Financial Failure and Depositor Quality: Evidence from Building and Loan Associations in California

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1354 August 2022 Financial Failure and Depositor Quality: Evidence from Building and Loan Associations in California Todd Messer Please cite this paper as: Messer, Todd (2022). “Financial Failure and Depositor Quality: Evidence from Building and Loan Associations in California,” International Finance Discussion Papers 1354. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2022.1354. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Financial Failure and Depositor Quality: Evidence from Building and Loan Associations in California Todd Messer∗ Federal Reserve Board July 2022 Abstract Flightiness, or depositor sensitivity to liquidity needs, can be an important determinant of financialdistress.Ileverageinstitutionaldifferences—thatattractdepositorswithvaryingflightiness—across buildingandloanassociationsinCaliforniaduringtheGreatDepression.Anewtypeofplan,theDayton plan, involved less restrictive savings plans and lower withdrawal penalties. Dayton plans in California were more likely to close during the Great Depression. Archival evidence on lending rates and returns supports the flightiness mechanism. Keywords: Bank Failures, Banks, credit unions, and other financial institutions, Building and Loan, Great Depression JEL Codes: N22, G23, G21 ∗IwouldliketothankChaewonBaek,ChristopherCampos,MarinaDias,EricHilt(editor),PeterMcCrory,MarthaOlney, MathieuPedemonte,GaryRichardson,ChristinaRomer,DavidRomer,JonathanRose,KennethSnowden,andthreeanonymous refereesforhelpfulcommentsandsuggestions.IwouldalsoliketothankseminarparticipantsatUCBerkeley.CharlesMcMurry and Jameson Weiss provided outstanding research assistance. I also acknowledge the All-UC Group in Economic History for providing funds for data collection. Any remaining errors are my own. The views in this paper are the responsibility of the authoranddonotnecessarilyrepresentthoseoftheFederalReserveBoardortheFederalReserveSystem. 1

1 Introduction The failure of individual financial institutions is often associated with poor macroeconomic conditions and financial instability. Conventional explanations for failure include liquidity shocks due to the maturity mismatch of assets and liabilities (e.g. the model of Diamond and Dybvig 1983) or insolvency due to impaired assets. Yet both of these explanations are limited in incorporating the institutional details of how financial institutions structure their liabilities. For instance, commercial banks can vary the deposit rate or restrict access to liquidity via withdrawal limitations. Relying on low-cost funding may attract depositors that are ex-ante more susceptible to liquidity shocks—i.e. more “flighty.” What appears across banks as surprise liquidity shocks is actually a function of their predetermined structures. Financial distress can be endogenous to the characteristics of depositors. Inthispaper,Istudytheroleofdepositorflightinessusingbuildingandloanassociations(B&Ls)inCalifornia during the Great Depression. B&Ls were lending institutions that specialized in loans against real estate, accounting for about one-third of the institutional residential mortgage market at their peak in the 1920s. During the Great Depression, there were a large number of closures via liquidation (both voluntary and involuntary)amongB&LsacrosstheUnitedStates.Snowden(2003)attributesthedeclinetothecombination of macroeconomic forces and B&Ls unique operating structure. While B&Ls would specialize in mortgage lending throughout their history, they continually innovated on the liability side of their balance sheet. The two dominant plans during the late 1800s and early 1900s were the serial and permanent plans.1 In California, these B&L plans issued withdrawable shares, a form of equity contract. Withdrawable shares were frequently structured into series, a form of forced savings plan fornewmembers.2 Inaddition,withdrawablesharesfeaturedpenaltiesthatmadeitdifficultformembersto access funds on short notice. By the early 1900s, California B&Ls began to adopt the Dayton plan (named after the city in Ohio, where they originated), which did away with series and removed many aspects of the savings plan and withdrawal penalties. In California, instead of only withdrawable shares, Dayton plans frequently issued investment certificates, a debt contract. Investment certificates differed from withdrawable shares because withdrawal penalties were lower, so members could more easily access the full value of their funds, and there was a less rigid savings program. Investment certificates were comparable with certificates of deposit at commercial banks with a few additional restrictions common to B&Ls (Clark and Chase 1927). The simultaneous existence in California of Dayton plans, which emphasized low-cost savings, and non- Dayton plans, which emphasized regular savings and higher withdrawal penalties, presents the key source of liability heterogeneity studied in this paper. Due to the gradual nature of B&L innovation, in some states therewereperiodsofoverlapduringwhichthereexistednon-Daytonplansthatlookedmorelike“traditional” B&LsandnewerDaytonplansthatwerecloserinspirittocommercialbanks.Thispaperstudiesthestateof CaliforniaduringtheGreatDepression,whichwasoneofthestatesandtimeperiodswiththemostoverlap. 1.B&Lsoriginallystartedastemporaryinstitutionswhereafewmemberswouldpoolsavingstomakemortgageloans.This originalB&Ls,knownastheterminatingplan,hadonlyone“series”thatmemberscouldparticipateinbypurchasingshares, which were equity contracts in the B&L. Yet new members were often difficult to attract due to the planned closure of the institutionplustherequirementofback-payingearliersavings. 2.These B&L plans would continuously create new series to accommodate new members. The key difference between the serialplanandpermanentplanisthattheserialplanwouldstartnewseriesthatmemberscouldbuyinto,reducingthetotal amount of back-pay for new members, while the permanent plan allowed each individual to essentially start their own series, eliminatinganyneedforback-pay. 2

Ileveragevariationacrossthesetwotypesofplanstounderstandwhethermemberflightinessledtodifferent rates of liquidation during the Great Depression. B&Ls in the Great Depression offer an exceptional laboratory to study the effect of flightiness on financial distress. First, the proliferation over the past few decades of different types of derivatives and investments has complicated both sides of financial balance sheets, making it difficult to disentangle the relative effects of specific liabilities. B&Ls at this time had very simple liability structures that allow me to focus on the flightiness issue. Second, even if one could find a modern institution with a simple liability structure, it is equally challenging to find settings where the asset side of the balance sheet is relatively homogenous across institutions. Whether one looks historically or in the present day, the types of loans made by either commercial or investment banks vary based on sector (e.g. mortgage, commercial) or maturity. However, B&Ls in California had assets that were almost completely in real estate loans and, due partly to legal restrictions,verysimilaracrossinstitutions.Finally,theclosureofB&Lsinthistimeperiodisalsoattractive to study because reverse causality is unlikely to be a major concern. B&Ls were unlikely to have caused the Great Depression. Field (2014) shows that the impact on the housing market during the Great Depression was comparativelysmall relativeto theGreat Recessionbut notesthatwhile “[w]ehaveabundant historical evidence that commercial bank failures can pose a systemic threat to an economy, it is less clear that this would have been so with building and loans.” Similarly, White (2014) finds little impact of the housing market in the 1920s on the financial system. Ibeginbyestimatingacross-sectionallinearprobabilitymodeltodeterminetheeffectofbeingaDaytonplan on the probability of closure. I rely in this specification on two measures of the Dayton plan: the reported plan of the B&L in the annual report, and an alternative measure that leverages the liability structure. As California Dayton B&Ls issued investment certificates, I also use an alternative measure that compares associations with relatively more investment certificates to those with relatively less. Of the two types of B&L’s, Dayton and non-Dayton, Dayton plans were much more likely to close using eithermeasure.Theresultsarerobusttoanumberofalternativespecificationsthatcontrolforlocaleconomic conditions, competition from other B&Ls or commercial banks, and other potential balance sheet effects. In a quite restrictive specification, I condition on only counties with multiple B&Ls and include city fixed effects and find similar results. These results suggest there is something fundamental about plan type that predicts closure. Fortheobservedclosureratestobeconsistentwiththeflightinesshypothesis,non-Daytonplansshouldhave higher costs for members to access savings. I estimate different measures of access costs for the two types of B&Ls. One measure of access costs is withdrawal penalties, which I define broadly as being unable to withdrawforfullbookvalue.Ialsolookatdues,whichweretherequiredpaymentseachmemberhadtopay at regular intervals. I find that Dayton B&Ls were less likely to have withdrawal penalties and had lower dues on average. This result suggests that being a member at a non-Dayton B&L was costlier than at a Dayton B&L. Pairing the balance sheet information with archival information hand recorded from the California State Archives (CSA) in Sacramento, California permits a deeper dive into the differences between the two types of B&Ls. Members should be willing to pay higher access costs only if returns were also higher. I leverage detailed archival data in unpublished annual reports. While these data are only available for the year 1931 and for a subset of B&Ls, they provide a glimpse into returns for members across institutions. I find that 3

returns were significantly higher for non-Dayton plans. This result is driven mainly by the difference in returns across the two types of instruments, as investment certificates had lower returns overall compared with withdrawable shares. Paired with the result on access costs, this suggests that non-Dayton plans had high access costs but attracted members via higher returns. Finally, I present characteristics of the members of the institutions themselves. I show that the average wealth per member held in non-Dayton B&Ls was significant higher than that held in Dayton plans. I also showthatduringtheGreatDepression,membersofDaytonplansweresignificantlymorelikelytopaycostly fees to access their funds. These two results suggest that members were fundamentally different across the plan types, and therefore point toward flightiness as an important reason for closure. Ialsoprovideanumberofadditionalteststoshowthattheassetsideofthebalancesheetsacrossplantypes was very similar. Historically, B&Ls in California were required to lend against real estate. They followed national trends in providing long-term amortized loans that had proven popular among B&Ls in other states. Additionally, I show that the net borrowing cost for members was essentially the same, suggesting little discrimination among borrowers across plan types. Average loan sizes were also similar. It is important to emphasize that B&Ls did not fail in the conventional sense. While deposits at commercial banks were debt contracts, which banks were required to repay on demand, withdrawable shares issued by B&Ls were equity contracts. These members of B&Ls were therefore investors in the institution, with the value of their investment supposedly linked to the success of the B&Ls. California was no exception, with liquidation requiring the vote of two-thirds of total members. However, this paper is interested in the role of ex-antedifferencesinliquidityneedsbydepositors.Iarguethatthepropensitytoliquidatewasnotdifferent acrosstheinstitutionsduetothefactthattheshareofborrowingmemberswassimilar.Whilethedistinction betweencommercialbanksandB&Lsisimportant,Idiscussinthelastsectionoftheresultsofthispaperon the importance of liability structure can be used to inform the theoretical literature on bank failure. Related Literature The idea that liquidity shocks cause bank failure goes back at least as far as Diamond and Dybvig (1983). Liquidity shocks have also been used to motivate financial contagion (Allen and Gale 2000) or fickle international capital flows (Caballero and Simsek 2020). Liquidity can also be seen as disciplining the behavior of bank management, such as in Diamond and Rajan (2001) or Calomiris and Kahn (1991). More recently, the Great Recession has revitalized work on bank distress, both in the domestic context (e.g. Ivashina and Scharfstein2010;Shin2009)andintheinternationalcontext(e.g.Ivashina,Scharfstein,andStein2015).My paper suggests that liquidity shocks are endogenous to banking structure. I discuss further how my paper can help better understand such models in Section 6. Asecond,smallerstrandoftheliteraturehasdirectlyexamineddepositorheterogeneity.O’GradaandWhite (2003) study the Emigrant Industrial Savings Bank and show that the effect of the runs during the panics in the mid-1800s depended on whether depositors were more informed. Using depositor level data in India, Iyer and Puri (2012) and Iyer, Puri, and Ryan (2016) also show that depositor relationships with the bank matter. Beshears et al. (2020) randomly allocate withdrawal penalties and find that high penalties attract more committed depositors. My paper builds on this work and suggests that depositors are aware of the institutional structure of the banks they use. Asmallsetofpapershavealsostudiedearlywithdrawalsintimedepositsrelativetodemanddeposits.While 4

anumberofthesepapersarefocusedonthesensitivityofinterestratestomarketinterestrates,othershave studiedtherelativeimportanceofbankrisk,findingasimilarpatternasinthispaperthathigherwithdrawal feesareassociatedwithhigherreturns(BikkerandGerritsen2018).Therelativegrowthofnonbankfinancial institutions in the first half of the 20th century led to a number of articles emphasizing that interest rate differentials alone could not explain this phenomenon. The role of time deposits vs. savings deposits (Smith 1959) and commercial banks and savings banks (Alhadeff and Alhadeff 1958) have been explored to argue that the availability of savings as an important factor. Finally, this paper also contributes to a large literature using the Great Depression to understand how and why banks fail. Bank failures during this time period have been found to be due to insolvency (Calomiris and Mason 1997; Postel-Vinay 2016) or illiquidity (Blickle, Brunnermeier, and Luck 2019; Richardson and Troost 2009). The building and loan sector, studied in detail in this paper, has received increased attention inrecentyears.WorkbySnowden(1997),Snowden(2003),Fleitas,Fishback,andSnowden(2018),Fishback et al. (2018), Rose and Snowden (2013), and Price and Walter (2019) has established the importance of B&Ls in the institutional mortgage lending market in the United States in the first half of the 20th century as well as their lasting influence on the structure of the residential mortgage contract. Other papers on nontraditional financial institutions include Mitchener and Richardson (2013), who study non-member country banksintheGreatDepressionandfindalargeroleforfinancialcontagiontocities.Ialsocontributetobetter understandingthedevelopmentofCalifornia’sfinancialsector.Anattractivefeatureofstudyingcommercial banks in California during the Great Depression is the state’s allowance of branch networks. Recent work examining California’s experience with branch banking include Carlson and Mitchener (2009) and Quincy (2019). 2 Historical Background 2.1 Evolution of Buildings and Loans in the United States B&L’s were one of the most important lenders in the U.S. institutional home mortgage market over the first few decades of the 20th century.3 B&Ls were marketed as safe vehicles for savings, which permitted them to grow quickly.4 Figure 1 plots mortgage debt held by B&Ls for all single-family residential structures in both millions of dollars and as a share of the total amount of institutional mortgage debt. During the 1910s, B&Ls took on an increasingly larger share of institutional mortgage debt. Their importance peaked at just over 33% in the 1920s before collapsing during the Great Depression. The number of B&Ls in the country also doubled in the 1920’s, from 5,869 in 1920 to 11,777 in 1930, with assets per association nearly doubling over that same time period (Bodfish 1935). This first B&L in the United States, the Oxford Provident Association in Frankfort PA, followed what was known as the terminating plan. A group of households would get together and put forward funds for initial stock purchases in the association and commit to future saving. These funds were then auctioned to these members,andthememberwhobidthehighestforfundswouldobtainamortgageloanfromtheassociation, The amount bid, the “premium,” was discounted from the gross amount the household was able to borrow. This mortgage was accompanied by periodic repayments towards interest, amortization, and installment 3.For more in-depth historical overviews of B&L institutions, see Clark and Chase (1927), Bodfish (1935), Riegel and Doubman(1927),andSnowden(1997),amongothers. 4.Pieplow (1931) called Building and Loan Associations “the safest, most convenient, and fairest earning institution that wehavetoaidapersonwhoreallydesirestosaveandinvestmoney.” 5

on stock payments. As members saved and borrowers repaid, new members would then become borrowers. However, payments pre-specified the end date of the last mortgage payment, following which the institution was liquidated. As the plan was inherently temporary, which ran counter to goals of long-term savings, B&Ls soon took on two related forms called the serial plan and the permanent plan. These plans allowed for several series of “withdrawable shares” to be issued, each maturing at different times. The serial plan, which came first, allowed different series of withdrawable shares to be issued at regular intervals so that new members no longer had to back-pay larger amounts of funds the later they entered the association. Instead, members would be on equal footing with others from the same series. Another innovation of the permanent plan was that it allowed investors to purchase withdrawable shares without paying prohibitively large back-payments to catch up to earlier members. In other words, members implicitly began their own new series when they joined.However,membersstillhadtocommittoalong-termsavingsplan,andtheseassociationsfrequently had high withdrawal penalties. Takingthisideatothelimit,B&LseventuallydevelopedintoaformknowncolloquiallyastheDayton plan. Institutions using this plan allowed individuals to make payments whenever they pleased rather than at a regular interval. There were typically lower withdrawal penalties, and members could usually withdraw money on request (Pieplow 1931). Dayton plans were most common in Ohio (hence the name Dayton) and a few other states in the country, including California. Dayton B&Ls frequently issued some sort of debt contract rather than relying solely on withdrawable shares. The Dayton B&Ls in Ohio actually accepted deposits, which led to the observation that the Dayton B&Ls were “open to the charge of being savings banks, a term frequently applied as a stigma” (Clark and Chase 1927). On the lending side, the premium on loans was eliminated for Dayton plans. There were therefore two broad classes of B&Ls operating during the 1920s: Dayton plans, which were more closely related to commercial banks and catered to short-term investors, and non-Dayton plans (serial and permanent plans), which required more of a commitment by members. Both plans specialized in local real estate by permitting only its members to borrow. Table 1, taken from Clark and Chase (1927), shows the distribution in 1923 for the United States as a whole. Terminating plans were almost completely eliminated, accounting for less than 1% of the total. Serial or permanent plans accounted for 87%, while Dayton plans accounted for a little over 11%. By 1935, the federal government had implemented a number of new laws targeting the B&L industry that made it possible for B&Ls to “federalize,” or join the Federal Home Loan Bank system (in the same manner ascommercialbankscouldbecomeFederalReservebanks).Snowden(2003)discusseshowtheselawshelped create the savings and loan industry that would come to persist for the following decades. 2.2 California Building and Loans The reported history of B&Ls in California traces back to 1893, when the first annual report of the Office of the Board of Commissioners of the Building and Loan Associations was issued and the Building and Loan Commission was created. The earliest reports only mention plan type in passing and focus instead on whether or not members planned to become borrowers.5 By the third annual report in 1895, the Dayton 5.The 1893 report does not mention Dayton plans or Permanent/Serial plans. Instead, this report defines B&Ls based on their scope of operation (Local/National), and whether or not members plan to eventually become borrowers (Type of Premium). The latter distinguishes between the types of withdrawable shares issued by B&Ls: free shares (non-borrower) vs. 6

plan began to be used by two institutions in California. By 1900 in the seventh annual report, California was well aware of the transition from Permanent/Serial to Dayton: “...the old Terminating association was succeeded by the Serial and is now fast being succeeded by the Dayton.”’ By 1905, this number jumped to 24 officially listed. As described in detail by Haveman and Rao (1997), although the Dayton plan grew in popularity, the coexistence of these different types of B&Ls continued throughout this time period into the 1920s. Non-Dayton B&Ls issued various forms of withdrawable shares, which were equity contracts featured elsewhere in the country (e.g., in New Jersey as discussed by Fleitas, Fishback, and Snowden 2018). There were two main forms of withdrawable shares: installment shares or full-paid shares. Installment shares created the forced savings plan, as individuals would commit to regular savings until their total savings reached the value of an individual withdrawable share. Full-paid shares allowed individuals to simply purchase the full value of an individual withdrawable share. Withdrawable shares typically had variable returns based on the dividends of the institution and featured higher withdrawal costs as shown later in this paper. DaytonB&LsinCaliforniawereuniqueinthattheyissuedinvestment certificates,whichdistinguishedthem from other Dayton plans elsewhere in the country. Along with having lower withdrawal penalties relative to withdrawable shares, investment certificates were a debt contract that featured a fixed rate of interest. These investment certificates were senior to withdrawable shares in the event of liquidation (Stanford Law Review1950;Bodfish1935).ClarkandChase(1927)viewthesecertificatesascomparablewithcertificatesof deposit, as they make it “possible to withdraw money quickly and take it elsewhere.” This made it easier to attract new members. Unlike withdrawable shares, California B&Ls were required to keep a reserve on hand for investment certificates of 10% for any amount up to $1 million with an additional percentage that scales with the amount issued (e.g., 3% for any amount in excess of $5 million). This reserve could be composed of a standard reserve fund and/or what was known as “guarantee stock.” Guaranteestock wasanotherdevelopmentintheevolutionofbothDaytonandnon-Daytonplans.Guaranteestock plans allowed some members to purchase non-withdrawable stock in the institution, which was essentially the initial capital. This allowed the institution to begin making a higher volume of loans more quickly andguaranteesomeformofinterestordividendpayoutsforinvestmentcertificatesandwithdrawableshares, respectively. The institution could use this guarantee stock as a reserve for investment certificates and could also presumably respond more easily to withdrawal requests for both investment certificates and withdrawablesharesbyhavingsomecapitalonhand.InCalifornia,mostB&Lshadguaranteestockbytheendofthe 1920s. While dividends were not guaranteed, guarantee-stockholders would typically receive excess earnings beyond those allotted to other liabilities.6 The general shift towards Dayton plans reflects a financial environment motivated towards an efficient movement of funds in the face of large migration into California.7 Haveman and Rao (1997) and Haveman, Rao, and Paruchuri (2007) argue that the shift towards the Dayton plan was due to values related to the pledged shares (borrower). Importantly, this “premium” is not a characteristic of Dayton plans, suggesting that all B&Ls in CaliforniastilloperatedasPermanent/Serialplans. 6.ClarkandChase(1927)emphasizethat“[t]hepresenceofthecapitaloftheguaranteestockholders(afundwhichremains permanentlyintheassociationbusiness),thelendingoperationsarenotgreatlyaffectedeitherbytheentranceorwithdrawal of the temporary funds. If losses should occur before the contract with the temporary investors is completed, they could be absorbedbytheguaranteestockholders.Installmentshares,investmentcertificates,...canbeissuedbysuchassociationswith fullassurancethattheearningscontractedforcanbepaid.” 7.HavemanandRao(1997)outlinetheevolutionofplantypeinCaliforniaanddiscussitscauses.Foundingaccountedfor three-fourthsofcompositionchangescomparedwithtransitioningtoanewtype. 7

Progressivemovement.AdesireforefficiencypushedB&Lsfromclub-likenon-Daytonplanstobureaucratic Dayton plans. This change was propagated by internal migration and immigration into California, which expandedthesizeoflocalfinancialnetworksandreducedtheabilitytobuildlong-termrelationships.Dayton plans, attractive due to their low withdrawal fees and ease of access, began to grow. This general shift towards efficiency is similar to the overall transformation of California banking. As described by Doti and Schweikart (1991), a substantial portion of early banking along the frontier was highly localized. By the early 1900s, following a series of panics and dishonest bankers, state regulation began taking form and bank examiners began conducting regular examinations, thereby streamlining bank reporting. Doti and Schweikart(1991)arguethattheseexaminationscreatedopaquereportsfromtheperspectiveofthedepositor. Depositors increasingly relied upon the specialists’ determination of banking safety (even if such specialists were potentially unqualified and received the job due to political connections). Taken together, both the Progressive movement described in Haveman and Rao (1997) and the increasing reliance on specialists as in Doti and Schweikart (1991) suggest an important role for flightiness. First, shifting towards more efficient banking systems may have attracted newer members and depositors. These newcomers may not have been financially savvy and relied more on external regulators for safety. Second, as individual members grew wary of their fellow members or understood less about their local institutions, they may have been more likely to wish to withdraw funds in the event of a bad shock. To withdraw funds in California, members would formally request to do so in writing with at least 30 days notice.Thememberwouldreceivesomeamountuptothefullvalueofwhathepaidin,althoughwithdrawal values(especiallyforwithdrawableshares)werefrequentlylessthanthebookvalue.B&Lswerethenrequired to use up to 50% of receipts in a given month to respond to withdrawal requests. In California, associations wererequiredtopayallwithdrawalrequestsonfilewithinayearorallreceiptswouldgotowardswithdrawals. This was also true for investment certificates, which were similar to deposits in that they represented debt. Ifwithdrawalswerenotpaidoutwithintwoyears,thestatecommissionerwouldhavethepowertoliquidate the B&L according to the 1929 Civil Code. In California, the different plan types were evident in their advertisements. non-Dayton plans would state the overall return and in some cases directly emphasize the forced savings component. The left panel of Figure 4 shows for the Guarantee Building and Loan Association in San Bernardino, a non-Dayton plan in my sample that closed in 1930, both the savings plan component (“save $10 every month for but six and a half year”) as well as the overall return (“[e]very dollar has earned 8 per cent return”). Compare this to the Dayton plan advertisements. The right panel of Figure 4 shows an advertisement for the Guaranty Building and Loan Association in San Jose, a Dayton plan in my sample that did not close. One of the key features of their investment plan is the advertised ability to withdraw essentially on demand. Advertisements such as these were documented in a B&L post-mortem, with the Select Committee of the California Assembly for the Purpose of Investigating the Building and Loan Situation in the State of California noting that “...a definite relationship between advertising and present conditions exists ... those associations most active in advertising for new investors are those associations which are today suffering...” (Dawson et al. 1935). Even Commissioner Louis C. Drapeau noted in 1935 that “[t]he impressions that building and loan associations offered high interest on savings invested with them, and that investors could have their money returned to them at their demand, were eagerly accepted and believed by the average investor” (Drapeau 1935). Figure2showsthedevelopmentoftotalassetsandthetotalnumberofassociationsforreportingB&Lsfrom 1920-1934. As elsewhere in the country, during the 1920s the number of associations and the total number 8

of assets were on the rise. The number of associations peaked in 1929 with 233 associations, whereas the totalvalueofassetspeakedin1930with$513,110,594.58.Likeelsewhere,therewasstrictregulationlimiting California B&Ls to mortgage loans.8 Figure 3 maps the location of B&Ls in California. The location of B&Ls unsurprisingly tracks the population of the state as a whole.9 The right panel of Figure 3 shows the distribution by plan type. Dayton plans were more common in the state as a whole but were not obviously overrepresented in any specific location. TheexplosivegrowthinB&LsinCaliforniabegantoattractnotice,andCaliforniastateofficialsalsostarted consideringadditionalregulation.10 AlthoughB&Lscontinuedtoincreaseinsizethrough1930,newcommissioner Charles Whitmore wrote to the governor that “Loan commitments by associations showed a decline for the year of 38 per cent” (Building and Loan Commissioner 1930). However, he did not see any cause for concern, writing that “conditions in many parts of the state show signs of returning normality, and more and better loans are now being offered for association investment.” The 1930s would be hard for B&’s, as the total number operating in California declined from 233 total associations in 1929 to 178 associations in 1934.11 In the 1932 annual report, new commissioner Friend W. Richardson,wrotethat“[t]heyear1932wasthemostcriticalinthehistoryofbuildingandloanassociations” (Building and Loan Commissioner 1933) Through 1934, the number of associations and the total amount of assets was on steady decline, as was common throughout the country. The relatively high closure rates of Dayton plans, which relied more on these investment certificates, was noted by contemporaries. In the Minority Report by the California Legislature, Chairman Frederick Peterson notes that “complaints were directed against stock organizations - particularly those ... affiliated with companies dealing in pass book and investment securities.” His recommendation was the elimination of this system, emphasizing that B&Ls had led investors to see passbooks and investment certificates as deposits (Peterson 1935). 2.3 Do B&Ls Fail? B&Ls were fundamentally different institutions than commercial banks. Commercial banks’ main source of liabilitiesweredepositorsthatowneddebtcontractsinthecommercialbank.Intheeventthatabankcouldn’t payoutdepositors,thenthebankcouldbeforcedtoclose.However,forB&Lsthewithdrawablesharesthey issued were in fact equity contracts. This meant that along with involuntary liquidation mentioned earlier, shareholders could choose to voluntarily liquidate the B&L or merge with another association. The voluntary liquidation option has been studied in the New Jersey context by Fleitas, Fishback, and 8.Forexample,TitleXVIoftheCaliforniaCivilCodein1929requiredloanstobesecuredbya“firstmortgageordeedof trust upon unencumbered real estate having an appraised value of not less than 25% in excess of the face of the loan” (with someexceptions).Asof1917,borrowerscouldrepaytheloanatanytime.Ifaborrowercouldn’tpayhisorherdebts,theB&L could,afteraperiodof6months,issuehimanoticeofdefaultinwriting.Iftheborrowerdidn’trepayhisorherdebtswithin 2 months, he or she is in default, and the association may, by law, purchase the property. No other restrictions were in place accordingtolaw,andassociationswerefreetosetotherterms. 9.Additionally, many cities did not have more than one B&L. Conditional on having at least one B&L, approximately two-thirdsofcitieshaveexactlyone.However,assumingB&Lscompetedwithcommercialbanksforsavings,thenthenumber of banking institutions per city is likely much higher. An important assumption in this paper is that B&Ls of a specific type arenot“theonlygameintown”forsavings. 10.Inthe1929annualreporttotheGovernorofCalifornia,CaliforniaStateCommissionerGeorgeWalkerwrotethatwhile “Buildingandloanlawshavebeenmateriallystrengthenedduringthelastfewyears,...additionalpowershouldbegrantedyour commissioner.Becauseoftheadvertisedprofitabilityofthebuildingandloanbusiness,promotersandothersareendeavoringto organizenewassociationsineverypartofthestateregardlessofthefactthatinmanysectionsthebusinessisalreadyoverdone” (BuildingandLoanCommissioner1929). 11.Following this year, it is difficult to track the total number operating. B&Ls were allowed to federalize, and the state commissionerdidnotcompilestatisticsonfederalizedB&Lsduetonewlypassedlegislation. 9

Snowden(2018).InNewJersey,voluntaryliquidationoccurredwhentwo-thirdsofmembers,eitherborrowing or non-borrowing, voted to liquidate. They find that the probability of liquidation rose when there was a higher share of non-borrowers. The same laws were in place in California. Paragraphs 83 and 87 of the 1891 California B&L act dictated that “dissolution” can also be either involuntary (if the association commits a crime or “unsafe practice”) or voluntary. In 1911, the commissioner was given the power to “revoke the license of any ... association ... [whose] solvency whereof may have become imperiled...” In California, voluntary dissolution always required a two-thirds majority, as in New Jersey. Alternatively, members could theoretically sell their withdrawable shares (or investment certificates) in informal secondary markets. Rose (2014) studies the markets in New Jersey but finds that these secondary markets were common throughout the country. There is evidence that these markets existed in California, as he finds that as late as 1934, share prices in San Francisco were 50 cents on the dollar. However, Rose (2014) finds that these markets were not fully mature until the late 1930s, making it unlikely that members could easily sell shares during the early stages of the Great Depression. The closure of a B&L was a complex affair. Whether an association chose to vote to close or to engage in a lengthy court battle to prove insolvency meant that, in some cases, years could go by before a result was determined. The analysis in this paper relies on the fact that closures were not necessarily quick but were driven by the surprise shock of the Great Depression and occurred by 1935. 3 Data I draw on historical data on B&Ls in California. I focus on California for a number of reasons. First, California has a non-trivial share of non-Dayton and Dayton plans, unlike almost every other state in the country. Second, the amount of money invested, in terms of assets per member, was higher relative to the UnitedStatesasawhole.In1923,assetspermemberinCaliforniawere$1,014.22comparedwith$486.96for the United States (Clark and Chase 1927). Members in California presumably relied more heavily on B&Ls as a source of investment making the B&L choice salient. Third, data availability makes California B&Ls attractive to study. Annual balance sheet and profit and loss data is available from the Annual Reports of the State Building and Loan Commissioner. Additionally, select underlying micro data from the annual reports has survived to provide additional insight into how B&Ls operated during the Great Depression. Finally,California’sBuildingandLoanLeaguewasactiveinpreventingso-called“National”B&Ls,orB&Ls headquartered outside of the state of California, from entering. Thus, nearly every B&L operated almost exclusively in California, limiting the effect of external factors in determining closure rates. 3.1 Public Annual Reports Data I use the appendices to the 1927, 1929, 1930, and 1935 annual reports to construct a cross-section of B&L balance sheets in California. The focus on these years is due both to data availability and economic history. First, the 1927 annual reports explicitly stated whether the institution was a Dayton plan. Data availability in 1927 was also at its highest. Along with balance sheets, which were available every year, the 1927 annual reportsalsohavedataonmembercontractssuchasdues,withdrawalvalue,anddividends.The1929annual reports provide baseline characteristics observed just prior to the onset of the Great Depression, avoiding 10

any effects from depressed aggregate economic conditions.12 This implicitly assumes that the onset of the Great Depression was sufficiently unexpected that the decision to start and operate a B&L by 1929 was independent of this aggregate shock. I use the cash-flow statements from the 1927 and 1930 annual reports, as the 1929 annual report does not include this information. I obtain the operating status from the 1935 annual report, which includes the effect of the Great Depression and limits the effect of federal programs, such as the Federal Home Loan Bank System, that may affect decisions to remain open. An example of a balance sheet for a Dayton B&L is displayed in Figure 6a. Starting from the top of the figure, there is demographic information about the B&L, such as the number of members/investors and shares (which appear to include both withdrawable shares and investment certificates). In the middle of the page there is balance sheet data. On the asset side, the large reliance on real estate loans is clearly visible. On the liability side, we can see the importance of investment certificates (listed as the third item). Finally, at the bottom of the page, one can see clearly that the association is labeled “Dayton Plan.” There is also additional data on dues and withdrawal value. Figure 6b presents a non-Dayton plan. The key difference is the reliance on withdrawable shares in liabilities, rather than investment certificates, and the listing of individual series at the bottom of the page. From the 1929 annual report, I record the complete balance sheet of each B&L. From the 1927 annual reports, I record the total number of members, the total number of shares, dues per certificate or withdrawable share, and any description of withdrawal value. While dues per share could differ across series for non-Dayton plans, in practice they did not. Using the reported plan type from the 1927 annual report should represent well the operations of the B&L, particularly how the managers aimed to attract new members. However, this measure ignores the fact that manyB&Lsissuedbothwithdrawableshares(typicalofnon-Daytonplans)andinvestmentcertificates(typicalof Daytonplans).I constructanalternativemeasureby consideringonlythe observedliability structure of the balance sheet. B&Ls reported “withdrawable shares” separately from “investment certificates.” I calculate the share of investment certificates relative to the sum of investment certificates and withdrawable sharesin1929foragivenB&L.IdiscretizethismeasurebycomparingittothemedianvalueacrossB&Ls.I call thisthe “liabilities”measureof theDaytonplan. Figure 5shows,for Dayton andnon-Dayton plans,the ratio of investment securities to the sum of investment securities and shares, meant to capture how much a B&L’s standard liabilities are in one or the other. 13 The preferred specification is to use the reported measure rather than the liabilities measure, as this probably more accurately captures the managerial decisions of how to attract new members, but I present results using both measures. From the 1935 annual report I record the operating status of B&Ls and the date in which the institution ceasedoperating.Thereasonsforceasingoperationsareclassifiedasoneofthefollowing:absorbed,removed, consolidated, transferred, merged, revoked, federalized, and liquidated (both voluntary and involuntary). I countasclosuresthoselistedasabsorbed,liquidated,transferred,andconsolidated.14 Ifthebusinessislisted 12.Prior to 1931, fiscal years were not uniform. The reports were filed at the end of the calendar year. The fiscal years are mostlytheDecemberoftheprioryearortheJuneofthecurrentyear,withapproximatelyhalfofeach. 13.Whilealargeshareofeachtypespecializeasexpected,thereareanumberofB&Lsthatdonot.Non-Daytonplanshaving a large share of investment certificates is likely due to plans changing over time to take advantage of new plan forms. The numberofB&Lsthatendogenouslychangeplantypeisnotadrivingforceintheoverallcomposition.HavemanandRao(1997) findthatthemajorityofchangesincompositionisduetoentry.Interestingly,therearealsoasignificantnumberofplansthat were originally Dayton plans have very low shares of investment certificates (in some cases they have none). It is difficult to knowexactlywhysuchinstitutionsexist.Onereasonmaybebecausetheannualreportsdidnotseparateinvestmentcertificates until the 1908 annual report. These institutions were likely always Dayton plans, but perhaps never changed the reporting of the investment certificates. Alternatively, over time Dayton plans may have preferred the traditional method of issuing series andsoswitchedtheirliabilitiesstructure. 14.absorbed, consolidated, and transferred occur when a B&L is bought by another B&L. I treat this as a closure as, 11

as removed, I list them as open, as these represent relocations or name changes. I drop B&Ls that closed prior to 1929. There are 55 closures from the 1927 listing of B&Ls in my sample. This number rises to 76 closures when using the liabilities measure, which relies on 1929 balance sheets (and so includes B&L’s started in 1927, 1928, and 1929).15 The final sample contains 164 non-federalized B&Ls active in 1927 in 1927 and 205 non-federalized active in 1929. In California in 1927, there were significantly more Dayton plans than non-Dayton plans. This is true not only in the state as a whole, but also within counties. From Figure 3, which shows the distribution of B&Ls and their type, we can see that the majority of counties with at least one B&L also have at least one of each type. 3.2 Archival Data I hand-record surviving archival data available from the CSA in Sacramento, California. I use raw copies of the detailed balance sheet data submitted by the B&Ls that were maintained by the Los Angeles office.16 Theserecordingsformthebasisofthereportedbalancesheetsintheannualreports.Alongwiththepublicly availableinformation,theyalsoincludeadditionalstatisticssuchaslendingratesandmemberreturns. These unpublished recordings contain a wealth of useful information.17 I observe the reported interest on mortgagelending,eithertheaverageorinsomecasessimplyalistorrangeofinterestratesonloanscurrently outstanding. There are also details on the average rate of interest on investment certificates. These archival reports are only available at 5-year intervals for a limited number of B&L’s (specifically, 1926, 1931, and 1936). I focus on the 1931 annual reports, which is the earliest year for which a substantial number of B&Ls havesurvivingbalancesheetdata.18 Ihandmatchthesedatatothe1927and1929balancesheetdata.Iam only able to match around half of the sample. For the remainder, the B&Ls either closed before 1931 or the reports did not survive. 3.3 Summary Statistics Summarystatisticsforthesetof205non-federalizedB&Lswith1929balancesheetinformationarereported in Table 2. Thistable includes mergedcityor county level data froma varietyof other sources.19 Of the164 institutions with 1927 balance sheet information, approximately three-quarters report as Dayton plans. Of the 204 institutions with balance sheet data in 1927, around 37% of B&L’s close according to my definition. Theaveragenumberofmembersandassetsarearound1,400and$2millionrespectively,althoughthelargest B&L’s have 9,000 members and $30 million in assets in 1929. according to the 1910 annual report, “The larger volume of assets, coupled with a good reserve, attracts the attention of the public,commandsrespect,andattractsmoreandbetterbusiness.”Iinterpretthisstatementassayingtheinstitutionswould haveclosedifnotconsolidatedwithalargerenterprise.Closediseitherduetoinvoluntaryorvoluntaryliquidation.Transferred implies that assets were shifted to another B&L. As the B&Ls where the assets are transferred are not started in the same year the transfer occurs, these do not seem to be simple relocations, which are listed separately. Rather, this appears to be somethingclosertoasaleoftheinstitution. 15.TableB4showsthedistributionofclosurecodesbyplantype,andmostareeitherabsorptions,transferals,orliquidations bythecommissioner.TableB5showsthedistributionofclosuresacrosstime,andmostclosuresoccurin1929-1931asexpected. 16.The CSA has records organized by either the San Francisco or Los Angeles office. The San Francisco office consists of recordsonlysince1968.TheLosAngelesofficehassomerecordsdatingbackasearlyasthe1900s. 17.AnexampleofmemberreturnsisgiveninFigureB3. 18.There are some reports in 1926 for an extremely limited number of B&Ls. There is balance sheet information on surviving1936annualreports.However,survivorbiasconcernsaremagnifiedevenmore.Additionally,anyinstitutionsthatwere federalizednolongerreportbalancesheetstothestateregulators. 19.SeeAppendixAforadetailedlistofsources. 12

To better understand the difference across institutions, a balance table for non-Dayton and Dayton plans is given in Table 3. Some key differences stand out. First and foremost, non-Dayton B&Ls were older. This is not unexpected; the historical development of B&Ls and the relatively recent development of the Dayton plan development predicts this age difference. Second, turning to the balance sheets, Dayton plans were larger in terms of both assets and members. Third, unsurprisingly, the composition of balance sheets differs asDaytonplansreliedoverwhelminglymoreoninvestmentcertificatesintheirliabilities(includingguarantee stock), while non-Dayton plans relied more heavily on withdrawable shares. Both make up more than half of their liabilities on average. If anything, Dayton B&Ls had more liquidity available in terms of cash ratios. Part of this was due to the legal requirements on maintaining reserves when issuing investment certificates. However, there is no significant difference among institutions in terms of real estate owned. Dayton plans were also more likely to be located in larger cities with more commercial banks. In the appendix, I show which factors play a predictive role in determining B&L plan choice.20 Age is by far the most important predictor of the Dayton plan. In all cases, the age variable is highly significant. Conditionalonage,oftheobservablelocalvariablesonlylogpopulationismarginallysignificant.Thisresult is consistent with the argument in Haveman and Rao (1997) that Progressive values and the desire for more efficient institutions in response to immigration led to the adoption of the Dayton plan. 4 Closure Rates I first show that the probability of closure for Dayton B&Ls was higher relative to non-Dayton B&Ls. I estimate the following regression model by ordinary least squares (OLS) Closure =α+βDayton +ΓX +ε (1) i i i i whereClosure isadummyvariableequalto1ifB&Liclosesbetween1929and1935,Dayton isadummy i i variable equal to 1 if the institution is a Dayton plan, X is a vector of controls at the B&L level, and ε is i i theerrorterm.Thecoefficientofinterest,β,representstherelativeincreaseinclosureratesforDaytonplans compared with non-Dayton plans. This coefficient is hypothesized to be positive, indicating that Dayton B&Ls were more likely to close. I show results using both the reported and liabilities measures. In a causal sense, the identifying assumption in this model is that the decision of whether or not to use the Dayton plan, or issue relatively more investment certificates, is uncorrelated with other determinants of closure that would be included in the error term ε . Some threats to this assumption are observable and can i be directly controlled for. First, the size of B&Ls may be an indicator of distress. If larger B&Ls are more diversified or more efficient, then the coefficient β may be biased, as Dayton B&Ls were on average slightly larger.Toaccountforthispossibility,IincludelogassetsascontrolsinX .Second,theageoftheinstitution i is frequently found to be an important determinant of closure. I use age group dummies to account for this concern.21 A third threat to identification is the vulnerability of the B&L due to the maturity mismatch of the balance sheet. While I have already argued that the structure of the asset side of the balance sheet is similar for both types of B&Ls, liquidity ratios differed across institutions. For example, all plans were 20.I regress the self-reported Dayton variable on B&L age as well as a number of other local indicators. I estimate this regressionviaOLS,probit,andlogit.TableB6showstheresults. 21.I bin the ages into decades: ages 1-4, 5-10, 11-20, 21-30, 31-40, and 40+. The 1-10 bin had the largest number of B&Ls, soIfurtherdivideditinto1-4and5-10tohavetwobinsofapproximatelyequalsizes.Thissecondsubdivisiondoesnotaffect theresults. 13

requiredtoholdreservesagainsttheoutstandingvalueofinvestmentcertificates.Thiswouldnaturallyimply that Dayton plans, which issued more investment certificates, had higher liquidity ratios. I include the cash ratio as a control to account for this possibility. Another set of threats to identification is local economic conditions, such as the size of the local population or commercial bank competition. Local banking competition may push B&Ls to take the Dayton plan. This competition may also result in higher closure rates if banking panics spread locally. This would bias the estimate of β upwards. I include both the log population and the log number of commercial banks in the city as controls to account for this possibility. I also show the results are robust to the inclusion of city fixed effects. The estimates of β support the hypothesis that Dayton plans did close at higher rates. Table 4 reports the resultsfromestimatingEquation(1)viaOLS.22 Thefirstcolumnreportsresultsfromthebivariateregression of closure on only the reported Dayton measure (without any controls). The point estimate of 0.224 (SE: 0.07) implies that Dayton institutions had higher closure rates on the order of around 22 percentage points. The second column includes B&L size and balance sheet controls. The coefficient β changes only slightly to 0.238 (SE: 0.08) but remains significant both economically and statistically. In the third column, I include the age dummies and the coefficient estimate again remains broadly unchanged but note that the standard errorswidenduetothehighcorrelationbetweenageandplantype.Finally,thefourthcolumnreportsresults including the local controls and the point again but remains of similar magnitude and is significant. I repeat this ordering in the last four columns using the liabilities measure of Dayton plan and a similar pattern emerges. The results are robust to a number of different specifications and sample selection decisions. Table 5 reestimates the benchmark specifications under alternative specifications. The first column simply replicates the third column of Table 4 for convenience. The second (sixth) column includes city fixed effects. The third column restricts the results to counties with at least one of each type of B&L present (while still including city fixed effects), and the estimate is unchanged. Finally, the fourth column drops the two largest counties: SanFranciscoandLosAngeles(stillincludingcityfixedeffects).AlthoughLosAngeleshadalargenumberof closingDaytonplans,thattheresultsarerobusttodroppingthesecitiesisstrongevidenceoftheimportance of plan type. The next four columns focus on the liabilities measure. A similar pattern emerges, and the results are highly significant with the city fixed effects across all specifications. The results in this section strongly support the hypothesis that Dayton plans had closure rates that were significantly higher than non-Dayton plans. In the next section, I dig deeper into the mechanism driving this result by using information on access costs, returns, and measures of liquidity needs. Before proceeding, I discuss a number of robustness checks regarding the stability of the results when including additional controls. I then briefly discuss a number of additional checks available in the appendix. Robustness Checks In the appendix, I investigate the stability of the coefficient estimate subject to other controls in order to address various identification concerns.23 I show that the results are not sensitive to balance sheet measures 22.Resultsarerobusttousingprobitorlogitspecifications. 23.SeeAppendixC.Alongwithrealestateownedandtheconcentrationindexmentionedinthisparagraph,Ialsoshowthat the results are robust to the inclusion of other measures of borrower characteristics, asset-side variables, and additional local controls. 14

of borrower quality. If Dayton plans borrowers were more ex-ante likely to default in general, then the differential closure rates I identify may simply be due to the impairment of assets. Real estate owned shares isausefulproxyfordefaultrisk.AsemphasizedbyFleitas,Fishback,andSnowden(2018),thisassetincludes foreclosed property taken on by the B&L. The results are robust to the inclusion of this control. I also show thattheownershipstructureoftheB&Lisnotaconcern.Fleitas,Fishback,andSnowden(2018)discusshow B&Ls in New Jersey could close with a 2/3 majority vote by shareholders and stockholders. The regulations on closure were similar in California. I construct a “concentration index”, which is the sum of withdrawable shares and guarantee stock as a share of assets. This measure captures how much the B&L relied on voting members. Including this measure does not affect the point estimate. I explore a number of additional robustness checks in the appendix. I show that dropping either involuntary closures or consolidations and transfers does not significantly affect the results. Dropping involuntary closures homes in on the liquidity decision by focusing on whether members would be willing to liquidate the institution to access funds. Dropping consolidations and transfers is a robustness check on the classification of closure codes. I also examine the decision to federalize. B&Ls that may have liquidated might instead choose to federalize instead. Due to the distress faced by B&Ls during the Great Depression, U.S. federal policy in the 1930s allowed B&Ls to federalize and join the Federal Home Loan Bank system, created in 1932. In the appendix I show that treating federalization as either closure or as an independent outcome in a multinomial logit framework does not affect the results. I also show that the results are not sensitive to survivor bias on the part of non-Dayton plans that survive earlier recessions. 5 Costs, Returns, and Lending Rates In this section, I investigate why there were higher closure rates at Dayton plans by focusing on the characteristics of the B&L plans’ liability structures. I first show suggestive evidence that non-Dayton B&Ls had higher access costs and higher withdrawal penalties for members. To account for higher costs, I then leverage the archival data to show that non-Dayton B&Ls attracted members by offering higher returns, but that lending rates loan characteristics were largely equal across the institutions. Taken together, I argue this framework resulted in having members that were less flighty (ex-ante less likely to need to access their funds during a shock). As additional evidence, I use reported withdrawal fees during the Great Depression to show that liquidity needs seemed higher at Dayton plans. For brevity, I focus on the reported measure in the tables that follow.24 5.1 Access Costs: Withdrawal Fees and Dues I begin by comparing the withdrawal penalties across plans in California. As elsewhere, in California, withdrawal penalties could first be in the form of timing restrictions or fees. The first page of the 1927 annual report notes that “many associations in the past have advertised that money might be withdrawn at will by the investor, and the public has come to expect it,” suggesting that in some cases individuals tended to believe they could withdraw on demand, with little to no penalty. In California, withdrawal of both withdrawable shares and investment certificates were subject to up to 30 days advance notice. Withdrawal fees in California were more lax than in other parts of the country. Clark and Chase (1927) note that Califor- 24.Resultsusingtheliabilitiesmeasureareavailableintheappendix.Theresultsarequalitatively,andinmostcasesquantitatively,similar. 15

nia is one of only two states that does not permit forfeiture of principal when investors withdraw either installment shares or investment certificates. Instead, entrance fees or withdrawal fees are charged. Clark and Chase (1927) note that these fees may be high enough to effectively reduce the principal if an investor withdraws too early. Withdrawal penalties were not explicitly listed in the 1927 balance sheets. Instead, information regarding the value of withdrawals was presented. Clark and Chase (1927) state that “[a]ssociations using the Dayton plan ... customarily repay to withdrawing members the full book value of their investment.” This statement suggests that withdrawal fees are low, but that whether a member receives book value is a good measure of withdrawal cost. The appendix to Clark and Chase (1927) also describes withdrawal fees as “Deductions from book value when shares are withdrawn before maturity.” Relative to book value, profits were more variableandwerepaidoutonlyonspecificdates.25 The1891annualreportinCaliforniaalsofoundthatthe average amount of profits paid out was only 50% of the total accrued, suggesting this is a good measure of withdrawal penalties. For Dayton plans, withdrawal values were listed in the 1927 annual report as either “Full Book Value” or “Dues plus Profits.” I treat Dues plus Profits as a withdrawal penalty. Non-Dayton plans explicitly listed the withdrawal value for each share series, as shown at the bottom of Figure 6b. If this withdrawal value was less than the listed book value, then I consider this a withdrawal penalty under the definition by Clark and Chase (1927). In no case is the total withdrawal value less than dues, so the penalty is on the returns rather than the principal itself. I compare withdrawal costs using the benchmark regression specification as in Table 4 but set the outcome variable to be a dummy equal to 1 if an institution has withdrawal penalties.26 The first column of the top panel of Table 6 shows that Dayton plans were significantly less likely to have withdrawal penalties (after controllingforB&Landlocalcontrols),beinglowerbyaround50%,usingthereportedmeasure,conditional on observable B&L characteristics. Inextanalyzecostsasproxiedbydues.Duesarewhatisowedateachmeetingforforcedsavingsplans.The traditional Dayton plan, as in Ohio, would not have any dues listed, but in California there could be forced savings plans even for investment certificates. For my purposes, I am interested in whether these dues were different between Dayton and non-Dayton B&Ls. If the dues structure is lower at Dayton plans relative to non-Dayton plans, then this means that the forced savings plan for Dayton plans was less restrictive, which I consider to be a lower cost. Dayton plans listed the dues per share (or per certificate) per month, and this appeared to be the same amount for all members. Non-Dayton plans listed dues per share for each series, as shown in Figure 6b.27 I compare dues directly in the second column. Dues per share were around 10 cents lower for Dayton plans according to the reported measure, or just over half of a standard deviation. Comparingonlyduespershareleavesoutthefactthatmembersofnon-DaytonB&Lsmayholdfewershares in total. This would mean that the total amount of dues paid could be the same across institutions. To account for this possibility, I examine the number of withdrawable shares and certificates per member. The third column shows that Dayton B&Ls had significantly lower log shares per member. In the last column, I show that a measure of total cost of dues per member (or the product of columns two and three) is 25.One would have to hold their savings in the institution until at least those dates to make a return. This contrasts with bookvalue,whichwouldnotbesubjecttodividenddates. 26.In this section, I control for age by using a dummy if the institution was incorporated prior to 1920. Due to the smaller sample,somepreviousagebinshadveryfewassociationssoIelecttopoolthem. 27.Forthesenon-Daytonplans,eachseriescouldtheoreticallyhavedifferentcosts.Inpracticethiswasnotthecase. 16

approximately $6 less for Dayton institution compared with non-Dayton institutions. Since Dayton plan total costs were only around $4.75 per member, then non-Dayton plans essentially had three to four times the cost, all else equal. In sum, the detailed data on withdrawal penalties and costs, when paired with the historical narrative, provide indicative evidence that accessing funds was more difficult at Dayton plans. In addition, members at Dayton plans had higher costs of membership, not only because they held more shares on average but in part due to Dayton plans charging lower dues. 5.2 Archival Data: Investor Return and Borrower Characteristics Having presented evidence that access costs were higher at non-Dayton B&Ls, I now use the archival data to study returns and lending rates. The detailed annual statements in the archives provide information unavailable in the publicly available reports that help to answer this question.28 I begin by showing that returns were higher at non-Dayton plans. The first column of the middle panel of Table 6 shows regression results of the observed investor rates on a dummy variable equal to 1 if an institution is listed as a Dayton plan, again controlling for B&L and local controls.29 Dayton plans were associated with returns that were lower by around 23 basis points. In the bottom panel, using the liabilities results in lower returns by about 14 basis points. Relative to variation in lending rates, this difference is economically meaningful. Returns, on average, were around 6 percent with a standard deviation of 37 basis points, so the result is approximately one-third to two-thirds of a standard deviation.30 High returns alone do not imply that investors are being compensated exclusively for giving up liquidity access. First, high returns could compensate members for their time screening or monitoring loans issued. This is unlikely to be the case. Even if the withdrawal fee is a screening tool for potential borrowers, it only matters for the share of members that do actually plan to borrow. For the remaining members, this fee purely affects liquidity access. While the early history of B&Ls in the 1800s involved members that were specificallylookingtofinanceahome,bythe1920sand1930sadvertiserswereclearlystressingjoiningB&Ls for purely savings reasons. In fact, as shown in the next column, the ratio of borrowers to members was if anything lower at Dayton plans, although the results are mixed depending on the definition used. Second, one may be concerned that returns reflect a risk premium and compensate members for actually extendinglower-qualityloans.ThethirdcolumnofthemiddlepanelofTable6showsthatlendingrateswere similar or even higher at Dayton plans by around 30 basis points. However, it is likely that the net lending ratesweremoreequalthanthissimplecomparisonsuggests.Daytonplanshadeliminatedthepremium(the amount, bid by the borrower, by which the gross value of the loan was reduced). Dayton plan members and borrowers likely internalized the premium, reflecting it in the lending rate rather than the net amount 28.TableB7comparessummarystatisticsforB&Lsthatdoanddonothaveavailablemicrodata.Observablecharacteristics aresimilar.However,thesampleofB&Lswithmicrodataissmallerandhaslowerclosurerates.Thisisexpectedduetousing 1931data,andsoresultsusingthesedatashouldbeinterpretedwithcautionduetopotentialissuesofsurvivorbias.Forthe exercisesusingthesedata,IuseallB&Ls(whetherornottheyfederalized)inordertoimprovestatisticalpower. 29.Returns are calculated as the weighted average of returns for investment certificates and withdrawable shares, where weightsaregivenbytherelativeshareofeach.Returnsfortheinstrumentitselfisgivenbythesimpleaverageofthereported returnsifmorethanoneislisted. 30.In unreported results, I check whether this is driven by composition differences (Dayton plans having more low-return investmentsecurities)orifDaytonplanssimplypaidoutlowerreturns.Forplanswithbothinvestmentcertificatesandwithdrawableshares,Iregressthereturnforthespecificsecurityonplantype.Theresultsaremixedanddependonwhichdefinition isused,andsoIelectnottoreportthem.Hence,themainresultislikelydrivenbycompositiondifferences. 17

borrowed.Finally,averageloansizesareroughlysimilaracrosstheinstitutions,asshowninthelastcolumn. Loansdonotappeartohavebeenriskiersimplybecausetheywerebigger,asshowninthelastcolumn. It is important to reiterate that all lending rates and returns are as of 1931 due to data availability. The regulatory landscape during the Great Depression significantly changed due to the passage of the Building andLoanActin1931,whichmadedatacollectionapriority.Using1931excludesB&Lsthatclosedinthelate 1920s, many of which were Dayton plans. One concern is that the Dayton plans that closed had offered high interest rates on investment certificates that they were unable to pay out, and thus closed. Unfortunately, given the data restrictions, I cannot exclude this as a possible explanation. 5.3 Member Liquidity Needs I now argue that the potential liquidity needs of members was higher for those at Dayton plans than non- Dayton plans. I have already shown evidence that members at Dayton plans held fewer withdrawable shares orinvestmentcertificatesthanthoseatnon-Daytonplans.Ifparvaluesweresimilarandindividualsinvested the same share of wealth at B&Ls across type, then this would suggest that Dayton plan members were of lower wealth than members at non-Dayton plans. Unfortunately, it is not immediately clear what the par value per share is from the available data, and without information on member characteristics it is even less clear whether investment behavior differs across plan type. A straightforward way to observe liquidity needs is to ask whether members were willing to pay costly fines, fees, or penalties to either access funds or stop regular savings plans during the Great Depression. Why would feesbea good way tomeasure liquidityneeds? Clark andChase (1927)describesuchfees asbeingan important tool to ensure regular savings. They note that “[i]t is well known that fees, fines, and forfeitures wereoriginallydesignedtoencouragepersistenceinsaving”(ClarkandChase1927).Ifsuchfeesareinplace to encourage thrift, then it follows that whenever members are willing to pay it is to deviate from savings plans due to liquidity needs. IuseobserveddifferencesinfeesandcalculatetherelativeincreaseinfeespaidduringtheGreatDepression. In1927andin1930,theprofitandlossaccountsontheannualstatementsincludedvariousmeasuresoffees. As the categories listed are different in the two years I record, I define as fees any line item that uses the words “fines” or “fees.” I then calculate the sum of all fees and divide by total assets in 1927. By dividing by assets in 1927, all changes are due to changes in fees and total assets exist simply to scale the outcome variable. One issue with this definition is that it includes fees paid by borrowers who are late on repayment, therebyincludingsomemeasureofex-postassetquality.However,asIhaveattemptedtoargueinthispaper thatassetqualityisrelativelysimilaracrossinstitutions,thedifferenceinfeespaidacrossinstitutionsshould largely reflect liquidity needs. Additionally, I can econometrically account for this difference in the analysis below. I estimate the following regression: Fees Assets =α +γ +λ(DAYTON ×Year )+ε it i t i t it where DAYTON is a dummy equal to 1 if the institution is a Dayton plan, Y1930 is a dummy equal to 1 i t if the year is 1930. α and γ are association and time fixed effects, respectively. i t Themaincoefficientofinterestisλ,whichrepresentstherelativeincreaseinfeesduringtheGreatDepression 18

forDaytonplans,relativetonon-Daytonplans,comparedwithtranquiltimes(priortotheGreatDepression). Ihypothesizethatλ>0,whichmeansthatfeesroserelativelymoreforDaytonplansrelativetonon-Dayton plans. This would imply that members at Dayton plans were willing to pay to withdraw their money, or at least stop using their savings plans, more than those at non-Dayton plans. I treat this as a test of liquidity because it implies that funds are more needed outside of a savings vehicle rather than inside. Thisspecificationisastandard2x2differences-in-differences.Theidentifyingassumptionsareparalleltrends (the difference in the outcome would have been the same in the absence of treatment) and exogeneity of treatment. I have already argued in this paper that the decision to have a Dayton plan is orthogonal to the beginning of the Great Depression, and so plans should not have been chosen anticipating this event. As for parallel trends, it is not possible to provide pre-trends as there are no data in the pre-period. Even so, the parallel trends assumption is likely to hold. The specification allows for differential levels of fees across plan types. What would be problematic would be if Dayton plans are increasing or decreasing fees over time relativetonon-Daytonplans.However,thereisnoevidencethatDaytonB&Lsweredisproportionately raising or lowering fees. If anything, Dayton plans would be lowering such fees to continue to compete with local banks, meaning any estimate of λ would likely be a lower bound. Thisspecificationisanimperfecttestofliquidityneedsfortworeasons.First,thisisatestofex-postliquidity needs,notex-ante.ThemainhypothesisofthispaperisthatDaytonB&Lsattractedindividualswithhigher liquidity needs, thereby endogenizing the probability of closure. Second, it could be the case that members at Dayton plans simply lost their jobs or sources of income. While this could be seen as a liquidity shock, it could also be interpreted as a net worth shock on the part of members. Taken together, this test is only suggestive of liquidity needs assuming individuals understand the risks ex-ante. However, it is arguably the best test I could perform. The last panel of Table 6 presents the results. The first column shows results using the reported measure for the 149 B&Ls with cash flow data in both 1927 and 1930, controlling for the same variables as in Table 4. The point estimate of 0.792 (SE: 0.24) on the interaction term suggests that there was a rise in fees as a share of total assets by approximately 0.79 percentage points. The second column shows results for the liabilities measure, and the point estimate is largely unchanged at 0.864 (SE: 0.32). This provides evidence that liquidity needs were an important difference between Dayton and non-Dayton B&Ls. 6 Discussion 6.1 The Flightiness Mechanism Howmightclosurehaveoccurredinpractice?First,Daytonplanmemberscouldhavebeenmoreaggressivein requesting withdrawal. Given the rule in California that required institutions to pay out withdrawals within two years, such aggressive demands would result in liquidation, potentially by the commissioner, if demands occurred relatively quickly. Second, Dayton plans, which typically featured guarantee stock and non-voting investment certificates, may have elected to liquidate quicker as the equity value of the institution fell. The equity value may fall if either existing members’ demand for liquidity raised the chances of insolvency, or if the B&L became unattractive to potential future members. Both of these methods are likely to have occurred. Investment certificates, (along with assets as shown in Figure 2) peaked in 1930. However, closure rates only began to spike in 1930, suggesting two distinct waves. 19

Table 7 shows how closures evolved over the Great Depression in California across types of closure. The first wave, in 1929-1930, experienced high numbers of consolidations and transfers. The second wave, while investment certificates were declining after 1930, experienced a higher rate of involuntary closures. Both periods are indicative of flighty members, albeit for different reasons. The first period, through 1930, saw a rush into investment certificates by members seeking safety and liquidity and the assumption that their investments would be easily withdrawable. Dayton plans, which marketed their investment certificates as precisely that, were happy to take the new members. The wave of consolidations and transfers may then represent a desire to reorganize B&Ls to take advantage of the high demand. Specifically, the consolidations and mergers resulted in chains of building and loans operating by oneholdingcompany.Thistransformationlargelyeliminatedthe“localcontactsandlocalsympathies,which wereoriginallyimportantcharacteristicsofbuildingandloanassociations”(BuildingandLoanCommissioner 1931). Flighty members, whose main focus was easy access to funds in the event of economic distress, were a symptom. The period after 1930 featured involuntary withdrawals. The sharp increase in investment certificates in the first wave were the precondition for the next wave after 1930. In this period, flightiness directly determines closure for any of the three potential reasons mentioned earlier in this subsection. Indeed, splitting the sample into closures in 1929-1930 and closures after 1930 shows a strong effect in this latter period for Dayton plans.31 6.2 Relation to Bank Failure Theory This paper provides empirical results that help inform the theoretical literature on bank failure. I find that bank liquidity shocks can be endogenous to the depositor base, which in turn is a function of the types of liabilities issued by the bank. Multiple equilibrium models that feature liquidity shocks, such as the benchmark model of Diamond and Dybvig (1983), typically assume that the probability of a liquidity shock is exogenous or at least that heterogeneity across depositors is orthogonal to the decision to withdraw funds. My results instead suggest an important role for depositor heterogeneity. One method of obtaining endogenous liquidity probabilities is to augment a bank failure model by allowing individuals to receive signals (e.g. Goldstein and Pauzner 2005). These models typically emphasize signals about the health of the bankortheeconomy.Incontrast,myresultssuggestthatsignalsmayalsobeafunctionofthecharacteristics ofdepositors.Alternatively,therearemodelsofbankingpanicswitharisk-aversesetofagents(Caballeroand Simsek 2013) or models studying flight to safety (Caballero and Farhi 2018). My results contribute to this theoretical literature by stressing that heterogeneity across instruments determine whether an institution’s liabilities are held by such risk-averse agents. The choice of how to structure liabilities to take into account asymmetric information about the flightiness of investors also has empirical support in my study. Offering investment contracts that any investor could purchase could be problematic in the event of a bad shock if such contracts attract flighty investors. B&Ls in California essentially engaged in a form of price discrimination across institutions. High-return, high-cost B&Ls attracted investors less likely to force a closure, while low-return, low-cost B&Ls were more likely to close. Whether these closures are efficient is beyond the scope of this paper. 31.AppendixTableC11re-estimatestheclosureregressionrestrictingtheclosurestothoseeitherbeforeorafter1930.Ifind that the effect persists in the second period under both measures, with a weaker estimate in the first period. Note that this exerciseismechanicallydownwardbiasedasitdropsB&Lsthatclosedineachsubsample. 20

Whether these penalties are efficient is related to a separate but related class of models that studies how liquidity mismatch can be a commitment device (e.g. Calomiris and Kahn 1991). That depositors can easily withdrawfundsmayactasacheckonthemanagementpracticesofbanks,especiallyifdepositorsrespondto negative news by switching banks. It is not clear whether withdrawal penalties would reinforce or limit this channel. Because it is more difficult to withdraw, bank managers may be less likely to perform due diligence on new lending. However, if flightiness is negatively correlated with financial sophistication, management may feel pressure to make higher-quality investments lest a larger base of informed/sophisticated depositors leave. 6.3 Conclusion In this paper, I study the role that flightiness plays in causing financial distress. I leverage institutional differences across B&Ls in California during the Great Depression. These institutions offer a unique laboratory to investigate flightiness because their liabilities greatly differed, with one type of plan, Dayton plans, offeringrelativelylowwithdrawalpenaltiesandallowingirregularsavingsplans.Atthesametime,theasset structure across institutions was very similar and focused almost exclusively on mortgage lending. I emphasize three main results. First, Dayton plans had a probability of closure during the Great Depression higher than other plans. Second, Dayton plans were less costly for members to join and participate in compared with other plans. Finally, Dayton plans had lower returns to members compared with non- Dayton plans. Taken together, these three results suggest that the access costs of non-Dayton B&L’s were an important factor in reducing closure rates, likely because it attracted a less flighty member who would be significantly less likely to need liquidity during the Great Depression. These higher withdrawal penalties were justified by offering higher returns to members. The results in this paper do not necessary imply that withdrawal penalties are socially optimal. As shown in this paper, B&Ls with withdrawal penalties also needed to pay out higher returns to members to attract investment in general. The need to pay out higher returns may affect lending rates and reduce demand for loanable funds, with potential spillover effects on local households and businesses. Future research can examine this tradeoff. The evolution of Dayton B&Ls has important implications for financial stability. First, some liquidity characteristics of liabilities can lead to financial instability. During the Great Recession, both money market mutual funds and investment banks such as Lehman Brothers experienced distress (Gertler and Gilchrist 2018). This distress took the form of a bank run in wholesale lending markets where investors could easily withdraw funds. Capital flows are known to be fickle in international finance (e.g. Caballero and Simsek 2020). Central bank digital currencies (CBDCs), which represent a more liquid and low-cost alternative to bank deposits, may be an arguably safer asset available to the private sector. If households have access to CBDCs (so-called retail CBDC’s as described by Boar and Wehrli 2021), this arguably safer asset may lead tofinancialinstabilityinfuturedownturnsiftheretailholdersofstandardbankdepositsare“flighty.” Second, consider the development of financial institutions due to innovation. The evolution of Dayton plans is not unlike the growth of trust companies (an early form of investment bank) at the turn of the 20th century. Noyes (1901) discusses how the rapid growth of trust companies from 1896-1901 was driven by competitionwithcommercialbanks;however,trustcompaniesweremorelightlyregulatedandcouldfinance riskier types with lower reserve requirements. Trusts therefore grew rapidly and were the main source of 21

financial instability during the Panic of 1907, subject to severe deposit withdrawals and contractions of lending (Moen and Tallman 1992) with subsequent effects on real economic activity (Frydman, Hilt, and Zhou2015).MuchlikethedevelopmentoftheDaytonB&Lsliabilitystructure,thefinancialinnovationthat resulted in rapid growth was also a source of fragility. 22

References Alhadeff, David A., and Charlotte P. Alhadeff. 1958. “The Struggle for Commercial Bank Savings.” The Quarterly Journal of Economics 72 (1): 1–22. Allen,Franklin,andDouglasGale.2000.“FinancialContagion.”Journalof Political Economy 108(1):1–33. Beshears, John, James J. Choi, Christopher Harris, David Laibson, Brigitte C. Madrian, and Jung Sakong. 2020. “Which early withdrawal penalty attracts the most deposits to a commitment savings account?” Journal of Public Economics 183:104144. Bikker,JacobA.,andDirkF.Gerritsen.2018.“DeterminantsofInterestRatesonTimeDepositsandSavings Accounts: Macro Factors, Bank Risk, and Account Features.” International Review of Finance 18 (2): 169–216. Bleemer, Zachary. 2016. Role Model Effects of Female STEM Teachers and Doctors on Early 20th Century University Enrollment in California. Technical report. Blickle, Kristian, Markus K. Brunnermeier, and Stephan Luck. 2019. Micro-Evidence From a System-Wide Financial Meltdown: The German Crisis of 1931. Boar, Codruta, and Andreas Wehrli. 2021. Ready, steady, go? - Results of the third BIS survey on central bank digital currency. Bodfish,Morton.1935.“TheDepressionExperienceofSavingsandLoanAssociationsintheUnitedStates.” InTheAmericanSavingsandLoanIndustry,1831-1935,editedbyDavidL.Mason.London:Pickering/ Chatto Limited. Building and Loan Commissioner of the State of California. 1927. Annual Report. . 1929. Annual Report. . 1930. Annual Report. . 1931. Annual Report. . 1933. Annual Report. . 1935. Annual Report. Caballero, Ricardo J, and Emmanuel Farhi. 2018. “The Safety Trap.” Review of Economic Studies 85 (1): 223–274. Caballero, Ricardo J., and Alp Simsek. 2013. “Fire Sales in a Model of Complexity.” Journal of Finance 68 (6): 2549–2587. . 2020. “A Model of Fickle Capital Flows and Retrenchment.” Journal of Political Economy 128 (6): 2288–2328. Calomiris, Charles W., and Charles M. Kahn. 1991. “The Role of Demandable Debt in Structuring Optimal Banking Arrangements.” The American Economic Review 81 (3): 497–513. Calomiris,CharlesW.,andJosephR.Mason.1997.“ContagionandBankFailuresduringtheGreatDepression: The June 1932 Chicago Banking Panic.” American Economic Review 87 (5): 863–83. 23

Carlson, Mark, and Kris James Mitchener. 2009. “Branch Banking as a Device for Discipline: Competition and Bank Survivorship during the Great Depression.” Journal of Political Economy 117 (2): 165–210. Carter,SusanB,ScottSigmundGartner,MichaelRHaines,AlanLOlmstead,RichardSutch,GavinWright, et al. 2006. Historical statistics of the United States: Millennial edition. Vol. 3. Cambridge University Press Cambridge. Clark, Horace Frisby, and Frank A. Chase. 1927. Elements of the Modern Building and Loan Associations. New York: The MacMillan Company. Courtemanche,Charles,andKennethSnowden.2011.“RepairingaMortgageCrisis:HOLCLendingandIts Impact on Local Housing Markets.” The Journal of Economic History 71 (2): 307–337. Dawson, Kennett B., James J. McBride, James B. Utt, and Ralph Louis Welsh. 1935. Report of Select Committee of the Assembly for the Purpose of Investigating the Building and Loan Situation in the State of California. DepartmentofSavingsandLoanRecords.RecordsoftheLosAngelesOffice.F3739.CaliforniaStateArchives. Diamond,DouglasW.,andPhilipH.Dybvig.1983.“BankRuns,DepositInsurance,andLiquidity.”Journal of Political Economy 91 (3): 401–19. Diamond, Douglas W., and Raghuram G. Rajan. 2001. “Liquidity Risk, Liquidity Creation, and Financial Fragility: A Theory of Banking.” Journal of Political Economy 109 (2): 287–327. Doti, Lynne Pierson, and Larry Schweikart. 1991. Banking in the American west. Norman, OK: University of Oklahoma Press, December. Drapeau, Louis C. 1935. Problems Confronting Associations. Field, Alexander J. 2014. “The Interwar Housing Cycle in the Light of 2001-2012: A Comparative Historical Perspective.” In Housing and Mortgage Markets in Historical Perspective, 39–80. NBER Chapters. National Bureau of Economic Research, Inc. Fishback, Price, Sebasti´an Fleitas, Jonathan Rose, and Kenneth Snowden. 2018. Collateral Damage: The Impact of Foreclosures on New Home Mortgage Lending in the 1930s. Working Paper, Working Paper Series 25246. National Bureau of Economic Research. Fishback, Price, and Shawn Kantor. 2018. New Deal Studies. Fishback,PriceV.,WilliamC.Horrace,andShawnKantor.2005.“DidNewDealGrantProgramsStimulate LocalEconomies?AStudyofFederalGrantsandRetailSalesduringtheGreatDepression.”TheJournal of Economic History 65 (1): 36–71. Fleitas,Sebastian,PriceFishback,andKennethSnowden.2018.“Economiccrisisandthedemiseofapopular contractual form: Building & Loans in the 1930s.” Journal of Financial Intermediation 36:28–44. Frydman, Carola, Eric Hilt, and Lily Y. Zhou. 2015. “Economic Effects of Runs on Early “Shadow Banks”: Trust Companies and the Impact of the Panic of 1907.” Journal of Political Economy 123 (4): 902–940. Gertler, Mark, and Simon Gilchrist. 2018. “What Happened: Financial Factors in the Great Recession.” Journal of Economic Perspectives 32 (3): 3–30. 24

Goldstein, Itay, and Ady Pauzner. 2005. “Demand–Deposit Contracts and the Probability of Bank Runs.” Journal of Finance 60 (3): 1293–1327. Haines, Michael, Price Fishback, and Paul Rhode. 2018. United States Agriculture Data, 1840 - 2012. Haveman, Heather A., and Hayagreeva Rao. 1997. “Structuring a Theory of Moral Sentiments: Institutional and Organizational Coevolution in the Early Thrift Industry.” American Journal of Sociology 102 (6): 1606–1651. Haveman, Heather A., Hayagreeva Rao, and Srikanth Paruchuri. 2007. “The Winds of Change: The Progressive Movement and the Bureaucratization of Thrift.” American Sociological Review 72 (1): 117– 142. Inter-university Consortium for Political and Social Research. 1999. United States Historical Election Returns, 1824 - 1968. Ivashina, Victoria, and David Scharfstein. 2010. “Bank lending during the financial crisis of 2008.” The 2007-8financialcrisis:Lessonsfromcorporatefinance,Journal of Financial Economics 97(3):319–338. Ivashina, Victoria, David S. Scharfstein, and Jeremy C. Stein. 2015. “Dollar Funding and the Lending Behavior of Global Banks.” Quarterly Journal of Economics 130 (3): 1241–1281. Iyer, Rajkamal, and Manju Puri. 2012. “Understanding Bank Runs: The Importance of Depositor-Bank Relationships and Networks.” American Economic Review 102 (4): 1414–45. Iyer, Rajkamal, Manju Puri, and Nicholas Ryan. 2016. “A Tale of Two Runs: Depositor Responses to Bank Solvency Risk.” The Journal of Finance 71 (6): 2687–2726. Mitchener, Kris James, and Gary Richardson. 2013. “Shadowy Banks and Financial Contagion during the Great Depression: A Retrospective on Friedman and Schwartz.” American Economic Review 103 (3): 73–78. Moen, Jon, and Ellis W. Tallman. 1992. “The Bank Panic of 1907: The Role of Trust Companies.” The Journal of Economic History 52 (3): 611–630. Noyes,AlexanderD.1901.“TheTrustCompanies:IsThereDangerintheSystem?”Political Science Quarterly 16 (2): 248–261. O’Grada, Cormac, and Eugene White. 2003. “The Panics of 1854 and 1857: A View from the Emigrant Industrial Savings Bank.” The Journal of Economic History 63 (1): 213–240. Peterson, Frederick. 1935. Minority Report. Pieplow, William L. 1931. “Century Lessons Commemorating One Hundred Years of Building and Loan Associations.” In The American Savings and Loan Industry, 1831-1935, edited by David L. Mason. London: Pickering / Chatto Limited. Postel-Vinay,Natacha.2016.“WhatCausedChicagoBankFailuresintheGreatDepression?ALookatthe 1920s.” The Journal of Economic History (New York, USA) 76 (2): 478–519. Price, David A., and John R. Walter. 2019. “It’s a Wonderful Loan: A Short History of Building and Loan Associations.” Richmond Fed Economic Brief, no. January. 25

Quincy, Sarah. 2019. Loans for the Little Fellow: Credit, Crisis, and Recovery in the Great Depression. Richardson, Gary, and William Troost. 2009. “Monetary Intervention Mitigated Banking Panics during the Great Depression: Quasi-Experimental Evidence from a Federal Reserve District Border, 1929-1933.” Journal of Political Economy 117 (6): 1031–1073. Riegel, Robert, and J. Russell Doubman. 1927. The Building and Loan Association. New York: John Wiley & Sons, Inc. Rose, Jonathan D. 2014. “The Prolonged Resolution of Troubled Real Estate Lenders During the 1930s.” Chap.7inHousingandMortgageMarketsinHistoricalPerspective,editedbyEugeneN.WhiteKenneth Snowden and Price Fishback, 245–284. Chicago: University of Chicago Press. Rose,JonathanD.,andKennethA.Snowden.2013.“TheNewDealandtheoriginsofthemodernAmerican real estate loan contract.” New Views of Roosevelt’s New Deal, Explorations in Economic History 50 (4): 548–566. Ruggles, Stven, Sarah Flood, Sophia Foster, Ronald Goeken, Jose Pacas, Megan Schouweiler, and Matthew Sobek. 2021. IPUMS USA: Version 11.0 [dataset]. Shin, Hyun Song. 2009. “Reflections on Northern Rock: The Bank Run That Heralded the Global Financial Crisis.” Journal of Economic Perspectives 23 (1): 101–19. Smith, Warren L. 1959. “Financial Intermediaries and Monetary Controls*.” The Quarterly Journal of Economics 73, no. 4 (November): 533–553. Snowden,KennethA.1997.“BuildingandloanassociationsintheU.S.,1880–1893:theoriginsoflocalization in the residential mortgage market.” Research in Economics 51 (3): 227–250. . 2003. “The Transition from Building and Loan to Savings and Loan.” Finance, Intermediaries, and Economic Development (January). Stanford Law Review. 1950. “Building and Loan Liquidation, California Style.” Stanford Law Review 3 (1): 60–87. SuperintendentofBanksoftheStateofCalifornia.1935.Twenty-First Annual Report of the Superintendent of Banks of the State of California : 1930. White, Eugene N. 2014. “Lessons from the Great American Real Estate Boom and Bust of the 1920s.” In Housing and Mortgage Markets in Historical Perspective, 115–158. NBER Chapters. National Bureau of Economic Research, Inc. 26

Figure 1: Rise and Fall of B&Ls during the Interwar Period Value(inmillions,ontheleft)andshareoftotalinstitutionalrealestatelending(inpercentagepoints,ontheright)bybuilding andloanassociationsintheUnitedStates.Source:Carteretal.(2006) 27

Figure 2: California B&Ls in the Great Depression Total assets (in millions, on the left) and total number (on the right) of California B&Ls over the period 1920-1935. Source: BuildingandLoanCommissioner(VariousYears) 28

Figure 3: County Distribution of California Building and Loan Associations (1927) TheleftpanelmapsthetotalnumberofCaliforniaB&L’sactivein1927.TherightpanelmapstheshareofDaytonplans.Both mapsareatthecountylevel.Source:BuildingandLoanCommissioner(1927) 29

Figure 4: Advertisements (a) Non-Dayton Plan (b) Dayton Plans The left panel shows an advertisement for a non-Dayton plan. The right panel shows an advertisement for a Dayton plan. Source:non-DaytonPlanAdvertisement:SanBernardinoSun,Volume57,Number31,Page8(1October1925);DaytonPlan Advertisement: Healdsburg Tribune, Number 54, Page 4 (9 January 1928); Accessed via UCR California Digital Newspaper Collection. 30

Figure 5: Investment Certificates Share of Liabilities This figure plots the ratio of investment certificates to the sum of investment certificates and withdrawable shares. Source: BuildingandLoanCommissioner(1929,1927) 31

Figure 6: Balance Sheets (a) Balance Sheet: Dayton Plan (b) Balance Sheet: Non-Dayton Plan Left panelshows asample B&L balance sheetfor a Dayton plan (as indicatedat the bottom of the figure). Right panelshows asample B&L balance sheetfor a non-Dayton plan.Source:BuildingandLoanCommissioner(1927) 32

Table 1: Distribution of Plans in the United States in 1923 Plan Number Percent Terminating 96 0.92% Serial/Permanent 9,121 87.04% Dayton 1,186 11.32% Other 76 0.73% Total 10,479 100% Serial/Permanentplanscalculatedasthesumof“AllPermanent”and“RegularPermanent.”Percentsharesmaynotaddup to100%duetorounding.Source:Page61ofClarkandChase(1927),author’scalculations. 33

Table 2: Summary Statistics Variable Obs Mean Std. Dev. Min Max P50 Dayton (Reported) 164 .774 .419 0 1 1 Members (Thousands) 164 1.446 2.053 .014 9.263 .583 Closure Dummy 205 .371 .484 0 1 0 Age (Years since Incorporation) 205 15.122 15.849 0 55 7 Total Assets (Millions) 205 1.909 3.835 .033 30.892 .593 Cash (% Assets) 205 4.627 4.466 0 25.867 3.395 Real Estate Owned (% Assets) 205 1.177 2.204 0 13.661 0 Shares (% Assets) 205 22.448 34.013 0 97.928 1.376 Investment Certificates (% Assets) 205 56.82 31.599 0 91.579 68.916 Investment Securities Share of Member Funds 205 .74 .382 0 1 .982 Number of Banks in City 205 8.059 9.441 0 25 3 City Population (Thousands) 205 349.195 487.379 .726 1238.048 52.513 “Closure Dummy” is a dummy variable equal to one if a building and loan association was absorbed, closed, consolidated, or transferred. “Investment Securities Share of Member Funds” calculated as investment securities divided by the sum of investmentsecuritiesandwithdrawableshares.“Age”calculatedasnumberofyearsopenasof1929.“Dayton(Reported)”and “Members(Thousands)”usedatafromthe1927annualreportsandsodropB&Lsformedin1927-1929.“InvestmentSecurities ShareofMemberFunds”istheratioofinvestmentcertificatestothesumofinvestmentcertificatesandwithdrawableshares, asdescribedinthetext.Source:BuildingandLoanCommissioner(VariousYears),SuperintendentofBanks(1935),Bleemer (2016),Fishback,Horrace,andKantor(2005),Haines,Fishback,andRhode(2018),FishbackandKantor(2018),Carlsonand Mitchener(2009) 34

Table 3: Dayton and Non-Dayton (Reported) (1) (2) (3) Variable Permanent Dayton Diff Closure Dummy 0.16 0.39 0.22** (0.37) (0.49) (0.09) Members (Thousands) 0.76 1.64 0.88** (1.22) (2.20) (0.38) Age (Years since Incorporation) 33.86 13.96 -19.90*** (11.88) (14.28) (2.57) Total Assets (Millions) 1.17 2.60 1.43* (2.01) (4.60) (0.78) Cash (% Assets) 2.86 4.47 1.61** (2.47) (3.83) (0.67) Real Estate Owned (% Assets) 0.91 1.56 0.65 (1.49) (2.56) (0.44) Shares (% Assets) 60.43 17.48 -42.95*** (33.09) (30.45) (5.80) Investment Certificates (% Liabilities) 27.63 64.98 37.35*** (31.77) (28.59) (5.48) Investment Securities Share of Member Funds 0.32 0.80 0.48*** (0.36) (0.34) (0.06) Number of Banks in City 4.65 8.76 4.11** (6.20) (9.62) (1.68) City Population (Thousands) 177.24 382.41 205.17** (303.59) (499.75) (86.57) Observations 37 127 205 “Closure Dummy” is a dummy variable equal to one if a building and loan Association was absorbed, closed, consolidated, or transferred. “Investment Securities Share of Member Funds” calculated as investment securities divided by the sum of investmentsecuritiesandwithdrawableshares.“Age”calculatedasnumberofyearsopenasof1929.“Dayton(Reported)”and “Members(Thousands)”usedatafromthe1927annualreportsandsodropB&Lsformedin1927-1929.“InvestmentSecurities ShareofMemberFunds”istheratioofinvestmentcertificatestothesumofinvestmentcertificatesandwithdrawableshares, asdescribedinthetext.Source:BuildingandLoanCommissioner,(VariousYears),SuperintendentofBanks(1935),Bleemer (2016),Fishback,Horrace,andKantor(2005),Haines,Fishback,andRhode(2018),FishbackandKantor(2018),Carlsonand Mitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 35

Table 4: Closure Rates Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.224∗∗∗ 0.238∗∗∗ 0.253∗∗ 0.195∗ (0.0749) (0.0803) (0.109) (0.109) Dayton (Liabilities) 0.225∗∗∗ 0.218∗∗∗ 0.161∗∗ 0.159∗ (0.0661) (0.0679) (0.0811) (0.0828) N 164 164 164 164 205 205 205 205 R-Squared 0.04 0.04 0.07 0.13 0.05 0.05 0.07 0.16 B&L Controls N Y Y Y N Y Y Y Age FE N N Y Y N N Y Y Local Controls N N N Y N N N Y ThistablepresentsresultsforthecoefficientβfromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closureiis adummyvariableequaltooneifBuildingandLoanAssociationiwasabsorbed,closed,consolidated,ortransferred.“Dayton (Reported)” is the plan type as reported in the 1927 annual reports, while “Dayton (Liabilities)” is a dummy equal to one if theassociationhasabove-medianinvestmentcertificatesasashareofliabilities,asdescribedinthetext.B&Lcontrolsinclude log assets and cash percentage. Age controls include age bin fixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. Source: Building and Loan Commissioner (Various Years), Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.HeteroskedasticityrobuststandardErrorsinparentheses. 36

Table 5: Closure Rates: Alternative Specifications Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.253∗∗ 0.297∗ 0.299∗ 0.446∗ (0.109) (0.174) (0.175) (0.227) Dayton (Liabilities) 0.161∗∗ 0.306∗∗∗ 0.304∗∗∗ 0.482∗∗ (0.0811) (0.114) (0.115) (0.201) N 164 117 114 53 205 150 145 67 R-Squared 0.07 0.33 0.32 0.62 0.07 0.40 0.39 0.56 B&L Controls Y Y Y Y Y Y Y Y Age FE Y Y Y Y Y Y Y Y City FE N Y Y Y N Y Y Y Sample Full Full Both No SF/LA Full Full Both No SF/LA Thistablepresentsresultsforthecoefficientβ fromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closurei isadummyvariableequaltooneifbuildingandloanassociationiwasabsorbed,closed,consolidated,ortransferred.“Dayton (Reported)” is the plan type as reported in the 1927 annual reports, while “Dayton (Liabilities)” is a dummy equal to one if theassociationhasabove-medianinvestmentcertificatesasashareofliabilities,asdescribedinthetext.B&Lcontrolsinclude log assets and cash percentage. Age controls include age bin fixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. The sample denoted “Full” is the benchmark sample of 164 B&L’s for the reported measure and 205 B&Ls for the liabilities measure. The sample denoted “Both” includes only counties whichcontainatleastoneofeachtypeofB&L(Daytonandnon-Dayton).“NoSF/LA”dropsB&LslocatedinSanFrancisco orLosAngeles.Source:BuildingandLoanCommissioner(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 37

Table 6: Evidence on the Role for Flightiness in Predicting Closure (a) Withdrawal Fees and the Costs of Membership by Type Withdrawal Penalty Dues Shares per Member Costs Dayton (Reported) -0.491∗∗∗ -0.0970∗ -0.651∗∗ -5.939∗ (0.0841) (0.0551) (0.326) (3.425) N 164 164 164 164 R-Squared 0.26 0.42 0.24 0.25 B&L Controls Y Y Y Y Age Controls Y Y Y Y (b) Archival Evidence: Member Returns and Loan Characteristics Return Borrower Share Lending Rate Log Avg Loan Size Dayton (Reported) -0.232∗∗ -0.0975∗ 0.306 -0.0102 (0.0975) (0.0585) (0.273) (0.110) N 97 97 97 97 R-Squared 0.13 0.19 0.08 0.07 B&L Controls Y Y Y Y Age Controls Y Y Y Y (c) Withdrawal Fees Difference in Difference Fees Ratio Dayton (Reported) X 1930 0.792∗∗∗ (0.244) N 298 R-Squared 0.82 B&L FE Y Year FE Y Thetopandmiddlepanelsshowresultsfromestimatingtheequationyi=αi+βDAYTONi+ΓXi+εi.Theoutcomesforthe toppanelinclude:“WithdrawalPenalty,”adummyequaltooneifaB&Lhaspenaltiesforwithdrawingfunds;“Dues”denotes thecostofduesin1927;“SharesperMember”istheratiooftotalsharestototalmembers;“Costs”istheproductof“Dues”and “SharesperMember,”ortotalcostspermember.Theoutcomesforthemiddlepanelinclude“Return,”whichistheweighted averageofreturnsforinvestmentcertificatesandwithdrawableshares,wheretheweightsaregivenbytherelativeproportionof each; “Borrower Share,”which denotesthe share ofmembers that areborrowing, “Lending Rate,”whichdenotes the average rateonmortgageloans,and“LogAvgLoanSize,”orthelogoftheratiooftheamountofloanstothenumberofloans.B&L controlsincludelogassetsandcashpercentage,andagecontrolsincludeagebinfixedeffects.Agecontrolsforthispanel,which usesarchivaldata,includeadummyequaltooneiftheassociationwasincorporatedafter1920duetothelimitedsamplesize. Thebottompanelestimatesdifferences-in-differencesspecificationsoftheformYit =αt+βi+γ(DAYTONi×1(t=1930)). Source:BuildingandLoanCommissioneroftheStateofCalifornia(1927),Department of Savings and Loan Records ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 38

Table 7: Closure Type over Time Closure Year Consolidations/Transfers Involuntary Other Total 1929 14 1 4 19 1930 10 3 9 22 1931 4 6 4 14 1932 0 5 1 6 1933 1 3 0 4 1934 0 0 2 2 1935 1 5 3 9 Total 30 23 23 76 Totalclosuresbyyearandtype.Closuretypeistakenfromthe1935annualreport.Involuntaryincludesanyinvoluntaryclosure thatresortsinliquidationorreorganizationbythecommissioner.Otherincludesabsorptionorvoluntaryliquidation.Source: BuildingandLoanCommissioneroftheStateofCalifornia(1935) 39

A Data Sources Table A1 displays the sources for each variable used in this paper. 40

secruoS ataD :1A elbaT yhpargoeG selbairaV ecruoS L&B serahS ,seulaV lawardhtiW ,seuD ,srebmeM )7291( renoissimmoC naoL dna gnidliuB L&B raeY noitaroprocnI ,teehS ecnalaB )9291( renoissimmoC naoL dna gnidliuB L&B emocnI seeF ,stnemyapeR naoL ,stpieceR latoT )0391 ,7291( renoissimmoC naoL dna gnidliuB L&B sedoC erusolC )5391( renoissimmoC naoL dna gnidliuB L&B scitsiretcarahC reworroB ,snruteR ,setaR gnidneL sdroceR naoL dna sgnivaS fo tnemtrapeD ytiC noitalupoP )6102( remeelB ytiC etaR eruliaF knaB ,sknaB fo rebmuN )9002( renehctiM dna noslraC ytnuoC )8291-6981( erahS etoV citarcomeD ,erahS .poP nabrU ,selaS liateR )5002( rotnaK dna ,ecarroH ,kcabhsiF ytnuoC atipac rep snaoL CLOH )8102( rotnaK dna kcabhsiF ytnuoC seulaV dnalmraF egarevA )8102( edohR dna ,kcabhsiF ,seniaH ytnuoC serahS gnitoV nacilbupeR/evissergorP 2191 )9991( hcraeseR laicoS dna lacitiloP rof muitrosnoC ytisrevinu-retnI ASU leveL ecirP ,gnidneL etatsE laeR )6002( .la te retraC 41

B Additional Information on Historical Context and Final Sample The first annual report in 1893 listed 146 active B&L’s. The oldest B&L in the state was the Germania Building and Loan Association of Sacramento, incorporated in 1872. Prior to 1893, some B&L’s were incorporated under various legislative environments in California. In 1893, all B&L’s were consolidated under the same California law. This law also created the state’s Building and Loan Commission, which started the series of annual reports used extensively in this paper. There are 204 B&L’s listed in the 1927 annual report. I drop the two foreign B&L’s that are headquartered in Utah. Of the remaining 202, 11 institutions are new and do not have sufficient balance sheet information, and 20 other close in 1927 and 1928. Finally, 7 institutions federalize.32 This brings the total sample down to 164 institutions. Using the new institutions between 1927 and 1929, the alternative definition of Dayton plan increases the sample size to 219, of which 14 are federalized. Some B&L’s listed as their origin cities that did not have population data. For these, the cities were either mergedintoalargercities(andwerethus“neighborhoods”)orthecitywasunincorporated.Thosecitiesthat were merged into a larger city include San Pedro, Van Nuys, Wilmington, Sherman, and North Hollywood, which became part of Los Angeles, and La Jolla, which became part of San Diego. I therefore assign those larger city populations. Newcastle and Cucamonga were unincorporated, so I assume these are Rocklin and Ontario, respectively. Table B1 shows the full summary statistics table, including those used in the robustness checks in Section C. TableB2showsthebalancetablefortheliabilitiesmeasureofDaytonplan.Thedifferencesbetweenthetwo types are similar to the results in the main text. Table B3 shows the balance table for the variables at the local level, including real retail sales changes, changes in real average farmland values, and bank failure rates. The distribution of closure codes by plan type can be seen in Table B4. The timing of closures can be seen in Table B5. Figure B1 plots the raw histogram of fees paid. Figure B2 shows the histogram of institution ages by plan type. Table B6 shows which observable variables most predict Dayton choice as of 1930. The results show that Age is an important predictor, which motivates my inclusion of age controls in the benchmark specification. Log population also matters for the reported plan type. Figure B3 shows an example Balance Sheet from the Archives. Table B7 shows a balance table to compare the available associations in the archival sample to the full sample. 32.14institutionsactuallyfederalize,but7ofthemincorporatebetween1927and1929. 42

Figure B1: Fees Share Figureshowsdistributionoffeesin1927and1930asshareofassetsin1927forDaytonandNon-Daytonplans.Source:Building andLoanCommissioner,(VariousYears) 43

Figure B2: Age Distribution by Type This figure shows the histogram of Building and Loan associations by plan type. Source: Building and Loan Commissioner (1935,1927) 44

Figure B3: Balance Sheet: Raw Archives Sample page from archival balance sheet information. Source: Inventory of the Dept. of Savings and Loan Records. Records oftheLosAngelesOffice.F3739:425-450.CaliforniaStateArchives 45

Table B1: Summary Statistics Variable Obs Mean Std. Dev. Min Max P50 Dayton (Reported) 164 .774 .419 0 1 1 Closure Dummy 205 .371 .484 0 1 0 Members (Thousands) 164 1.446 2.053 .014 9.263 .583 Age (Years since Incorporation) 205 15.122 15.849 0 55 7 Total Assets (Millions) 205 1.909 3.835 .033 30.892 .593 Cash (% Assets) 205 4.627 4.466 0 25.867 3.395 Real Estate Owned (% Assets) 205 1.177 2.204 0 13.661 0 Shares (% Assets) 205 22.448 34.013 0 97.928 1.376 Investment Certificates (% Assets) 205 56.82 31.599 0 91.579 68.916 Investment Securities Share of Member Funds 205 .74 .382 0 1 .982 Concentration Index (1929) 205 35.35 31.708 0 99.894 22.412 1930 Loan Repayments 180 1.139 7.897 0 72.811 0 Number of Banks in City 205 8.059 9.441 0 25 3 City Population (Thousands) 205 349.195 487.379 .726 1238.048 52.513 “Closure Dummy” is a dummy variable equal to one if a building and loan association was absorbed, closed, consolidated, or transferred. “Investment Securities Share of Member Funds” calculated as investment securities divided by the sum of investmentsecuritiesandwithdrawableshares.“Age”calculatedasnumberofyearsopenasof1929.“Dayton(Reported)”and “Members(Thousands)”usedatafromthe1927annualreportsandsodropB&Lsformedin1927-1929.Source:Buildingand LoanCommissioner(VariousYears),SuperintendentofBanks(1935),Bleemer(2016),Fishback,Horrace,andKantor(2005), Haines,Fishback,andRhode(2018),FishbackandKantor(2018),CarlsonandMitchener(2009)Inter-universityConsortium forPoliticalandSocialResearch(1999) 46

Table B2: Balance Table - Liabilities Measure (1) (2) (3) Variable Permanent Dayton Diff Closure 0.22 0.45 0.22*** (0.42) (0.50) (0.07) Age 28.12 8.67 -19.45*** (16.06) (11.10) (1.92) Members (Thousands) 1.17 1.63 0.46 (1.77) (2.21) (0.33) Total Assets (Millions) 1.53 2.10 0.56 (2.29) (4.40) (0.57) Secur. Share of Liabs 0.26 0.98 0.71*** (0.31) (0.04) (0.03) Shares (% Liabilities) 63.89 1.88 -62.02*** (29.77) (3.69) (2.58) Cash (% Assets) 3.52 5.18 1.66** (3.95) (4.62) (0.65) Real Estate Owned (% Assets) 1.36 1.09 -0.27 (2.52) (2.03) (0.33) 1930 Loan Repayments 1.22 1.10 -0.12 (9.11) (7.18) (1.23) Banks 25.18 39.69 14.51*** (31.32) (36.08) (5.13) City Population (Thousands) 294.22 376.48 82.26 (446.88) (505.63) (72.25) Observations 68 137 205 Agecalculatedasnumberofyearsopenasof1927.Closureisadummyvariableequalto1ifaBuildingandLoanAssociation wereabsorbed,closed,consolidated,ortransferred.Source:BuildingandLoanCommissioner,(VariousYears),Superintendent of Banks (1935), Bleemer (2016), Fishback, Horrace, and Kantor (2005), Haines, Fishback, and Rhode (2018), Fishback and Kantor(2018),CarlsonandMitchener(2009). ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 47

Table B3: Local Balance Tables (a) Reported Measure (1) (2) (3) Variable Non-Dayton Dayton Diff Real Retail Sales (1929-1933) -0.45 -0.45 0.00 (0.11) (0.08) (0.02) Avg. Farmland Value (1920-1925) -0.09 -0.03 0.06 (0.23) (0.20) (0.04) Avg. Farmland Value (1925-1935) 0.32 0.15 -0.16** (0.42) (0.33) (0.07) Bank Failure Rate 0.12 0.17 0.05* (0.14) (0.14) (0.03) Observations 37 127 205 (b) Liabilities Measure (1) (2) (3) Variable Non-Dayton Dayton Diff Real Retail Sales (1929-1933) -0.45 -0.45 0.00 (0.10) (0.09) (0.01) Avg. Farmland Value (1920-1925) -0.08 0.00 0.08*** (0.21) (0.19) (0.03) Avg. Farmland Value (1925-1935) 0.23 0.15 -0.08 (0.37) (0.31) (0.05) Bank Failure Rate 0.15 0.18 0.03 (0.14) (0.14) (0.02) Observations 68 137 205 Agecalculatedasnumberofyearsopenasof1927.ClosureisadummyvariableequaltooneifaBuildingandLoanAssociation were absorbed, closed, consolidated, or transferred. Source: Building and Loan Commissioner, (Various Years), Fishback, Horrace,andKantor(2005),Haines,Fishback,andRhode(2018),FishbackandKantor(2018),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 48

Table B4: Distribution of Closures by Closure Code Reported Liabilities Closure Code Dayton Non-Dayton Dayton Non-Dayton Absorbed 10 2 15 3 Removed 3 0 3 0 Consolidated 6 2 9 1 Transferred 24 3 26 5 Relocated 1 0 1 0 Federalized 4 1 11 3 Converted 1 0 1 0 In Liquidation (Commissioner) 16 0 20 2 In Liquidation (Receiver) 1 0 2 0 Liquidated (Commissioner) 0 1 0 1 Liquidated (Voluntarily) 3 1 5 1 Liquidated (Receiver) 0 0 1 0 Source:BuildingandLoanCommissioneroftheStateofCalifornia(1927,1935) 49

Table B5: Distribution of Closures by Year Reported Liabilities Year Dayton Non-Dayton Dayton Non-Dayton 1929 14 3 14 5 1930 9 2 20 2 1931 12 0 12 2 1932 3 0 5 1 1933 2 1 3 1 1934 2 0 2 0 1935 7 0 8 1 ClosuretimingforB&L’sinthesample.Thedifferentialnumberofclosuresinthetwomeasuresisdueto17closuresbyB&L’s formedin1927and1928.Source:BuildingandLoanCommissioneroftheStateofCalifornia(1927,1935) 50

Table B6: Determinants of Dayton Plan Dayton Dayton Dayton Dayton Dayton Dayton Age -0.0131∗∗∗ -0.0524∗∗∗ -0.0947∗∗∗ -0.0170∗∗∗ -0.0515∗∗∗ -0.0855∗∗∗ (0.00188) (0.00939) (0.0184) (0.00184) (0.00715) (0.0126) Log Population (1930) 0.0707∗∗ 0.356∗∗ 0.750∗∗ 0.0513 0.193 0.349 (0.0345) (0.157) (0.300) (0.0350) (0.132) (0.227) Log Banks (1930) -0.0639 -0.155 -0.428 -0.0552 -0.213 -0.349 (0.0538) (0.229) (0.422) (0.0533) (0.204) (0.349) Percent Urban 0.000407 -0.00578 -0.0143 0.000334 0.000358 -0.000312 (0.00221) (0.0102) (0.0183) (0.00214) (0.00807) (0.0140) Mean Repub Share 1896-1928 0.00603 0.0137 0.0118 0.00312 0.0110 0.0139 (0.0109) (0.0480) (0.0821) (0.0111) (0.0432) (0.0716) Avg. Farmland Value 1920-1925 0.100 0.379 0.811 0.160 0.542 1.002 (0.227) (1.050) (1.795) (0.227) (0.856) (1.480) Bank Failure Rate 0.0776 0.493 1.080 -0.136 -0.632 -0.924 (0.221) (0.993) (1.744) (0.214) (0.843) (1.447) Constant 0.126 -1.650 -3.168 0.370 -0.667 -1.212 (0.474) (2.016) (3.382) (0.497) (1.894) (3.116) N 164 164 164 205 205 205 R-Squared 0.31 0.35 Dayton Measure Reported Reported Reported Liabilities Liabilities Liabilities Estimator OLS Probit Logit OLS Probit Logit ThefirstcolumnestimatesviaOLSequationsoftheformDaytoni=α+βAgei+ΓX a(i) +εi.Thesecondandthirdcolumns use probit and logit respectively. Source: Building and Loan Commissioner of the State of California (1927,1935), Bleemer (2016),SuperintendentofBanks(1935),Haines,Fishback,andRhode(2018),Fishback,Horrace,andKantor(2005) 51

Table B7: Balance Table for Micro Sample (1) (2) (3) Variable Missing Micro Data Diff Dayton (Reported) 0.79 0.74 -0.05 (0.41) (0.44) (0.07) Age 14.92 14.72 -0.21 (15.96) (15.67) (2.14) Members (Thousands) 1.63 1.23 -0.39 (2.14) (1.93) (0.31) Total Assets (Millions) 2.17 1.50 -0.67 (4.78) (2.42) (0.50) Secur. Share of Liabs 0.76 0.73 -0.03 (0.39) (0.38) (0.05) Shares (% Liabilities) 21.43 23.25 1.82 (35.18) (33.36) (4.63) Cash (% Assets) 4.60 4.52 -0.08 (4.44) (4.37) (0.60) Real Estate Owned (% Assets) 1.19 1.12 -0.06 (2.02) (2.32) (0.30) 1930 Loan Repayments 2.51 0.08 -2.43** (11.87) (0.70) (1.11) Banks 37.01 33.85 -3.16 (35.52) (34.79) (4.76) City Population (Thousands) 362.28 346.85 -15.44 (490.64) (486.92) (66.16) Observations 103 116 219 Agecalculatedasnumberofyearsopenasof1927.ClosureisadummyvariableequaltooneifaBuildingandLoanAssociation were absorbed, closed, consolidated, or transferred. The difference in samples comes from the inclusion of Federalized B&Ls fortheexercisesusingthemicrosample.Source:BuildingandLoanCommissioner,(VariousYears),SuperintendentofBanks (1935), Bleemer (2016), Fishback, Horrace, and Kantor (2005), Haines, Fishback, and Rhode (2018), Fishback and Kantor (2018),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 52

C Robustness Specifications Coefficient Stability I first show the sensitivity of the main coefficient estimate of β in Equation 1 to the inclusion of additional controls. To frame the discussion, Figure C1 plots the estimate of the coefficient β, corresponding to the marginal effect on closure rates of the self-reported Dayton measure, changes when adding different controls. The first row of the figure labeled “Benchmark” presents the coefficient estimate and 90% standard error bars from the results in Table 4. The results are not sensitive to balance sheet measures of borrower quality. If Dayton plans borrowers were more ex-ante likely to default in general, then the differential closure rates I identify may simply be due to the impairment of assets. Real estate owned shares is a useful proxy for default risk. As emphasized by Fleitas, Fishback, and Snowden (2018), this asset includes foreclosed property taken on by the B&L. The point estimate does not change much when including this estimate, and the estimate is now significant at the 90% level. As I use 1929 measures, this control does not capture whether or not borrowers were more likely to default during the Great Depression. Rather, this measure captures a loan quality measure of whether, in tranquil times, borrowers were more likely to default. I can instead use 1930 loan receipts to capture repayments during the Great Depression as a proxy for borrower quality and the maturity of the loan portfolio. I construct this measure using principal repayments, rather than interest repayments, to better capture this potential latter effect. Note that using this variable as a control drops B&L’s that closed prior to submitting a 1930 annual report, so the estimate of β is downward biased. Still, the point estimate is little changed. Finally, I show the result is robust to including the average loan to value ratio.33 InextfocusonmeasuresthataccountfortheownershipstructureoftheB&L.Iincludetheratioofguarantee stocktothesumofinvestmentcertificatesandshares.Thebenefitsoftheguarantee-stockplanisthatB&L’s couldattractfundsquickerandbeginlendingoperationsearlierClarkandChase1927.Oneconcernwouldbe that of guarantee stock acted as a form of insurance. Similar to deposit insurance, higher levels of guarantee stock could signal to members that the institution is safe, and thus are less likely to close. The fourth row of Figure C1 shows this is not the case by including the share of guarantee stock as a control. Another concern relates to how B&L’s could function in a zombie status as withdrawals were not paid out immediately. These “zombie” B&L’s were common across the country (Snowden 2003). Fleitas, Fishback, and Snowden (2018) discuss how B&Ls in New Jersey could close with a 2/3 majority vote by shareholders and stockholders. The regulations on closure was similar in California. I construct a “concentration index”, whichisthesumofwithdrawablesharesandguaranteestockasashareofassets.Thismeasurecaptureshow much the B&L relied on voting members. Including this measure does not affect the point estimate. An additional concern may be that the mass of people at Dayton plans was higher, all else equal. The benchmark specification includes log total assets as a measure of size due to the number of members being unavailable in the 1929 annual reports. This measure of size accounts for a measure of scale that takes into account how crowded the association is. Haveman and Rao (1997) discuss how institutions had to become more efficient to handle growing numbers of members because relationship lending would be more difficult. That the coefficient is little changed suggests that the efficiency across plans is relatively similar. Finally, I explore include additional local controls. While I have already shown the results are robust to 33.IcalculatetheaverageloantovalueratiobyfirstcalculatingtheaverageloansizeforeachB&L.Theaverageloansizeis given by total lending in 1927 relative to total members in 1927 (since the 1929 statements do not include total members). I thendividethisbytheaveragehousevalueforthecity,takenfromthe1930Census(Rugglesetal.2021). 53

city fixed effects, these controls are also illustrative to rule out specific sources of bias. First, it could be that Dayton plans preferred urban areas, with more potential members, which may have been vulnerable to the Great Depression. Controlling for the urban share of the population does not affect the point estimate. Second, political factors across California may have led some areas to enact different zoning regulations or even to respond to the Great Depression differently. I show that political party, as measured by the average Republican vote share from 1896-1928, does not affect the results. Similarly, the Progressive vote share in the Election of 1912 also does not affect the point estimate. I then include measures of the potential severity of the recession. I caution that these controls could be considered a bad control if the presence of Dayton institutions led to worse real outcomes during the Great Depression. Still, given how “small” B&L’s were in some areas, it is useful to show that the magnitude of the coefficient is little changed when including this variable. As discussed by Courtemanche and Snowden (2011), declining values of B&L shares may also have led members to delay loan payments to obtain HOLC financing. I show that including HOLC loans per capita does little to the point estimate. Results are also robust to using the bank failure rate from Carlson and Mitchener (2009) in the city. At the county level, results are robust to log change in retail sales by county from Fishback and Kantor (2018) or the growth in log change in real farmland values from Haines, Fishback, and Rhode (2018). Similarly, I include the growth in farmland values during the boom period using data from Haines, Fishback, and Rhode (2018). This coefficient captures whether areas with high growth in land value simply had more Dayton plans due to entry in response to good conditions. The coefficient is again little changed. Figure C2 repeats this exercise using the liabilities measure of Dayton plan. Given that I do not observe the number of members in 1929, I do not include this control as it would lower the sample size. Second, the concentration index greatly raises the point estimate. There is a high correlation between this variable and the liabilities measure because both are constructed using the securities share. For all other controls, the coefficient estimate is again robust to their inclusion. Probit/Logit Specification I show results using probit/logit models. Suppose closure is modeled in the following latent variable framework, where y∗ denotes some latent variables. i  1 y∗ >0 Closure = i (2) i 0 otherwise y∗ =α+βDayton +ΓX +η (3) i i i i Ishowresultsusingaprobitspecification(assumingη isnormallydistributed)andusingalogitdistribution i (assuming that the error term η is distributed by the standard logistic distribution), and estimate this i equation via maximum likelihood. The results of this estimation, presented in Table C1 are shown as odds ratios. The coefficient estimate in Column (1) implies that Dayton institutions are two and a half times more likely to close compared with Non-Dayton institutions. As before, the point estimate is stable when including the same sets of controls as in the table in the main text, as shown in Columns (2)-(6). Cox Specification TableC2presentsresultsusingaCoxspecification.Theresultsfortheliabilitiesmeasurearestrong,andallpointestimatessuggestthatnon-Daytonplansaremorelikelytocloseearlier. 54

Closure Characteristics Tables C3 and C4 drop types of closures to investigate where the variation in closure rates is coming from. Assuming that the type of closure is uncorrelated with the plan type, then all coefficients are mechanically downward biased in these tables. Table C3 drops involuntary closures. As expected, the coefficient falls somewhat, although remains broadly stable across the specifications. For the liabilities meeasure, the point estimate stays large and significant. TableC4dropsconsolidationsandtransfers.Again,asexpected,thecoefficientfalls.However,thecoefficient estimateremainsbroadlystableacrossthereportedspecification.Fortheliabilitiesspecification,theestimate fall somewhat when including the full set of controls on the liabilities measure. This is solely because of the agecontrols(theestimateisunchangedwhenincludingonlyB&Landcitycontrols,unreported). Therefore, for non-consolidations and transfers, there is a somewhat stronger negative correlation between age and involuntary closure. This result is not worrisome, as closures due to consolidations are also indicative of flightiness. Federalization Inthebaselinesample,IchosetodropB&L’sthatarefederalizedratherthancloseorstay open as state institutions. The reason for this is because it is not clear why B&L’s choose to federalize. It couldbethatweakB&L’sthatmighthaveclosedchosetofederalizebecauseofadditionalaccesstoliquidity. On the other hand, strong B&L’s may have federalized because, as explained by Snowden (2003), much of the legislation was written by the B&L operators of the time period. Because of the unknown relative ordering in terms of the outcomes, specifically with respect to the decision to federalize vs. close, I estimate a multinomial logit model. Pr(Result ∈{Closure,Federalize})=α+βDAYTON +ΓX +ε (4) i i i i TheresultsarepresentedininTableC5.Thebaselevelisstayingopen,soallcoefficientsshouldbeinterpreted as the relative risk ratio of the listed result happening relative to staying open for a given change in the independent variable. The independent variable of interest is DAYTON . For closed institutions, as above, i we see that the effect of being a Dayton institution increases the probability of closure. However, there is no significant effect of being a Dayton institution on the probability of being federalized. If anything, being a Daytonplanreducestheprobabilityoffederalizing.Iinterprettheseresultsnotassayingthatfederalization was completely random, but rather the decision to federalize was unrelated to the institution’s liability structure. Alternatively, one might think that federalized BL(cid:32) ’s should not be dropped from the sample. One argument may be that these institutions are not in danger of closing, just changing their regulator for idiosyncratic reasons.TableC6showsthattheresultsareunchangedwhenincludingtheseinstitutionsasopenthroughout the sample. Dropping 1929 Closures IpresentresultsdroppingallB&L’sthatclosedin1929.TableC7presentsthe results of this estimation. Ordered Logit Specification I have so far assumed that the timing of closure is irrelevant. However, it may be the case that non-Dayton plans closed earlier than Dayton plans due to the Great Depression, but Dayton plans closed over time as deflation raised interest rates. To estimate whether or not Dayton 55

plans closed earlier than non-Dayton plans, I follow Postel-Vinay (2016) in estimating an ordered logistic model. ClosureOrder∗ ={j;κ ≤y∗ ≤κ } (5) i j−1 j y∗ =βDayton +ΓX +ε (6) i i i where κ are estimated cutoff value. ClosureOrder is an ordered variable of closure for institution i as j i described below. y∗ is a latent variable estimated as the linear combination of controls. The main variable of interest, Dayton is the type of the institution (Dayton vs. non-Dayton), and X is a vector of additional i i controls. Here,thedependentvariableClosureOrder isnolongersimplyadummyvariableindicatingclosure.Instead, i this variable is equal to the number of years an institution survives from 1927 through 1935.34 For example, if a B&L is alive in 1929 but closes in 1932, this value is 3. If it survives into 1936, then it takes on the highest value of 7. Table C8 shows the distribution of banks in this way, broken down by type. We can see that both Dayton and Non-Dayton plans had high rates of closure early on in the Depression before settling down a bit. At first glance, it also looks like Dayton plans not only had higher closure rates throughout the time period, but also had a peak slightly earlier, in 1929. The results of estimating Equation 5 are shown in Table C9. Coefficients are again expressed as odds ratios. Column (1) shows, consistent with earlier results, that Dayton B&L’s had a significantly lower chance of survivinglongerintotheRecessionthanDaytonB&L’s.ThepointestimateimpliesthattheoddsofaDayton plansurvivinganotheryearis0.34timesthanthatforpermanentplans.Thisresultisstablewhenincluding other controls, shown in the remaining columns, or using the alternative measure of Dayton plan. Survivor Bias TableC10showsresultstakingintoaccountsurvivorbias.Inthefirstfourcolumns,Ishow results dropping other subsamples that would be affected by survivor bias for non-Dayton plans that may be stronger because they have survived previous recessions. The first column repeats the results from the benchmark analysis. The next column drops all permanent B&Ls that entered prior to the 1890s. The next columns drops permanent B&Ls that enter between 1890 and 1906. Both results are statistically significant and similar in magnitude. The next column drops all non-Dayton plans that enter prior to 1906 (which leaves only 10). While the point estimate does not move much, it is insignificant. The liabilities measure of the Dayton plan is useful here since it is a bit more balanced. Dropping institutions Dayton plans that were born prior to 1906, the result is a statistically significant (and larger) coefficient estimate. Split by Closure Wave Table C11 shows results splitting into pre-1930 and post-1931. 34.Withthesevalues,κ−1 andκ9 areequaltominusinfinityandinfinity,respectively. 56

Figure C1: Stability of Dayton Coefficient: Reported Measure Thisfigureplotsthevalueand90%standarderrorbandsforthecoefficientβ fromestimatingEquation(1)usingthereported measure of plan type as the Dayton dummy. Each row includes the indicated control as well as the benchmark controls: cash ratio,logassets,andagebinfixedeffects.TherowlabeledbenchmarkistheresultfromTable(4).RealEstateOwnedisreal estateownedasashareoftotalassetsin1929.1930LoanRepaymentsarethetotalreceiptsonloanprincipalrelativetototal receipts in 1930. The Loan to Value Ratio is the average loan for an institution divided by the average home value in a city. Guarantee Stock Share is total guarantee stock as a share of total liabilities in 1929. The Concentration Index is the sum of guarantee stock share and the withdrawable share of total assets, meant to capture how concentrated voting rights are. Log Members(1927)isthelognumberofmembersreportedin1927,asthe1929annualreportsdonotlisttotalmembers.Percent Urbanistheshareofthepopulationinthecountythatliveinurbanareas.Avg.Repub.VoteShare1896-1928istheaverage of the republican vote share between 1896-1928. Progressive/Republican is the voting share for the Progressive party in the 1912 election. HOLC Loans per Capita is the per capita amount of HOLC lending in the county. The bank failure rate is the failure rate of banks in the city as in Carlson and Mitchener (2009). Real retail sales is the decline in retail sales per capita from 1929-1933. Average farmland value comes from the agricultural census, and controls are either 1920-1925 or 1925-1935. Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009),Fishback,Horrace, andKantor(2005),Haines,Fishback,andRhode(2018),FishbackandKantor(2018) Heteroskedasticityrobuststandarderrorsinparenthesesdenotedbytheerrorbars. 57

Figure C2: Stability of Dayton Coefficient: Liabilities Measure Thisfigureplotsthevalueand90%standarderrorbandsforthecoefficientβ fromestimatingEquation(1)usingtheliabilities measure of plan type as the Dayton dummy. Each row includes the indicated control as well as the benchmark controls: cash ratio,logassets,andagebinfixedeffects.TherowlabeledbenchmarkistheresultfromTable(4).RealEstateOwnedisreal estateownedasashareoftotalassetsin1929.1930LoanRepaymentsarethetotalreceiptsonloanprincipalrelativetototal receipts in 1930. The Loan to Value Ratio is the average loan for an institution divided by the average home value in a city. Guarantee Stock Share is total guarantee stock as a share of total liabilities in 1929. The Concentration Index is the sum of guarantee stock share and the withdrawable share of total assets, meant to capture how concentrated voting rights are. Log Members(1927)isthelognumberofmembersreportedin1927,asthe1929annualreportsdonotlisttotalmembers.Percent Urbanistheshareofthepopulationinthecountythatliveinurbanareas.Avg.Repub.VoteShare1896-1928istheaverage of the republican vote share between 1896-1928. Progressive/Republican is the voting share for the Progressive party in the 1912 election. HOLC Loans per Capita is the per capita amount of HOLC lending in the county. The bank failure rate is the failure rate of banks in the city as in Carlson and Mitchener (2009). Real retail sales is the decline in retail sales per capita from 1929-1933. Average farmland value comes from the agricultural census, and controls are either 1920-1925 or 1925-1935. Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009),Fishback,Horrace, andKantor(2005),Haines,Fishback,andRhode(2018),FishbackandKantor(2018) Heteroskedasticityrobuststandarderrorsinparenthesesdenotedbytheerrorbars. 58

Table C1: Closure Rates: Logit and Probit Models Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.695∗∗ 0.772∗ 1.177∗∗ 1.330∗ (0.272) (0.413) (0.483) (0.780) Dayton (Liabilities) 0.633∗∗∗ 0.492∗ 1.042∗∗∗ 0.906∗∗ (0.201) (0.260) (0.340) (0.455) N 164 164 205 205 164 164 205 205 B&L Controls N Y N Y N Y N Y Age Controls N Y N Y N Y N Y Local Controls N Y N Y N Y N Y Model Probit Probit Probit Probit Logit Logit Logit Logit This table presents results from estimating Equation (3) using probit and logit specifications. Closurei is a dummy variable equaltooneifBuildingandLoanAssociationiwasabsorbed,closed,consolidated,ortransferred.Thereportedmeasureisthe plantypeasdescribedbythe1927annualreports,whiletheliabilitiesmeasureisadummyequaltooneiftheassociationhas above-medianinvestmentcertificatesasashareofliabilities,asdescribedinthetext.B&Lcontrolsincludelogassetsandcash percentage.Agecontrolsincludeagebinfixedeffectsasdescribedinthetext.Localcontrolsincludelogcitypopulationandlog numberofcommercialbanksinthecity.Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),Carlson andMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 59

Table C2: Results using Cox Model Closure Closure Closure Closure Dayton (Reported) 0.945∗∗ 1.157∗∗ (0.438) (0.576) Dayton (Liabilities) 0.800∗∗∗ 0.621∗∗ (0.285) (0.296) N 926 926 1127 1127 B&L Controls N Y N Y Age Controls N Y N Y This table presents results from estimating a Cox model. Closurei is a dummy variable equal to one if Building and Loan Associationiwasabsorbed,closed,consolidated,ortransferred.Thereportedmeasureistheplantypeasdescribedbythe1927 annualreports,whiletheliabilitiesmeasureisadummyequaltooneiftheassociationhasabove-medianinvestmentcertificates asashareofliabilities,asdescribedinthetext.B&Lcontrolsincludelogassetsandcashpercentage.Agecontrolsincludeage binfixedeffectsasdescribedinthetext.Localcontrolsincludelogcitypopulationandlognumberofcommercialbanksinthe city.Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 60

Table C3: Dropping Involuntary Closures Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.158∗∗ 0.178∗∗ 0.194∗ 0.163 (0.0726) (0.0787) (0.111) (0.109) Dayton (Liabilities) 0.203∗∗∗ 0.200∗∗∗ 0.171∗∗ 0.174∗∗ (0.0641) (0.0653) (0.0712) (0.0734) N 147 147 147 147 182 182 182 182 R-Squared 0.02 0.03 0.06 0.09 0.05 0.05 0.06 0.11 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y .ThistablepresentsresultsfromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closureiisadummyvariable equaltooneifBuildingandLoanAssociationiwasabsorbed,voluntaryclosureconsolidated,ortransferred,withinvoluntary closuresdropped.Thereportedmeasureistheplantypeasdescribedbythe1927annualreports,whiletheliabilitiesmeasure is a dummy equal to one if the association has above-median investment certificates as a share of liabilities, as described in the text. B&L controls include log assets and cash percentage. Age controls include age bin fixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. Source: Building and Loan Commissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) .∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 61

Table C4: Dropping Consolidations and Transfers Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.204∗∗∗ 0.210∗∗∗ 0.212∗∗ 0.148∗ (0.0601) (0.0648) (0.0887) (0.0870) Dayton (Liabilities) 0.182∗∗∗ 0.177∗∗∗ 0.0872 0.0952 (0.0632) (0.0652) (0.0743) (0.0788) Constant 0.0606 0.0247 0.0124 -0.390 0.145∗∗∗ 0.0793 -0.0506 -0.307 (0.0418) (0.326) (0.321) (0.359) (0.0450) (0.317) (0.313) (0.374) N 139 139 139 139 175 175 175 175 R-Squared 0.04 0.05 0.07 0.15 0.04 0.04 0.07 0.21 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y ThistablepresentsresultsfromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closurei isadummyvariable equaltooneifBuildingandLoanAssociationiwasabsorbed,closed,withconsolidationsandtransfersdropped.Thereported measureistheplantypeasdescribedbythe1927annualreports,whiletheliabilitiesmeasureisadummyequaltooneifthe association has above-median investment certificates as a share of liabilities, as described in the text. B&L controls include log assets and cash percentage. Age controls include age bin fixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. Source: Building and Loan Commissioner, (Various Years), Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 62

Table C5: Including Federalization as an Outcome (Multinomial Logit Results) Closure Closure Closure Closure Closure Closure Closure Closure Close of Business Dayton (Reported) 3.246∗∗ 3.454∗∗ 4.192∗∗ 3.771∗ (1.568) (1.724) (2.934) (2.964) Dayton (Liabilities) 2.836∗∗∗ 2.759∗∗∗ 2.186∗ 2.469∗ (0.964) (0.950) (0.881) (1.141) Federalize Dayton (Reported) 0.530 0.863 0.159∗ 0.0669∗∗ (0.421) (0.807) (0.162) (0.0714) Dayton (Liabilities) 1.743 2.001 0.980 1.118 (1.080) (1.279) (0.564) (0.722) N 171 171 171 171 219 219 219 219 Chi Squared 7.13 16.12 241.94 399.24 9.59 24.96 718.05 634.64 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y Results for estimating Equation 4, where Closure is when an association absorbed, closed, consolidated, or transferred and Federalizeiswhenanassociationisfederalized.Thereportedmeasureistheplantypeasdescribedbythe1927annualreports, while the liabilities measure is a dummy equal to one if the association has above-median investment certificates as a share ofliabilities,asdescribedinthetext.B&Lcontrolsincludelogassetsandcashpercentage.Agecontrolsincludeagebinfixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses.Exponentiatedcoefficients 63

Table C6: Including Federalized B&Ls as Closure Closure Closure Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.224∗∗∗ 0.232∗∗∗ 0.249∗∗ 0.199∗ (0.0709) (0.0760) (0.103) (0.102) Dayton (Liabilities) 0.207∗∗∗ 0.196∗∗∗ 0.147∗ 0.148∗ (0.0631) (0.0644) (0.0785) (0.0793) Constant 0.150∗∗∗ 0.204 0.189 -0.294 0.208∗∗∗ 0.0526 -0.0415 -0.364 (0.0568) (0.374) (0.367) (0.443) (0.0481) (0.334) (0.336) (0.405) N 171 171 171 171 219 219 219 219 R-Squared 0.04 0.04 0.07 0.12 0.04 0.04 0.06 0.14 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y ThistablepresentsresultsfromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closurei isadummyvariable equal to one if Building and Loan Association i was absorbed, consolidated, closed, transferred, or federalized. The reported measureistheplantypeasdescribedbythe1927annualreports,whiletheliabilitiesmeasureisadummyequaltooneifthe association has above-median investment certificates as a share of liabilities, as described in the text. B&L controls include log assets and cash percentage. Age controls include age bin fixed effects as described in the text. Local controls include log city population and log number of commercial banks in the city. Source: Building and Loan Commissioner, (Various Years), Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 64

Table C7: Closure Rates: Linear Probability Model Dropping 1929 Closures Closure Closure Closure Closure Closure Closure Closure linprob8 Dayton (Reported) 0.221∗∗∗ 0.222∗∗∗ 0.232∗∗ 0.164 (0.0657) (0.0714) (0.105) (0.107) Dayton (Liabilities) 0.242∗∗∗ 0.227∗∗∗ 0.160∗∗ 0.160∗∗ (0.0629) (0.0648) (0.0765) (0.0781) N 147 147 147 147 186 186 186 186 R-Squared 0.05 0.05 0.10 0.17 0.06 0.07 0.10 0.22 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y ThistablepresentsresultsforthecoefficientβfromestimatingEquation(1):Closurei=α+βDaytoni+ΓXi+εi.Closureiis adummyvariableequaltooneifBuildingandLoanAssociationiwasabsorbed,closed,consolidated,ortransferredafter1929. Thereportedmeasureistheplantypeasdescribedbythe1927annualreports,whiletheliabilitiesmeasureisadummyequal tooneiftheassociationhasabove-medianinvestmentcertificatesasashareofliabilities,asdescribedinthetext.B&Lcontrols includelogassetsandcashpercentage.Agecontrolsincludeagebinfixedeffectsasdescribedinthetext.Localcontrolsinclude logcitypopulationandlognumberofcommercialbanksinthecity.Source:BuildingandLoanCommissioner,(VariousYears), Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 65

Table C8: Classification of Closure Timing for Ordered Logit Self-Reported Liabilities Year Value Non-Dayton Dayton Total Non-Dayton Dayton Total 1929 0 3 14 17 5 14 19 1930 1 2 9 11 2 20 22 1931 2 0 12 12 2 12 14 1932 3 0 3 3 1 5 6 1933 4 1 2 3 1 3 4 1934 5 0 2 2 0 2 2 1935 6 0 7 7 1 8 9 Survive 7 31 78 109 36 93 129 Total 37 127 164 48 157 205 Timingofclosurebyplantype.Forself-reported,IdenoteDaytonasanyinstitutionwithatleast50%relianceoninvestment securities.Source:BuildingandLoanCommissioneroftheStateofCalifornia(1927) 66

Table C9: Ordered Logit Specification Closure Closure Closure Closure Closure Closure ologit7 ologit8 Dayton (Reported) 0.342∗∗ 0.316∗∗ 0.256∗ 0.304 (0.177) (0.169) (0.182) (0.234) Dayton (Liabilities) 0.384∗∗∗ 0.380∗∗∗ 0.463∗∗ 0.461∗∗ (0.134) (0.136) (0.167) (0.179) N 164 164 164 164 205 205 205 205 Chi-Squared 4.32 4.74 8.31 17.52 7.48 7.46 9.36 21.23 B&L Controls N Y Y Y N Y Y Y Age Controls N N Y Y N N Y Y Local Controls N N N Y N N N Y ThistablepresentsresultsforthecoefficientβfromestimatingEquation(5).Thereportedmeasureistheplantypeasdescribed bythe1927annualreports,whiletheliabilitiesmeasureisadummyequaltooneiftheassociationhasabove-medianinvestment certificatesasashareofliabilities,asdescribedinthetext.B&Lcontrolsincludelogassetsandcashpercentage.Agecontrols includeagebinfixedeffectsasdescribedinthetext.Localcontrolsincludelogcitypopulationandlognumberofcommercial banksinthecity.Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses.ExponentiatedCoefficients 67

Table C10: Survivor Bias Closure Closure Closure Closure Closure Closure Dayton (Reported) 0.253∗∗ 0.268∗∗ 0.252∗∗ 0.213 (0.109) (0.134) (0.119) (0.170) Dayton (Liabilities) 0.161∗∗ 0.275∗∗∗ (0.0811) (0.0936) N 164 147 149 133 205 160 R-Squared 0.07 0.06 0.07 0.05 0.07 0.09 B&L Controls Y Y Y Y Y Y Age FE Y Y Y Y Y Y Drop non-Day Start After 1890 Drop non-Day End Before 1890 1906 1906 1906 This table presents results for the coefficient β from estimating Equation (1): Closurei =α+βDaytoni+ΓXi+εi. Results drop non-Dayton plans that enter between the years listed. Closurei is a dummy variable equal to one if Building and Loan Associationiwasabsorbed,closed,consolidated,ortransferred.Thereportedmeasureistheplantypeasdescribedbythe1927 annualreports,whiletheliabilitiesmeasureisadummyequaltooneiftheassociationhasabove-medianinvestmentcertificates asashareofliabilities,asdescribedinthetext.B&Lcontrolsincludelogassetsandcashpercentage.Agecontrolsincludeage binfixedeffectsasdescribedinthetext.Localcontrolsincludelogcitypopulationandlognumberofcommercialbanksinthe city.Source:BuildingandLoanCommissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.HeteroskedasticityRobustStandardErrorsinParentheses. 68

Table C11: Split by Wave Closure Closure Closure Closure Dayton (Reported) 0.133 0.215∗∗∗ (0.108) (0.0819) Dayton (Liabilities) 0.173∗∗∗ 0.160∗∗ (0.0618) (0.0765) N 137 136 170 186 R-Squared 0.03 0.09 0.06 0.10 B&L Controls Y Y Y Y Age Controls Y Y Y Y Wave Pre-1930 Post-1930 Pre-1930 Post-1930 This table presents results for the coefficient β from estimating Equation (1): Closurei =α+βDaytoni+ΓXi+εi. Results aresplitbyclosurewave.Closurei isadummyvariableequaltooneifBuildingandLoanAssociationiwasabsorbed,closed, consolidated,ortransferred.Thereportedmeasureistheplantypeasdescribedbythe1927annualreports,whiletheliabilities measureisadummyequaltooneiftheassociationhasabove-medianinvestmentcertificatesasashareofliabilities,asdescribed inthetext.B&Lcontrolsincludelogassetsandcashpercentage.Agecontrolsincludeagebinfixedeffectsasdescribedinthe text. Local controls include log city population and log number of commercial banks in the city. Source: Building and Loan Commissioner,(VariousYears),Bleemer(2016),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 69

D Additional Balance Checks Table D1 shows a balance table for the Dayton plans across institutions in and out of the sample of banks from the archives. Table D2 shows a balance table for the non-Dayton plans across institutions in and out of sample of B&Ls from the archives. 70

Table D1: Balance Table: Dayton Only (1) (2) (3) Variable Missing Micro Data Diff Members (Thousands) 1.86 1.48 -0.38 (2.26) (2.16) (0.40) Closure 0.69 0.16 -0.52*** (0.47) (0.37) (0.07) Total Assets (Millions) 3.43 1.99 -1.45* (6.15) (2.89) (0.82) Cash (% Assets) 4.45 4.48 0.03 (3.69) (3.96) (0.69) Real Estate Owned (% Assets) 1.63 1.50 -0.13 (2.28) (2.76) (0.46) Concentration Index (1929) 29.11 26.85 -2.27 (31.91) (26.92) (5.23) Secur. Share of Liabs 0.78 0.82 0.04 (0.37) (0.32) (0.06) Shares (% Liabilities) 20.18 15.47 -4.71 (33.81) (27.78) (5.47) 1930 Loan Repayments 2.03 0.10 -1.93 (11.25) (0.85) (1.33) Banks 38.76 36.77 -1.99 (36.19) (35.82) (6.46) City Population (Thousands) 351.01 405.63 54.61 (481.63) (514.81) (89.93) Observations 54 73 127 “Closure Dummy” is a dummy variable equal to one if a building and loan Association was absorbed, closed, consolidated, or transferred. “Investment Securities Share of Member Funds” calculated as investment securities divided by the sum of investmentsecuritiesandwithdrawableshares.“Age”calculatedasnumberofyearsopenasof1929.“Dayton(Reported)”and “Members(Thousands)”usedatafromthe1927annualreportsandsodropB&Lsformedin1927-1929.Source:Buildingand LoanCommissioner,(VariousYears),SuperintendentofBanks(1935),Bleemer(2016),Fishback,Horrace,andKantor(2005), Haines,Fishback,andRhode(2018),FishbackandKantor(2018),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 71

Table D2: Balance Table: Non-Dayton Only (1) (2) (3) Variable Missing Micro Data Diff Members (Thousands) 1.31 0.50 -0.81* (1.98) (0.45) (0.41) Closure 0.33 0.08 -0.25* (0.49) (0.28) (0.13) Total Assets (Millions) 1.83 0.86 -0.97 (3.32) (0.83) (0.70) Cash (% Assets) 2.71 2.93 0.22 (3.01) (2.23) (0.88) Real Estate Owned (% Assets) 1.57 0.59 -0.98* (2.21) (0.88) (0.50) Concentration Index (1929) 69.80 63.95 -5.85 (33.43) (30.55) (11.06) Secur. Share of Liabs 0.28 0.34 0.06 (0.37) (0.36) (0.13) Shares (% Liabilities) 66.27 57.62 -8.64 (34.50) (32.74) (11.70) 1930 Loan Repayments 0.03 0.00 -0.03 (0.11) (0.00) (0.02) Banks 12.25 18.96 6.71 (22.22) (24.27) (8.30) City Population (Thousands) 86.43 220.82 134.39 (189.84) (340.04) (105.72) Observations 12 25 37 “Closure Dummy” is a dummy variable equal to one if a building and loan Association was absorbed, closed, consolidated, or transferred. “Investment Securities Share of Member Funds” calculated as investment securities divided by the sum of investmentsecuritiesandwithdrawableshares.“Age”calculatedasnumberofyearsopenasof1929.“Dayton(Reported)”and “Members(Thousands)”usedatafromthe1927annualreportsandsodropB&Lsformedin1927-1929.Source:Buildingand LoanCommissioner,(VariousYears),SuperintendentofBanks(1935),Bleemer(2016),Fishback,Horrace,andKantor(2005), Haines,Fishback,andRhode(2018),FishbackandKantor(2018),CarlsonandMitchener(2009) ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 72

E Additional Mechanisms Table - Liabilities Measure Table E1 shows results for the flightiness mechanism using the liabilities measure. The results are qualitatively, and in most cases quantitatively, similar to the main text. 73

Table E1: Evidence on the Role for Flightiness in Predicting Closure (a) Withdrawal Fees and the Costs of Membership by Type Withdrawal Penalty Dues Shares per Member Costs Dayton (Liabilities) -0.131 -0.0165 -1.678∗∗∗ -8.884∗∗∗ (0.0885) (0.0416) (0.213) (2.088) N 164 164 164 164 R-Squared 0.17 0.41 0.43 0.39 B&L Controls Y Y Y Y Age Controls Y Y Y Y (b) Archival Evidence: Member Returns and Loan Characteristics Return Borrower Share Lending Rate Log Avg Loan Size Dayton (Liabilities) -0.138∗∗ 0.0529 0.188 -0.114 (0.0673) (0.0345) (0.144) (0.0795) N 120 120 120 120 R-Squared 0.06 0.21 0.08 0.13 B&L Controls Y Y Y Y Age Controls Y Y Y Y (c) Withdrawal Fees Difference in Difference Fees Ratio Dayton (Liabilities) X 1930 0.864∗∗∗ (0.318) N 298 R-Squared 0.82 B&L FE Y Year FE Y Thetopandmiddlepanelsshowresultsfromestimatingtheequationyi=αi+βDAYTONi+ΓXi+εi.Theoutcomesforthe toppanelinclude:“WithdrawalPenalty,”adummyequaltooneifaB&Lhaspenaltiesforwithdrawingfunds;“Dues”denotes thecostofduesin1927;“SharesperMember”istheratiooftotalsharestototalmembers;“Costs”istheproductof“Dues”and “SharesperMember,”ortotalcostspermember.Theoutcomesforthemiddlepanelinclude“Return,”whichistheweighted averageofreturnsforinvestmentcertificatesandwithdrawableshares,wheretheweightsaregivenbytherelativeproportionof each; “Borrower Share,”which denotesthe share ofmembers that areborrowing, “Lending Rate,”whichdenotes the average rateonmortgageloans,and“LogAvgLoanSize,”orthelogoftheratiooftheamountofloanstothenumberofloans.B&L controlsincludelogassetsandcashpercentage,andagecontrolsincludeagebinfixedeffects.Agecontrolsforthispanel,which usesarchivaldata,includeadummyequaltooneiftheassociationwasincorporatedafter1920duetothelimitedsamplesize. Thebottompanelestimatesdifferences-in-differencesspecificationsoftheformYit =αt+βi+γ(DAYTONi×1(t=1930)). Source:BuildingandLoanCommissioneroftheStateofCalifornia(1927),Department of Savings and Loan Records ∗p<0.1,∗∗p<0.05,∗∗∗p<0.01.Heteroskedasticityrobuststandarderrorsinparentheses. 74