between August 22 and November 30, 1932, identical

34

to the restricted control group. Table A.3 in the Online Appendix displays the results

showing that restricting the treatment group does not alter the effect of the disclosure on

the deposits-to-assets ratio.

5.5 Empirical specification to analyze bank failure rates

The performance of a bank can also be measured by whether or not the bank failed. I examine

the effect of the revelation on bank failure rates between August 22, 1932 and January 26,

1933 using a duration model to isolate the effect of stigma before the control group was

revealed on January 26, 1933. The benefit of using a duration model is that it incorporates

the history of bank failures over time where I can exploit knowing the exact date a bank

suspended. I define a failed bank as a bank that did not emerge from conservatorship or was

liquidated, identical to the methodology used by Calomiris and Mason (2003). Furthermore,

I also test whether the failure of revealed banks was accelerated due to the revelation.

I estimate the duration model using maximum likelihood estimation (MLE). The like-

lihood can be written as a function of the hazard and survival functions. Let i index all

banks that were operating as of June 30, 1932, i.e. all banks in both the treatment and

control groups. If a bank did not suspend operations by June 30, 1933, it is treated as a

right-censored observation in the analysis. Let t index the duration time in weeks since the

August 22, 1932 list was published. Let θ(t;X) be the hazard function of the probability

that bank i failed at time t, conditional on surviving until t, with time-invariant covariates

X. For a discrete time duration model, the likelihood function is:

L =p∏p−1

[θ(tp; X)

tp−1∏s=0

(1− θ(s; X))]

(2)

35

To estimate the hazard function, θ(t; X), I use a parametric hazard function following a

Weibull distribution. Under the Weibull assumption, the hazard function takes the form:

θ(t; X) = ρλ(X)(tλ(X))ρ−1 (3)

λ(X) = exp {Xβ} (4)

The parameter ρ is the scale parameter of the hazard function. For example, if ρ̂ is estimated

to be between 0.5 and 1, then the hazard function is increasing at a decreasing rate. The

shape of the baseline hazard denotes the average bank failure rate over time.

The specification used in the analysis is estimated on a weekly basis from August 22,

1932 to January 26, 1933. However, the ideal hazard specification would also incorporate

bank-level time-varying covariates to account for time-varying differences in the baseline

hazard each week. Because I only observe bank balance sheet information every six months,

I estimate the baseline hazard using time-invariant covariates fixed as of June 30, 1932, the

first balance sheet date I observe prior to the publication of the list on August 22, 1932.

Therefore, the following specification estimates the probability a revealed bank failed by

January 26, 1933 compared to a non-revealed bank. I estimate ρ and β1 under the following

parameterization:

ln(λ) = α + β1Revealedi + γXi,June 30, 1932 + σu (5)

where u follows a Type-1 extreme value distribution and σ is a shape parameter that is

equivalent to 1ρ. Hence, β1, the coefficient of interest, measures the logged probability that

a revealed bank failed by January 26, 1933 compared to non-revealed banks. Revealed is a

dummy equal to 1 if the bank was publicly revealed on August 22, 1932. Xi,June 30, 1932 is a

36

vector of controls measured at June 30, 1932 and includes the log of bank assets and the log

of the loan amount bank i received from the RFC in 1932.

5.6 The effect of the August 22, 1932 list on bank failure rates

Table 7 shows that revealed banks were more likely to fail after the list was published

compared to non-revealed banks. The exponentiated regression coefficients for the hazard

model and the accelerated time-to-failure model are displayed. Revealed banks were 61.4%

more likely to fail by January 26, 1933 (when the control group was revealed) compared

to non-revealed banks. Moreover, conditional on failure, revealed banks failed 21.4% faster

than non-revealed banks after the publication of the list (1 - 0.7859). This translates to an

acceleration of nearly five weeks. For a graphical representation of this test, Figure A.3 in

the Online Appendix displays the cumulative hazard estimates using a Weibull distribution

for revealed banks and non-revealed banks.

Although it is difficult to estimate the marginal contribution of the revelation on deposit

withdrawals and likelihood of failure given the poor macroeconomic conditions during the

Great Depression, the hazard analysis provides some perspective on the economic importance

of the main results. Taken alone, it is difficult to assess the importance of a 5-7% drop in

the deposits-to-assets ratio at revealed banks compared to non-revealed banks. However, in

1932 banks, on average, lost 81% of their deposits before failing (Wicker (1996)). Taking this

fact into consideration, along with the 61.4% increase in probability that a revealed bank

would fail after the publication of the list, suggests that the marginal contribution of the

revelation was considerable. Given the many problems banks faced during this period, the

stigma introduced by the revelation was a problem that could have been fixed by Congress.

37

Table 7: Effect of the Published List on LOLR Borrower Failure RatesThis table presents the parametric hazard function following a Weibull distribution estimates ofthe effect of the list publication on August 22, 1932 on bank failure rates. Revealed is a dummythat equals 1 if the bank was revealed on August 22, 1932 (revealed banks) conditional on survivinguntil August 21, 1932. Column (1) provides the exponentiated hazard coefficients presenting theprobability of failure by June 30, 1933 while Column (2) provides the exponentiated acceleratedtime-to-failure coefficients. Xi,June 30, 1932 are fixed bank-level controls measured prior to August22, 1932 including the log of bank assets and the log of the loan amount accepted from the RFC.Standard errors are presented in parentheses. All continuous variables are winsorized at the 1%level. ***, **, and * indicate significance at the 1%, 5%, and 10% respectively.

(1) (2)Hazard Accelerated Time-to-Failure

Treatmenti 1.6137* 0.7859*(0.4361) (0.1089)

Xi,June 30, 1932 Yes Yes

σ 0.5036 0.5036Obs. 1064 1064

Wald chi2(3) 4.58 4.58

38

5.7 Banks that did not borrow from the Reconstruction Finance

Corporation

After all RFC borrowers were revealed, it would interesting to observe if depositors differ-

entially withdrew from banks that borrowed from the RFC compared to banks that never

borrowed. However, this comparison introduces a sample selection bias into the analysis.

For example, suppose that banks with loans from the RFC experienced larger deposit with-

drawals than banks without loans. Is this result driven by depositors withdrawing more from

the banks with loans because they had borrowed from their LOLR, or because those banks

were somehow unobservably worse?

Nonetheless, I collected balance sheet data for 130 randomly selected banks that never

approached the RFC for a loan.23 First, Figure 7 displays the trends in bank size (total

assets) for banks that never borrowed from the RFC, revealed banks and non-revealed banks.

Initially, banks that would receive an RFC loan were similar in size to banks that would

receive no loans. However, after 1932, the size of the borrowing banks dropped faster than

the non-borrowing banks, despite receiving loans from the RFC.23These balance sheet data are also from Rand McNally Bankers’ Directory.

39

Figure 7: Trends in Bank Size

This figure illustrates the average percent change in total bank assets for banks that borrowed fromthe RFC (revealed and non-revealed banks) and banks that did not borrow from the RFC (neverborrowed banks), normalized by June 30, 1930 total asset levels. Total assets for each banks is thesum of its loans and discounts, bonds and securities, cash due from other banks, and real estateowned. The darker shaded area between June 30, 1932 and December 31, 1932 denotes the periodthe Revealed group was revealed (August 22, 1932). The second lighter shaded area denotes theperiod the control group was revealed (January 26, 1933).

Second, Figure 8 illustrates the trend in the deposits-to-assets ratio for banks that never

borrowed from the RFC, revealed banks and non-revealed banks. Initially, banks that re-

ceived no loans had less funding from deposits than banks that received loans. Therefore,

these non-borrowing banks had larger alternate sources of funding. As the Depression wore

on, banks that received no loans were able to maintain higher deposit funding compared

to banks that received loans. Third, Figure 9 displays the trends in total liabilities minus

deposits over total assets (the remaining sources of funding for banks). We can see that

banks that did not receive RFC loans had a larger share of alternate funding sources. These

alternate funding sources included paid-up capital, bank surplus and profits, and loans from

40

Figure 8: Trends in Deposit Levels

This figure illustrates the average ratio of total deposits over assets for banks that borrowed fromthe RFC (revealed and non-revealed banks) and banks that did not borrow from the RFC (neverborrowed banks). The darker shaded area between June 30, 1932 and December 31, 1932 denotesthe period the treatment group was revealed (August 22, 1932). The second lighter shaded areadenotes the period the control group was revealed (January 26, 1933).

41

other banks. Paid-up capital is a function of the bank president’s own capital. The share

of alternate funding sources for borrowing banks increased after 1932 because they received

loans from the RFC.

Figure 9: Trends in Remaining Liabilities

This figure illustrates the average ratio of total liabilities minus deposits over assets for banks thatborrowed from the RFC (revealed and non-revealed banks) and banks that did not borrow from theRFC (never borrowed banks). The darker shaded area between June 30, 1932 and December 31,1932 denotes the period the treatment group was revealed (August 22, 1932). The second lightershaded area denotes the period the control group was revealed (January 26, 1933).

Finally, Table 8 displays summary statistics comparing banks that never borrowed from

the RFC to non-revealed banks as of June 30, 1932, the last balance sheet date I observe

prior to the list being published on August 22, 1932. It shows that non-revealed banks and

never-borrowed banks are statistically different with regards to size, loans, and bonds. In

summary, banks that never borrowed from the RFC were fundamentally different from banks

that borrowed. This suggests that any analysis comparing the performance of banks that did

not borrow from their LOLR to banks that did borrow would suffer from a sample selection

42

bias.

Table 8: Summary Statistics for Never Borrowed BanksThis table presents summary statistics for the sample of non-revealed and never borrowed banksprior to the publication of the bank borrower list on August 22, 1932. There are 713 non-revealedbanks that borrowed from the RFC and 130 randomly selected banks that never approached fromthe RFC. The panel presents bank-level statistics as of June 30, 1932. The loans-and-discounts-to-assets ratio is the book value of the bank’s loans outstanding scaled by total assets. The bonds-and-securities-to-assets ratio is the book value of the bank’s bond portfolio scaled by total assets.The last ratio represents how each type of bank, on average, funds its assets adjusting for depositlevels.

Non-Revealed Banks (713) Never Borrowed (130)Obs. Mean Std. Dev Obs Mean Std. Dev Diff. in Means

total assets 713 4664.2 6348.3 130 15812 27079 ***log(total assets) 713 7.95 0.92 130 8.44 1.63 ***

DepositsAssets

713 0.69 0.12 130 0.72 0.23 *LoansAssets

713 0.65 0.15 130 0.58 0.17 ***BondAssets

713 0.34 0.14 130 0.42 0.17 ***Liabilities−Deposits

Assets713 0.31 0.12 130 0.28 0.22 ***

5.8 Further robustness

In the Online Appendix, I provide supplementary tables that present alternative specifica-

tions which test for stigma. First, Table A.4 in the Online Appendix shows that adding

bank-level loan variables controlling for the length of time since the bank had a loan autho-

rized, the number of times the bank approached the RFC for a loan, and the amount the bank

borrowed do not alter the effect of the disclosure on the deposits-to-assets ratio. Second,

Table A.5 in the Online Appendix illustrates that subtracting the bank’s total borrowings

from the RFC from its total assets to control for the change in the balance sheet after the

bank borrowed has little effect on the results. The effect of the revelation for revealed banks

after the list was published continues to be economically large and statistically significant.

43

Third, the results qualitatively hold when the outcome variable is changed to the log level of

deposits. Fourth, Table A.6 in the Online Appendix shows that revealed banks were quali-

tatively more likely to fail by June 30, 1933 compared to non-revealed banks but the effect

is mitigated due to the control group disclosure on January 26, 1933. Fifth, a condition of

the analysis is that the sample only consists of banks that chose to borrow from the LOLR.

It could be that the deposits-to-assets ratio for all RFC borrowers was decreasing at a rate

different to banks that never borrowed during this period. However, this is also a strength

of the analysis because it highlights the potential significance of the sample selection bias of

comparing banks that chose to borrow from the LOLR to banks that chose not to borrow.

Finally, because the deposits-to-assets ratio for non-revealed banks meets the levels of the

revealed banks after all RFC borrowers were revealed, it is difficult to raise a theory that

creates a drop in the deposits-to-assets ratio for the revealed banks compared to the non-

revealed banks, and then sees the ratios merge after the non-revealed banks were revealed,

that is uncorrelated with the August 22, 1932 list being published.

6 Managing the information environment: inter-city

comparisons

6.1 Empirical specification

Having found evidence of the stigma problem, I next ask how policy makers can use this in-

formation to improve the design of emergency lending facilities and manage the information

environment during a crisis to alleviate the problem. One possibility is to resolve a coordina-

44

tion problem between banks that results from banks reluctance to borrow from the discount

window (Armantier et al. (2014)). Because of stigma, a bank’s decision to borrow from the

LOLR involves predicting the actions of other banks. If a bank believes that other banks are

reluctant to borrow, its own belief in stigma is reinforced, and, as a result, the bank decides

not to borrow from the LOLR. To identify the optimal policy design to address this coordi-

nation problem, I must determine the conditions under which market participants did not

target withdrawals at revealed banks in comparison to non-revealed banks. Because banking

markets during the Great Depression were mostly defined by city boundaries (Heitield et al.

(2013)), I can compare depositor reactions between two types of defined banking markets,

i.e. cities, where the important variation between each type of city was how the informa-

tion about RFC borrowers was released. The first type of banking market includes cities

where nearly all banks were revealed to have borrowed from the LOLR, while the second

type includes cities where only few banks were revealed to have borrowed. If information

was revealed that nearly all banks in a city borrowed from the RFC, did depositors react

differently than depositors in cities where few banks in a city were revealed?

To test depositor reactions across cities, I create two dummy variables: I{Revealed ≤

1−x} and I{Revealed ≥ x} where x ε[ 50%, 100%] is the percentage of revealed banks within

a city as of June 30, 1932, prior to the publication of the list. I{Revealed ≤ 1 − x} equals

1 when the percentage of revealed banks within that city was beneath 1 − x. Similarly,

I{Revealed ≥ x} equals 1 when the percentage of revealed banks within that city was above

x. Using the nearly identical OLS regression from above, I now add two additional terms,

two triple interactions, that incorporate the percentage of revealed banks within a city. For

bank i in city c at half-year t:

45

Deposits

Assets i,t= α + β1 Revealedi × I{t = List} × I{Revealed ≤ 1− x}c

+β2 Revealedi × I{t = List} × I{Revealed ≥ x}c

+β3 Revealedi × I{t = List} + αRemaining Interaction Terms

+γXi,t−1 + ηt + δi + εi,t

(6)

β1 measures the change in the deposits-to-assets ratio following the publication of the

list for revealed banks in cities where the percentage of revealed banks was beneath 1 − x.

Similarly, β2 measures the change in the deposits-to-assets ratio following the publication of

the list in cities where the percentage of revealed banks exceeded x. Finally, the coefficient

β3 measures the change in the deposits-to-assets ratio following the publication of the list for

revealed banks compared to non-revealed banks relative to their controls.24 I also include

lagged control variables Xi,t−1, bank fixed effects δt, and time fixed effects ηt for the same

reasons explained in Section 5.1 although the results are robust to using fixed effects at the

city level.25 Standard errors are clustered at the bank level according to Bertrand et al.

(2004) and all continuous variables are winsorized at the 1% level to avoid outliers driving

the estimation results.

24Note that I do not include the remaining terms of Revealed × I{Revealed ≤ 1 − x}, Revealed ×I{Revealed ≥ x}, Dec. 31, 1932 and Revealed since they are not identified once I include half-year and bankfixed effects.

25The estimation results are also robust to including triple interaction terms for when the control groupis revealed.

46

6.2 Summary statistics

Table 9 displays summary statistics describing the distribution of revealed and non-revealed

banks across cities as a function of the total number of banks in each city as of June 30, 1932.

The percentage of revealed banks within a city equals the number of revealed banks divided

by the total number of banks in that city. The total number of banks within a city includes

the number of revealed banks, the number of non-revealed banks, and the number of banks

that never borrowed from the RFC. The 1,064 banks with loans authorized before August

22, 1932 were located in 817 cities where the average percentage of revealed banks within

a city was 22%. Although there were cities where all banks were revealed, those locations

were limited to cities with a single bank. Requiring a city to have at least two banks reduces

the maximum percentage of revealed banks within a city to 75%. The average percentage of

revealed banks within a city with at least two banks was 12%.

Table 9: Summary Statistics across CitiesThis table presents summary statistics describing the distribution of revealed and non-revealedbanks across cities as a function of the total number of banks in each city as of June 30, 1932 (priorto the publication of the list on August 22, 1932). The percentage of revealed banks within a cityequals the number of revealed banks in that city divided by the total number of banks in that city.The total number of banks within a city includes the number of revealed banks, the number ofnon-revealed banks, and the number of banks that never borrowed from the RFC. The 1,064 banksin the data (351 + 713) were in 817 cities.

As of June 30, 1932 % of revealed banks within a cityNo. of cities mean max

All cities 817 22% 100%Cities with 1 bank 252 53% 100%

Cities with at least 2 banks 565 12% 75%Cities with at least 3 banks 318 11% 75%Cities with at least 4 banks 204 10% 75%Cities with at least 5 banks 140 11% 60%

47

6.3 The effect of the August 22, 1932 list on deposits across cities

First, I consider depositor reactions in cities where all banks were revealed. According to

Table 9, cities where all banks were revealed were in cities with a single bank. Table 10

displays the DID estimates of comparing revealed banks to non-revealed banks in cities

with a single bank. The coefficient on Revealedi × I{t = List} is insignificant implying

no difference in the deposits-to-assets ratio for revealed banks compared to non-revealed

banks. Examining depositor reactions in cities with a single bank is analogous to examining

depositor reactions in cities where all banks were revealed. Unfortunately, it is difficult to

distinguish if depositors were processing information differently or if their actions resulted

from their inability to move their funds.

48

Table 10: Effect of the Published List on LOLR Borrower Deposits in Cities with1 BankThis table presents the DID estimates of the effect of the list published on August 22, 1932 onthe deposits-to-assets ratio in cities with a single bank. Revealed is a dummy that equals 1 if thebank was revealed on August 22, 1932 (revealed banks). There are 252 cities with a single bankwhere 143 revealed banks are in the treatment group and 109 non-revealed banks in the controlgroup.Revealedi × I{t = List} is a dummy equal to one for revealed banks after the August 22list was published, and zero otherwise. Revealedi × I{t = List + 1} is a dummy equal to onefor banks revealed after the retroactive January 26, 1933 list was published, and zero otherwise.Revealedi × I{t = List− 1} is a dummy equal to one for revealed banks before the August 22 listwas published, and zero otherwise. Xi,t−1 is a vector of controls identical to those used in Table3. Standard errors are clustered at the bank level and presented in parentheses. All continuousvariables are winsorized at the 1% level. ***, **, and * indicate significance at the 1%, 5%, and10% respectively.

deposits-to-assetsRevealedi × I{t = List} -0.0196

(0.0296)

Xi,t−1 YesBank Fixed Effects δi Yes

Half-Year Fixed Effects ηt Yes

Obs. 1361R2 0.6815

I address this concern in two ways. First, I only examine cities with a minimum of two

banks and I use a threshold of revelation. The benefit of using a threshold is that if some

banks in a city were revealed, then alternative banking options in that city were available

such as the US postal savings system (O’Hara and Easley (1979)). I start with x = 66%

although the results are robust to using alternate thresholds. Table 11 presents the DID

estimates of comparing the performance of revealed banks in cities where more than 66%

of banks were revealed and revealed banks in cities where fewer than 33% of banks were

revealed to non-revealed banks. Revealed banks, in cities where at least 66% of banks were

49

revealed, did not experience a drop in the deposits-to-assets ratio compared to non-revealed

banks (-0.0827 + 0.1148). In these cities, information was revealed simultaneously. However,

revealed banks, in cities where fewer than 33% of banks were revealed, experienced a drop

in the deposits-to-assets ratio of 6.89 percentage points compared to non-revealed banks (-

0.0827 + 0.0138). This translates to a 10% relative drop in the deposits-to-assets ratio with

respect to the pre-publication level. In these cities, information was revealed asymmetrically.

Moreover, from Table 11, we can also see that revealed banks, in cities where at least 66% of

banks were revealed, had higher deposit-to-assets ratios than revealed banks in other cities.

By construction, this signifies that deposit outflows in cities where nearly all banks were

revealed were less than deposit outflows in cities where few banks were revealed.26 This

implies that banks were worse off with respect to the deposits-to-assets ratio in cities where

few banks were publicly revealed. In these cities, the presence of stigma was larger, consistent

with the rationale behind banks’ reluctance to borrow from the discount window. Hence,

revealing nearly all bank identities simultaneously mitigated the effect of stigma.26If deposit outflows at revealed banks in cities where few banks were revealed were greater than deposit

outflows at revealed banks where nearly all banks were revealed, it must follow that the sum of depositoutflows across all banks in cities where few banks were revealed must be greater than the sum of depositoutflows across all banks in cities where nearly all banks were revealed.

50

Table 11: Effect of the Published List on LOLR Borrower Deposits in Cities with≥2 BanksThis table presents the DID estimates of the effect of the list published on August 22, 1932 on thedeposits-to-assets ratio across cities with at least two banks. Revealedi×I{t = List} × Revealed ≥66% is a dummy equal to one for revealed banks after the August 22 list was published in cities whereat least 66% of banks were revealed, and zero otherwise. Revealedi×I{t = List} ×Revealed ≤ 33%is a dummy equal to one for revealed banks in cities where fewer than 33% of banks were revealed,and zero otherwise. Revealed is a dummy that equals 1 if the bank was revealed on August 22,1932 (revealed banks). There are 351 revealed banks in the treatment group and 713 non-revealedbanks in the control group. Revealedi × I{t = List} is a dummy equal to one for revealed banksafter the August 22 list was published, and zero otherwise. Remaining interactions include alladditional interactions from the triple interactions. Xi,t−1 is a vector of controls identical to thoseused in Table 3. Standard errors are clustered at the bank level and presented in parentheses. Allcontinuous variables are winsorized at the 1% level. ***, **, and * indicate significance at the 1%,5%, and 10% respectively.

deposits-to-assetsRevealedi × I{t = List} × Revealed ≥ 66% .1148**

(0.0603)Revealedi × I{t = List} × Revealed ≤ 33% 0.0138

(0.0440)Revealedi × I{t = List} -0.0827***

(0.0301)

Remaining Interactions YesXi,t−1 Yes

Bank Fixed Effects δi YesHalf-Year Fixed Effects ηt Yes

Obs. 4942R2 0.6594

The potential mechanism explaining these results is the following. Depositors, in cities

where nearly all banks were revealed, were more likely to assume that the remaining non-

revealed banks in their city also borrowed from the RFC. Revealing nearly all banks was

suggestive of an aggregate adverse shock rather than an idiosyncratic shock to a specific bank.

On the other hand, depositors, in cities where few banks were revealed, were more likely to

51

interpret the news as reflective of each bank’s financial weakness rather than reflecting an

aggregate adverse shock. I make two assumptions for this mechanism to hold. First, assume

banks lent locally implying the default risk to their loan portfolios was limited to businesses

within the city of the bank’s location (Hautcoeur et al. (2013)). Second, assume that some

cities were hit harder than others during the Great Depression (Wicker (1996)). Table A.6

in the Online Appendix presents the DID estimates of the effect on the deposits-to-assets

ratio for alternative threshold levels and the results are robust for values of x ≥ 60% and

qualitatively robust for x = 50%. The effect is also qualitatively robust when increasing the

minimum number of banks within a city.

Finally, these results imply that emergency lending facilities that reveal information about

bank borrowers simultaneously can mitigate stigma. This can be achieved by providing a

coordination mechanism at the lending facility that allows banks to jointly request loans

simultaneously from the LOLR. In the case that identities are revealed, it is likely that more

than one bank identity would be simultaneously revealed, thus preventing an individual

bank from being targeted when approaching the LOLR for a loan. The design of the facility

implies that revealing information simultaneously (which can take the form of all banks

being revealed or no banks being revealed) dominates revealing information asymmetrically

(where few banks are revealed). Although the results do not explicitly show that keeping all

bank identities private dominated revealing all bank identities simultaneously, we can make

this inference based on the result that stigma exists and assumptions made in the existing

literature. Ideally, central banks should prevent the release of bank borrower identities during

a financial crisis but this may not be possible given the idiosyncratic leakage of discount

window loans during the recent crisis.27 Finally, these conclusions shed light on why banks27Armantier et al. (2014) identify some media articles that leaked discount window borrowing of large

banks and other idiosyncratic channels.

52

may have expected less stigma from TAF loans rather than loans from the discount window.

Because I show that the presence of stigma was worse for the aggregate system, from a policy

maker perspective, forcibly requiring banks to accept LOLR funding also alleviated stigma.

The public believed that borrowing from the LOLR was acceptable because it seemed that

many banks had chosen to borrow, solving the coordination problem between banks. The

stigma associated with receiving emergency loans was mitigated.

7 Ex-post bank behavior

Thus far, I have only considered the reactions of depositors to the August 22, 1932 publication

of the list. However, an aim of the central bank is to ease liquidity needs for solvent yet

illiquid banks during a crisis where, in turn, these banks may use the funding to maintain

loans to their consumers (Freixas et al. (1999)). Did banks behave differently after their

identities were revealed? Answering this question with these data, however, is difficult. The

lack of detail regarding a bank’s loan and bond portfolio makes it difficult to say precisely

how revealed banks adjusted their balance sheet in response to the withdrawal of deposits

after the list was published due to data limitations. For example, I do not observe if revealed

banks stopped issuing new loans after the publication of the list or if they wrote down the

value of existing loans.28 Nonetheless, in the Online Appendix, I examine the impact to28I can somewhat address this concern by exploring call reports from Federal Reserve member banks that

also borrowed from the RFC. Anecdotally, revealed banks reduced their real estate mortgages and calledin loans from other banks in response to depositors withdrawing their funds. Moreover, revealed banksincreased their bond portfolios by purchasing more US government securities and state or municipal bonds.A few also bought railroad bonds. However, it is difficult to be precise about how revealed banks modifiedtheir balance sheet after the publication of the list due to data limitations. I intend to address this limitationin future work.

53

loans and bonds at revealed banks compared to non-revealed banks but do not observe the

detailed compositions of loans/discounts or bonds/securities.

Table A.7 in the Online Appendix presents the results of the DID estimation of the

effect of the August 22, 1932 publication of the list on loans/discounts and bonds/securities

scaled by total assets. Column (1) shows a qualitatitive drop in the loans-and-discounts-to-

assets ratio (that the coefficient on Revealedi × I{t = List}is not statistically significant).

However, Column (2) presents the result for loans when the performance of revealed banks

is compared to the restricted control group. The loans-and-discounts-to-assets ratio dropped

by 4.2 percentage points at revealed banks compared to non-revealed banks. This translates

to a 6.2% relative drop in the loans-to-assets ratio with respect to the pre-publication level.

Columns (3) and (4) show the results of the DID estimation on bonds and securities.

The bonds-and-securities-to-assets ratio increased by 3-4 percentage points at revealed banks

compared to non-revealed banks after the publication of the list. The increase in the bonds-

and-securities-to-assets ratio translates to a 10-12.5% relative increase at revealed banks

compared to non-revealed banks with respect to the pre-publication level, a significantly

large increase.

8 Conclusion

In this paper, I provide evidence of the stigma problem. Using data from an unexpected

disclosure of a partial list of banks that secretly borrowed from the LOLR during the Great

Depression, I find evidence of the stigma problem in that depositors withdrew more deposits-

to-assets from banks included on the list in comparison to banks left off the list. This result

54

sheds light on why banks wish to avoid approaching the discount window. However, if the

identities of nearly all banks in a given banking market were simultaneously revealed to have

borrowed from the LOLR, the presence of stigma was mitigated. Symmetric information

revelation motivated depositors to interpret borrowing from the LOLR as reflecting an ag-

gregate negative shock rather than reflecting those banks’ financial weakness. Thus, cities

where information was revealed simultaneously were associated with lower deposit outflows

than cities where information was revealed asymmetrically. These conclusions provide ev-

idence that it is in the policy maker’s interest to prevent runs on banks associated with

stigma to support the financial system during a crisis. Revealing the identities of all banks

simultaneously dominates revealing the identities of few banks asymmetrically.

This paper helps to explain how the LOLR’s management of the information environment

can have strong adverse impacts on the health of the financial system. Designing emergency

lending facilities that allow banks to coordinate requests for LOLR loans will mitigate stigma,

solve the coordination problem between banks, and implicitly encourage banks to borrow

from the LOLR. Solving the stigma problem is crucial to effectively fight future financial

crises.

Primary Sources

Chronicle, Commercial Financial, 1932a, President hoover acts as decision is announced tomake public, loans by rfc - president seeks means to avert serious effect, Commercial &Financial Chronicle Aug 18, 1932.

Chronicle, Commercial Financial, 1932b, Publicity to be given to loans made by rfc, Com-mercial & Financial Chronicle August 20, 1932.

Chronicle, Commercial Financial, 1933, Atlee pomerene assails publication of names of bankswhich received aid from rfc, Commercial & Financial Chronicle March 14, 1933.

55

Company, Rand McNally &, Ed. January 1930 to July 1933, Rand McNally Bankers Direc-tory (Rand McNally & Company).

Corporation, Reconstruction Finance, 1932, US Reconstruction finance corporation quarterlyreport.

FDIC, 1920-1936, Federal deposit insurance corporation (fdic) data on banks in the unitedstates, Inter-university Consortium for Political and Social Research 7.

of Economic Analysis, US Bureau, 2002, Survey of current business.

Service, Moody’s Investor, Ed. 1932 and 1933, Moody’s: Banks, Insurance, Real Estate,Investment Trusts (Moody’s Investor Service).

Times, New York, 1932, Report of the rfc listing some of loans and showing financial condi-tion, New York Times August 22, 1932.

Secondary Sources

Armantier, Olivier, Eric Ghysels, Asani Sarkar, and Jr. Jeffrey Shrader, 2014, Discountwindow stigma during the 2007-2008 financial crisis, Federal Reserve Bank of New YorkStaff Report no. 483, Available at http://ssrn.com/abstract=1754558.

Benmelech, Efraim, 2012, An empirical analysis of the fed’s term auction facility, CATOPapers on Public Policy 2, 57–91, CATO Papers on Public Policy.

Bernanke, Ben, 2009, The federal reserve’s balance sheet: An update, in Speech at the FederalReserve Board Conference on Key Developments in Monetary Policy.

Berry, John, 2012, Bloomberg vs. the fed, The Magazine of International Economic PolicyWinter, 65–77.

Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, 2004, How much should wetrust differences-in-differences estimates?, The Quarterly Journal of Economics 119, 249–275.

Butkiewicz, James L., 1995, The impact of a lender of last resort during the great depression:The case of the reconstruction finance corporation, Explorations in Economic History 32,197–216.

Calomiris, Charles W., and Joseph R. Mason, 2003, Fundamentals, panics, and bank distressduring the depression, The American Economic Review 93, 1615–1647.

56

Calomiris, Charles W., Joseph R. Mason, Marc Weidenmier, and Katherine Bobroff, 2013,The effects of reconstruction finance corporation assistance on michigan banks survival inthe 1930s, Explorations in Economic History 50, 526–547.

Ebersole, J. Franklin, 1933, One year of the reconstruction finance corporation, The Quar-terly Journal of Economics 47, 464–492.

Ennis, Huberto M., and John A. Weinberg, 2013, Over-the-counter loans, adverse selection,and stigma in the interbank market, Review of Economic Dynamics 16, 601–616.

FOMC, 2007, Fomc meeting minutes from september 18, 2007, in Transcript.

Freixas, Xavier, Curzio Giannini, Glenn Hoggarth, and Farouk Soussa, 1999, Lender of thelast resort: A review of the literature, Financial Stability Review - Bank of England 7,151–167.

Friedman, Milton, and Anna Jacobson Schwartz, 1963, A Monetary History of the UnitedStates (Princeton University Press).

Furfine, Craig, 2003, Standing facilities and interbank borrowing evidence from the federalreserves new discount window, International Finance 6(3), 329–348.

Geithner, Timothy F., 2014, Stress Test: Reflections on Financial Crises, first edition(Crown).

Gorton, Gary, and Andrew Metrick, 2013, The federal reserve and panic prevention: Theroles of financial regulation and lender of last resort, The Journal of Economic Perspectives27, 45–64.

Gorton, Gary, and Guillermo Ordonez, 2014, Collateral crises, American Economic Review104(2), 343–378, http://www.nber.org/papers/w17771.

Gorton, Gary, and Ellis Talman, 2014, How do banking panics end?, Working Paper.

Hautcoeur, Pierre-Cyrille, Angelo Riva, and Eugene N. White, 2013, Floating a "lifeboat":the banque de france and the crisis of 1889, Journal of Monetary Economics November.

Heitield, Erik, Gary Richardson, and Shirley Wang, 2013, Modes of contagion in the bankingpanic of 1930, Working Paper.

Imbens, Guido W., and Joshua D. Angrist, 1994, Identification and estimation of localaverage treatment effects, Econometrica 62, 467–475.

Kleymenova, Anya, 2013, Consequences of mandated bank liquidity disclosures, WorkingPaper.

57

Madigan, Brian F., 2009, Bloomberg versus board of governors of the federal reserve system,declaration of brian f. madigan, Case number 08 Civ. 9595 (LAP) .

Mason, Joseph R., 2001, Do lender of last resort policies matter? the effects of reconstructionfinance corporation assistance to banks, Journal of Financial Services Research September(20:1), 77–95.

Mason, Joseph R., and Daniel Schiffman, 2003, Too-big-to-fail, Government Bailouts, andManagerial Incentives: The Case of Reconstruction Finance Corporation Assistance tothe Railroad Industry during the Great Depression (Benton E. Gup, ed. Westport, CT:Greenwood Press).

Mitchener, Kris James, and Gary Richardson, 2013, Shadowy banks and financial conta-gion during the great depression: A retrospective on friedman and schwartz, AmericanEconomic Review 103, 73–78.

O’Hara, Maureen, and David Easley, 1979, The postal savings system in the depression.,Journal of Economic History Rep. 3rd ed. Vol. XXXIX., 741–753.

Olson, James S., 1977, Herbert Hoover and the Reconstruction Finance Corporation 1931-1933 (The Iowa State University Press).

Robins, JM, 1989, The analysis of randomized and non-randomized aids treatment trialsusing a new approach to causal inference in longitudinal studies, Health service researchmethodology: a focus on the AIDS .

Todd, Walker F., 1992, History of and rationales for the reconstruction finance corporation,Federal Reserve Bank of Cleveland Economic Review QIV, 22–35.

Treasury, 2008, Uk parliamentary report ordered by the house of commons, Available athttp://www.publications.parliament.uk/pa/cm200708/cmselect/cmtreasy/56/5602.htm.

Wicker, Elmus, 1996, The Banking Panics of the Great Depression (Cambridge UniversityPress).

Winters, Bill, 2012, Review of the bank of englands framework for pro-viding liquidity to the banking system, Working Paper. Available athttp://www.bankofengland.co.uk/publications/Documents/news/2012/cr2winters.pdf.

58