Policy Uncertainty and Bank Mortgage Credit

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Policy Uncertainty and Bank Mortgage Credit Gazi I. Kara and Youngsuk Yook 2019-066 Please cite this paper as: Kara,GaziI.,andYoungsukYook(2019). “PolicyUncertaintyandBankMortgageCredit,” FinanceandEconomicsDiscussionSeries2019-066. Washington: BoardofGovernorsofthe Federal Reserve System, https://doi.org/10.17016/FEDS.2019.066. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) 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 Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Policy Uncertainty and Bank Mortgage Credit GAZI I. KARA∗ YOUNGSUK YOOK† Federal Reserve Board Federal Reserve Board May 2019 ABSTRACT We document that banks reduce supply of jumbo mortgage loans when policy uncertainty increases as measured by the timing of US gubernatorial elections in banks’ headquarterstates. Thereductionislargerformoreuncertainelections. Weutilizehighfrequency,geographicallygranularloandatatoaddressanidentificationproblemarising fromchangingdemandforloans: (1)themicroeconomicdataallowforstate/time(quarter) fixed effects; (2) we observe banks reduce lending not just in their home states but also outside their home states when their home states hold elections; (3) we observe importantcross-sectionaldifferencesinthewaybankswithdifferentcharacteristicsrespond topolicyuncertainty. Overall,thefindingssuggestthatpolicyuncertaintyhasarealeffect onresidentialhousingmarketsthroughbankscreditsupplydecisionsandthatitcanspill overacrossstatesthroughlendingbybanksservingmultiplestates. Keywords: BankMortgageCredit;HousingMarket;PolicyUncertainty;Gubernatorial Elections ∗FederalReserveBoardofGovernors;e-mail: gazi.i.kara@frb.gov; †FederalReserveBoardofGovernors;e-mail: youngsuk.yook@frb.gov. Theviewsexpressedinthisarticle are those of the authors and not necessarily of the Federal Reserve System. We thank Matthew Carl for his excellentresearchassistance.

1. Introduction Theuncertaintiesassociatedwithpossiblechangesingovernmentpolicyorleadershipcan affect the behavior of firms through various channels, such as industry regulation, monetary and trade policy, and taxation. In particular, financial institutions, which operate in heavily regulated industries, likely face more uncertainty than nonfinancial firms when the political landscapechanges,andtheirresponsetosuchchangesmayhaveariffleeffectintheeconomy because of their role as intermediaries. The recent financial crisis is an example highlighting the role played by financial institutions as both originators and transmitters of shocks to the economy. Banks’mortgagecreditsupply,inparticular,hastraditionallyrepresentedasubstantial fraction of their loan supply, and the availability of mortgage credit has been documented tohaveimportantimplicationsforfinancialstability.1 Thispaperinvestigateshowpolicyuncertaintyaffectsbanks’mortgagelendingdecisions. Theoretically,ourpredictionisguidedbymodelsofinvestmentunderuncertainty(e.g.,Bernanke (1983) and Bloom, Bond, and Van Reenen (2007)), where firms become cautious and hold backoninvestmentinthefaceofuncertaintyiftheinvestmentisatleastpartiallyirreversible. These models lead to our prediction that banks reduce their investment in mortgage markets, thatis,supplyofmortgageloans,whenpolicyuncertaintyishigh. Empirically,however,there are challenges in identifying the effect of uncertainty on bank lending. First, uncertainties affect all economic agents including households, who are also likely to cut back on durable goods consumption and housing investment when facing higher uncertainty. Thus, any observable change in bank lending is an equilibrium outcome reflecting both credit supply from banks and demand from households. Second, a relationship between uncertainty and banks’ investment decision can be endogenous as the economic downturn itself can generate a great deal of political uncertainty. Thus, establishing a causal relationship requires an exogenous measureofpoliticaluncertainty. This paper attempts to address these challenges in two important ways. First, we take advantage of rich supervisory data: confidential Home Mortgage Disclosure Act (HMDA) 1See, for example, Mian and Sufi (2009), Adelino, Schoar and Severino (2014), and Favilukis, Ludvigson, andVanNieuwerburgh(2017). 1

data provide daily loan-level mortgage information. Unlike the public version of HMDA data, which is shown at an annual frequency, the confidential version provides exact loan dates, allowing us to evaluate the data at a higher frequency. We aggregate the daily data to mergewithbanks’quarterlyfinancialinformation. Aquarterlyfrequencyallowsustocontrol for changing demand dynamics better than an annual frequency. It also captures possible short-term effects of uncertainty on banks’ behavior that may be averaged out in annual data. Another advantage of using HMDA data is the availability of loan-level information, which helps address the identification challenge in two aspects. First, the micro-level data allow us to exploit cross-sectional differences across banks and evaluate whether banks with varying characteristics respond to political uncertainty differently. Second, the data allows a more geographicallygranularexaminationofbanks’lendingdecisions: banksmayexhibitdifferent lending behavior across states in response to different demand for mortgage credit. A bank’s lendingaggregatedatthenationalleveldoesnotrevealcross-sectionalvariationswithinabank servingmultiplestates. Weevaluatebanks’lendingdecisionsatthestatelevel,controllingfor each state’s time-varying demand for mortgage credit and other local economic conditions affectingbanks’lendingdecisions. Second, we employ a plausibly exogenous measure of policy uncertainty: the timing of U.S. gubernatorial elections. A state’s gubernatorial election increases policy uncertainty for banksheadquarteredinthestatebecauseapossiblechangeinstategovernmentleadershipcan leadtochangesinvariousstatepolicies,includingstatetaxes,subsidies,budget,andprocurement (Peltzman (1987), Besley and Case (1995), Colak, Durnev, and Qian (2017)). A state’s governor also has a strong influence over the appointment of the head of the state banking regulators, who in turn hold various regulatory powers such as chartering, rulemaking, supervision, and enforcement (Saiz and Semenov (2014), Labonte (2017)). These election dates are predetermined by law and are independent of the states’ economic conditions. Furthermore, different states hold gubernatorial elections in different years, allowing us to net out nationalbusinesscycleeffects. Infact,severalpreviousstudieshaveusedelectiontimingasa 2

quasi-natural experimental setting to identify the link between policy uncertainty and various economicoutcomes.2 For our analyses, we aggregate the daily loan-level HMDA data between 1990 and 2014 at the bank, state, and quarter level and merge with banks’ quarterly financial information and data on 323 gubernatorial elections across 48 U.S. states.3 We employ a differencein-difference methodology to exploit time-series variations within a bank as well as crosssectional variations across banks. Specifically, we are able to compare a bank’s lending behavior in election quarters and non-election quarters. We are also able to compare, at a given pointintime,banksfacingelectionsintheirhomestatesandthosethatarenotbecausebanks headquarteredindifferentstatesfacegubernatorialelectionsindifferentyears. We focus on the type of loans that we consider relatively more irreversible—loans held in banks’ balance sheets and jumbo loans. Models of investment under uncertainty suggest that irreversibility increases the information value of waiting to invest, causing investment to vary negativelywithfluctuationsinpolicyuncertaintyovertime. First,comparedtoloansthatwere justoriginated,loansthatbankshaveheldintheirbalancesheetsforsometime,so-calledseasonedloans,arenoteasytodispose,makingthemarelativelyirreversibleinvestment.4 Loans canbecomedelinquentwhileinbanks’possession,makingitdifficultforbankstosellatalater date. Even well-performing loans have to meet various requirements to be sold as seasoned loans. For example, for Fannie Mae to buy seasoned loans, it requires, among other things, that the mortgage satisfy Fannie Mae’s current applicable mortgage eligibility requirements, thatthecurrentvalueofthepropertynotbelessthantheoriginalvalue,andthattheborrower’s ability to pay not have changed adversely (Fannie Mae (2014)). Freddie Mac has similar requirements (Freddie Mac (2016)). Second, jumbo loans, those with an amount exceeding the conforming loan limit, cannot be purchased or securitized by government-sponsored enterprises (GSEs) such as Fannie Mae and Freddie Mac. The lack of government support makes 2ExamplesofinternationalstudiesusingthetimingofnationalelectionsareJulioandYook(2012)andJulio andYook(2016). U.S.studiesusingU.S.gubernatorialelectionsincludeGaoandQi(2013),Colaketal. (2017), Jens(2017),andAtanassov,Julio,andLeng(2016). 3WeexcludeNewHampshireandVermont, whichholdelectionseveryotheryearasopposedtoeveryfour years. 4Banks often dispose loans soon after they are originated by either selling to the government-sponsored enterprisesorbypoolingascollateralforprivatelabelmortgage-backedsecurities. 3

jumbo loans less liquid than conforming loans, thus more irreversible.5 In fact, most jumbo mortgagesareheldbytheoriginallenderwhileconformingloansareoftensolduponorigination. Theresultssupportourprediction: Ourbaselineestimationresultsshowthatbankscutthe volume of jumbo loans they either originate and hold or purchase and hold each quarter by approximately13%to25%comparedwhennon-electionquarters,controllingforvariousbank characteristics. The number of jumbo loans also declines by 4% to 6%. Figure 1 depicts the estimated jumbo mortgage credit cycle around elections. These estimates reflect changes in lendingbehaviorinbothbanks’homestatesandforeignstatesinwhichtheyprovidemortgage lending. Theresulthastwoimportantimplications. First,policyuncertaintymattersforbanks’ mortgagelendingdecisions. Thatis,policyuncertaintyhasarealeffectonresidentialhousing markets through banks’ supply of mortgage credit. Second, policy uncertainty in one state has a spillover effect to other states through lending by financial institutions serving multiple states. We address the identification problem in a few different ways. First, our regressions include state-time fixed effects, which help control for the time-varying demand for mortgage credit and other local economic conditions affecting banks’ lending decisions. Second, we exploit the fact that many banks in our sample lend outside their home states as well. If the decline in lending in election quarters is solely driven by changes in demand in banks’ home state, banks are unlikely to reduce lending in foreign states. We find that banks also reduce lending in their foreign states, not just in their home states, when their home states hold elections, suggesting that the pattern in the data is unlikely to be driven by changing demand in their home states. Finally, we examine whether there is heterogeneity across banks intheirsensitivitytoelectoraluncertainty. Ourpremiseisthatthechangeinlendingbehavior will vary with banks’ characteristics if it was driven by supply rather than demand for loans. In particular, we consider two bank characteristics. First, we compare state-chartered banks and nationally chartered banks serving the same state, and find that state-chartered banks reducejumbomortgagelendingmore, implyingthatpotentialchangestostatebankregulations 5Formoreonjumboloans,seeAmbrose,LaCour-Little,andSanders(2004),LoutskinaandStrahan(2009), Adelino,SchoarandSeverino(2014),amongothers. 4

following elections create an additional layer of uncertainty for state banks compared with national banks. The second characteristics we consider is banks’ risk-taking behavior. We construct three risk indicators based on three measures of risk-taking, respectively: z-score, equity ratio, and credit risk. We find that more risky banks reduce the supply of jumbo mortgage credit a bit more than less risky banks, possibly because more risky banks are more vulnerabletochangesinpolicyregimes. We then exploit election characteristics to further examine whether the result is indeed driven by the uncertainty generated by elections. If the reduction in lending was driven by electoral uncertainty, the effect will likely be larger when there is a higher degree of uncertaintyovertheelectionoutcomeand,hence,overfuturepolicy. Wefindthatmortgagelending cycles around elections are more pronounced in close races in which the outcome is highly uncertain. The decline in bank lending is also more severe in elections in which incumbent governors do not seek re-election due to binding term limits. Elections lacking incumbent candidates are likely more competitive and the uncertainty about election outcome is likely higher. These results suggest that jumbo mortgage credit supply declines more when uncertaintyabouttheelectionoutcomeishigher. We conduct various robust checks such as using pseudo-election dates and different subsamples and confirm that our results remain qualitatively unchanged. We also consider an alternativemeasureofjumbomortgagecredit–thevolumeandnumberofjumbomortgageloans banksoriginateregardlessofwhethertheyholdorselltheloans. Wefindthatthejumbomortgage credit cycle is present, though less pronounced. To the extent that origination variables capturerelativelymorereversibleinvestment,thisresultsupportstheviewthattheinvestmentuncertainty relation is likely more negative for more irreversible assets. Finally, we explore a broadersampleofloanstoincludebothjumboandnon-jumboloansandfindasimilarpattern althoughthedeclineinthequarterlyvolumeissmallerthanthatinthejumboloanmarket. Our work contributes to understanding how policy uncertainty affects housing markets throughitseffectonfinancialinstitutions. Canes-WroneandPark(2014)documentthathome prices and home sales decline in the year leading up to gubernatorial elections. However, their finding is an equilibrium outcome, reflecting both mortgage credit supply and demand 5

effects. While isolating supply and demand effects is generally a challenging task, we make an important first step in separating the policy uncertainty effect coming from the supply side by exploiting the fact that the timing of gubernatorial elections is exogenous and staggered acrossstatesandthefactthat manybanksinoursampleprovidemortgagecreditinboththeir homestatesandforeignstates. Ourworkisalsorelatedtothestudieslinkingpolicyuncertaintyandfinancialinstitutions’ creditsupply. MostrelatedtoourstudyisGissler,Oldfather,andRuffino(2016),whichshow a negative correlation between banks’ perceived uncertainty over the regulation of qualified mortgages and their mortgage lending. Bordo, Duca, and Koch (2016) document that bank credit growth is negatively related to Baker, Bloom, and Davis (2016)’s economic policy uncertainty (EPU) index. Berger, Guedhami, Kim, and Li (2018) document that banks hoard moreliquidityasEPUincreases. Usingloan-leveldatafromItaliancreditregistry,Alessandri and Bottero (2017) document a reduction in banks’ approval rates of commercial and industrialloansandanincreaseinthedurationofanapprovalprocesswhenEPUishigh. TheEPU index, which captures the frequency of news articles indicating uncertainty about economic policy, is very useful in gauging a country’s changing policy uncertainty over time. However, as noted by Baker, Bloom, and Davis (2016), identifying a causal relation between the EPU index and economic activities is challenging because policy responds to economic conditions and is likely to be forward looking. Also, because it is measured at the country level, it is not straightforwardtodisentangletheuncertaintyeffectfromthenationalbusinesscycleeffect. In syndicated loan markets, Kim (2017) utilizes national elections around the world to establish acausalinferencebetweenpolicyuncertaintylendersfaceandfirms’borrowingcost. Finally, our finding about the cross-state spillover of policy uncertainty adds to the literature on the role that multi-market banks play in the cross-market spillover of economic shocks. Peek and Rosengren (1997, 2000) show that Japanese banks transmitted shocks that originatedfromJapanin1990stotheU.S.bycuttingbackthecommercialrealestatelending in their U.S. branches. Berrospide, Black, and Keaton (2016) examine banks that operated in multipleU.S.metropolitanareasduringthehousingmarketcollapseof2007-09anddocument that the banks, in response to high overall mortgage delinquencies in some markets that they were serving, reduced mortgage lending in other markets. Schnabl (2012) also documents 6

a spillover of the effect of Russian debt default to foreign banks’ lending in Peru. Cetorelli andGoldberg(2011),DeHaasandVanHoren(2012),andGiannettiandLaeven(2012)study the spillover effect in the case of cross-border lending by banks exposed to shocks during the financialcrisisof2007–09. 2. Data We obtain daily mortgage loan information between 1990 and 2014 from the confidential Home Mortgage Disclosure Act (HMDA) data. The HMDA of 1975 is a law requiring most banks,savingsandloanassociations,creditunions,andconsumerfinancecompaniestoreport every mortgage application received. As a result, the data provide a substantial coverage of theUnitedStatesmortgagemarket. Avery,Brevoort,andCanner(2007)estimatethatHMDA covers approximately 80% of all home loans nationwide in 2006.6 The mandatory reporting threshold for depository institutions has changed over time but almost all commercial banks are included in the data. In 2014, for example, any bank with assets above $43 million, with a branch in a metropolitan statistical area, and that originated at least one mortgage loan had to file a HMDA report. The HMDA data provide detailed information on loan applications and originations such as the date of an application and origination, loan amount and location, approval status, lender information as well as the information on mortgage applicants such as theirincome,sex,andrace. We clean raw HMDA data, taking similar steps as those in Loutskina and Strahan (2009). We drop mortgages originated by savings institutions, mortgage bankers, credit unions, and othernonbanklenders. WethendropmortgagessubsidizedbytheFederalHousingAuthority, the Veterans Administration, or other government programs. We also drop applications with missingcharacteristicssuchasloansize,propertylocation,orthebank’sapprovaldecisionon the loan. We only keep home purchase loans for owner-occupied, principal dwelling homes. Finally,toexcludeoutliers,wedropindividualmortgageloanssmallerthan$10,000orlarger than$10million. 6Avery,Brevoort,andCanner(2007)provideanextensivediscussionofHMDAdata. 7

We identify jumbo loans using the county-level conforming loan limits provided by the Federal Housing Finance Agency (FHFA) for one-unit properties.7 Prior to 2007:Q3, conformingloanlimitsweresetatthenationallevelandwereadjustedannuallytoreflectinflation, increasingfrom$187,450in1990to$417,500in2006.8 Startingin2007:Q3,conformingloan limitshavevariedacrosscountiesdependingonwhetheracountybelongstoageneralorhigh costarea. Accordingly,weapplyFHFA’snation-wideloanlimitstodatapriorto2007:Q3and county-level loan limits to data starting in 2007:Q3. Approximately 25% of counties in the HMDA data do not have conforming loan limits listed in the FHFA data. For these counties, wereplacemissingvalueswithconformingloanlimitsforgeneralareas. Next, we aggregate the loan-level information at the state, bank, and quarterly level to merge with banks’ quarterly financial information and the data on 323 U.S. gubernatorial elections across 48 states between 1990 and 2014. We exclude an observation if a bank does not originate, purchase or deny at least one loan, jumbo or not, in a given state in a given quarter. This step helps ensure banks in our sample have a footprint in the state’s mortgage market. Because jumbo loans are not originated or purchased as frequently as conforming loans are, we apply the following procedure to distinguish banks that do not operate in the jumbo loan market from those that operate but happen to add no new jumbo loans to their balance sheets in a given quarter: For each 4-year election cycle, we only consider banks that either originate and hold or purchase and hold jumbo loans at least three out of four quarters in the year before an election. These data cleaning procedures result in 207,535 observations atthebank/state/quarter-leveland49,597observationsatthebank/quarterlevel. Table1summarizestheloanandbankcharacteristicsinformationatthebank/quarterlevel. Note that banks’ financial information is not available at the state level. Each quarter, banks either originate and hold or purchase and hold an average of $11.14 million worth of jumbo loans and an average number of 17 jumbo loans nationwide. The median values are much smaller, suggesting a large variation across banks in their presence in the jumbo loan market. Note that these loans held are those that are not sold and hence held on the balance sheet 7https://www.fhfa.gov/DataTools/Downloads/Pages/Conforming-Loan-Limits.aspx 8ExceptforAlaskaandHawaiiwherelimitsare50percenthigher. 8

of a bank at least until the end of the calendar year.9 These loans represent about 0.28% of banks’ total assets in each quarter. That is, about 0.28% of banks’ assets worth of new jumbo loansareaddedtobanks’balancesheetseachquartereitherthroughoriginationsorpurchases. This is in addition to existing jumbo loans in banks’ balance sheets. When averaged at the bank/state/quarter level, the ratio is about 0.06% (untabulated). The ratio is smaller because the ratio constructed at the bank/state/quarter level uses for the denominator a bank’s assets consolidatedacrossallstatesinwhichthebankoperateswhileusingforthenumeratorabank’s jumbo mortgage activity at the state level. Turning to origination variables, the volume and numberofjumboloanoriginationisabitlargerthanthecorrespondingvolumeandnumberof loansheldwith$14.82millioninvolumeandabout25loansinnumberperquarternationwide. This suggests that some of the jumbo loans banks originate are sold within the same calendar year. Table 1 also reports banks’ quarterly financial information drawn from so-called Call Reports. Each quarter, commercial banks must file either “Consolidated Reports of Condition and Income for a Bank with Domestic and Foreign Offices” (FFIEC 031) or “Consolidated Reports of Condition and Income for a Bank with Domestic Offices Only” (FFIEC 041). These Call Reports contain banks’ detailed financial statements. They also provide information on the location of a bank’s headquarter, allowing us to identify the home state of each bank. Weutilizethemerger-adjustedversionofthepublicCallReportdata. Banksthatdonot file a HMDA form are excluded from our sample. The table shows that banks in our sample haveanaverageof$6.8billioninassets. Coredepositsareabout69%oftotalassetsandaverage return on equity is about 3% each quarter. These banks hold about 21% of their assets in theformofhomemortgagesloans,whichconsistoffirstandsecondlienmortgagesandhome equityloans. Threebankriskmeasuresarereported: z-score,equityratio,andcreditrisk. Finally,theelectioninformationisprimarilyobtainedfromtheCQPressVotingandElections Collection and is supplemented by Guide to U.S. Elections by Kalb (2015). All states in our sample have gubernatorial elections every 4 years. We exclude New Hampshire and Vermont, which have elections every two years. Table 2 summarizes the characteristics of 9BanksarerequiredbytheHMDAtoreportwhethertheyhavesoldaloanbytheendofthecalendaryearin whichitwasoriginated. 9

gubernatorialelections. Electionsinourdatahaveanaveragevotemarginof15.8%wherethe vote margin is defined as the percentage difference of votes between the winner and runnerup. Usingthisinformation,weconstructanindicatorvariable,Close,whichissettooneifan electionoutcomewasdeterminedbylessthana5-percentmarginandzerootherwise. Abouta quarter of elections in our sample are classified as close. Similarly, we construct an indicator variable, Wide, which is set to one if an election outcome was determined by more than a 15-percent margin and zero otherwise. Next, Term limited is an indicator variable showing whetheranincumbentgovernorfacesatermlimitimposedbythestateselectoralrulesornot. In a quarter of elections in our sample, incumbent governors do not seek re-election due to term limits. Finally, the last row reports that new governors are elected in about a half of elections,leadingtoachangeinleadership. 3. Methodology A key feature of our empirical setting is that we use the timing of gubernatorial elections as a proxy for exogenous variations in policy uncertainty. The timing of elections is fixed by electorallawandoutofthecontrolofanindividualbank,andhence,independentofeconomic conditions. Furthermore,differentstatesholdgubernatorialelectionsindifferentyears,allowing us to net out national business cycle effects. Our main variables of interest are election quarterindicators,whichallowustocapturethemortgagelendingdynamicsaroundelections. This setup enables us to exploit variations within a bank over time by comparing a bank’s lending behavior in election quarters and non-election quarters. In addition, because banks headquartered in different states face gubernatorial elections in different years, we are able to compare, at a given point in time, banks facing elections in their home states and those that are not. In essence, we employ a difference-in-differences methodology and estimate the followingspecification: 1 Y = α +α + ∑ β Elect +X(cid:48)θ+ε . (1) i,s,t i,s s,t k i,h,t+k i,s,t k=−2 10

The specification includes bank-state fixed effects (α ) and state/time fixed effects (α ). i,s s,t Including a full set of state/time fixed effects helps control for the time-varying demand for mortgagecreditandotherlocaleconomicconditionsaffectingbankslendingdecisionsineach state. The state-time fixed effects are analogous to firm-time fixed effects used in studies that focusonidentifyingthechangesinsupplyfromdemandforC&Iloansbycontrollingfortimevarying observed and unobserved heterogeneity across borrowing firms (e.g, Jime´nez et al (2012,2014)). Inoursetting,includingafullsetofstate-timefixedhelpscontrolforobserved and unobserved heterogeneity across states that borrow from banks to identify changes in the mortgage loan supply.10 Note that state-time fixed effects do not absorb the election effect becausemanybanksinoursamplelendnotonlyintheirhomestatesbutalsoinforeignstates. Forthedependentvariable,weconsiderseveralvariablescapturingbanks’lendingbehavior in the jumbo loan market. The main dependent variable we use is log(1+Volume held), where Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quarter t. Jumbo loans are originated or purchased relatively infrequentlycomparedwithconformingloans. Thus,itispossibleVolumeheld becomeszero because a bank cut jumbo lending to zero in some quarters rather than because a bank does not operate in the jumbo loan market. To ensure that such observations are not excluded, we addonetoVolumeheld beforetakingthelogarithm.11 Election quarter indicators, Elect (k = -2, -1, 0, 1), are set to one if bank i’s home i,h,t+k statehholdsagubernatorialelectioninquartert+k,andzerootherwise. Whilethedependent variableisdefinedbasedonthestateinwhichabankextendsaloan,electionquartervariables are defined based on a bank’s home state to capture the uncertainty arising from a bank’s home-state election. Elect is the quarter leading up to an election, the three-month period t fromSeptemberthroughNovemberoftheelectionyear. Becauseelectionstakeplaceinearly Novemberandbecausethereissomelagbetweenloanapprovalandorigination,thisdefinition captures the quarter leading up to an election more precisely than the last calendar quarter 10ThisidentificationstrategybuildsuponKhawajaandMian(2008)whousedfirmfixedeffectstocontrolfor creditdemandbycomparingtheborrowingbythesamefirmfrombanksthatdifferintheirexposuretoaliquidity shock. 11Notethatourdatacleaningprocedure,detailedinsection2,excludesobservationsforwhichVolumeheldis zerolikelybecauseabankdoesnotoperateinthejumboloanmarket. 11

before an election, which is from July to September.12 Coefficients on the election dummy variables can be interpreted as the difference in the within-bank conditional mean mortgage lending,controllingforotherdeterminantsoflending. Finally, the specification includes various time-varying bank characteristics (X) that can affect banks mortgage lending decisions over time. We lag all bank-level controls by one quartertoalleviateapotentialendogeneityconcern. Size,definedasthelogarithmofabank’s totalinflationadjustedassets,mayhelpexplainbanks’lendingdecisioniflargerbanksbehave differently than small ones in the mortgage market. We also include home mortgage, defined as the sum of first lien and junior lien residential real estate loans and home equity loans as a fraction of total assets. A bank’s mortgage lending decision can be affected by its business strategy as reflected in its concentration on home mortgage relative to its size. A bank’s dependence on core deposits, measured as the ratio of core deposits to total assets, can affect a bank’s willingness to extend mortgage credit. Core deposits can encourage risk-taking due to its stable nature as a funding source and deposit insurance associated with core deposits. Finally, a bank’s profitability, measured by return on equity, may also affect its mortgage lendingdecision. Standarderrorsaredoubleclusteredatthebankandstatelevel. 4. Mortgage Lending around Gubernatorial Elections 4.1. Bank–level Analysis We start with a bank-level analysis using data aggregated at the bank/quarter level, subsequently followed by a more granular analysis at the bank/state/quarter level. Table 3 shows the bank-level results. The first column uses log(1+Volume held) as the dependent variable, where Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in quarter t across all states. Coefficients of Elect , Elect , and Elect t−2 t−1 t are all negative and all but one are statistically significant. The pattern is similar using as the dependentvariablelog(1+Numberheld)incolumn (2)andvolumeheld/lag(assets)incolumn 12TheresultsaresimilarwhenwedefineElect asthethreemonthsfromAugusttoOctober. i,h,t 12

(3), respectively. These results imply that banks’ jumbo mortgage lending aggregated across allstatesdeclineswhenbanksfacegubernatorialelectionsintheirhomestates. Animportantdrawbackofthisspecificationisthatitcannotaddresstheidentificationproblem rising from changing loan demand at the state level. Different states hold gubernatorial elections in different years, resulting in varying degrees of uncertainty shocks across states in a given year. These state-level changes in demand cannot be accounted for by including nationwidemacrotrendsortimetrends. Inthenextsub-section,weutilizebank/state/quqarter leveldatatocontrolfortime-varyingdemandatthestatelevel. 4.2. Baseline Results: Bank/State–level Analysis Table4reportstheestimationresultsofthebaselinespecification(specification(1))atthe bank/state/quarter level. The first column uses log(1+Volume held) as the dependent variable. Note that Volume held is now defined at the state level. That is, Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quarter t. Coefficients of Elect , Elect , and Elect are all negative and statistically significant, t−2 t−1 t suggesting that banks reduce jumbo mortgage lending when banks’ home states hold elections. The magnitude of the reduction is economically large: The point estimates of the three coefficients range between -0.122 and -0.225, implying that, in the quarters leading up to an election,bankscutthevolumeofjumbomortgagesupplybybetween13%(=exp(0.122)−1) and 25% (= exp(0.225)−1) relative to the volume in non-election quarters, controlling for variousbankcharacteristics. Theelectioneffectweakensafteranelection,butdoesnotgoawayswiftly. Thecoefficient onElect remainsnegative,thoughsmallerinmagnitudethanthoseonpre-electionquarter t+1 variables. This lagged response is quite plausible considering that it takes time for a bank to process loan applications and originate loans. Thus, loans likely appear in banks’ books with some lags. In addition, while the uncertainty over an election outcome is resolved upon an election, there is some lingering uncertainty about the elected governor’s administration and agenda, more so in the case of a newly elected governor. Jens (2017), for example, points out 13

that stock market volatility is higher for several months after a new governor is elected than whenanincumbentisre-elected.13 The next column of table 4 uses log(1+Number held) as the dependent variable, where Numberheld isthenumberofjumboloansbankieitheroriginatesandholdsorpurchasesand holdsinstatesinquartert. Theresultissimilar: CoefficientsofElect ,Elect ,andElect t−2 t−1 t areallnegativeandsignificant. Themagnitudeofcoefficientsissmallerwiththereductionof 4%to6%inthenumberofloanscomparedwithnon-electionquarters,controllingforvarious bank characteristics. These results suggest that larger jumbo loans are likely affected more in electionyears. We find that coefficients of bank-level control variables generally have signs consistent with the literature. Bank size, measured as lagged bank assets, is positively correlated with jumbomortgagelendingincolumns(1)and(2),implyingthatlargebankshavemorepresence inthejumbomortgagemarket. Homemortgagesalsohavepositivecoefficients. Thatis,banks with a higher concentration in the mortgage market extend more jumbo loans. Banks relying moreoncoredepositsalsotendtoengagemoreinjumbomortgagelending. For robustness, the last column considers as the dependent variable the ratio of Volume held to the bank’s assets from a year ago, multiplied by 100.14 Because a bank’ assets are notbrokendowntothestatelevelinCallReports,weuseforthedenominatorabank’sassets consolidatedacrossallstatesinwhichthebankoperateswhileusingforthenumeratorabank’s jumbomortgageactivityatthestatelevel. Thismakestheratiosmallerthanifstate-levelbank assetswereused. Alsonotethatthenumeratorcapturesthejumboloansbanksnewlyacquired andheldinagivenquarter,whichisverysmallcomparedwithexistingjumboloansinbanks’ books. The regression result is qualitatively similar to those in the first two columns: All three pre-election quarter variables have negative and significant coefficients. As expected, the coefficients are small, ranging between -0.006 and -0.009. This means that the ratio of newly held jumbo loans to a bank’s assets declined between -0.006% and -0.009% in each 13See, also, Biakowski, Gottschalk, and Wisniewski (2008), Boutchkova, Doshi, Durnev, and Molchanov (2011),andKelly,Pastor,andVeronesi(2016). 14Theloanvolumeisscaledbyassetsfourquartersagoratherthanbyassetsinthepreviousquartertomitigate thepotentialseasonalityissueassociatedwiththequarterlyfrequencyofthedata. 14

of the pre-election quarters. This is a quite sizable change compared with the mean ratio of 0.06%. Overall, the results have two important implications. First, policy uncertainty matters for banks’ mortgage lending decisions. Second, the reduction in lending captured in the regressionsreflectsthereductioninbothbanks’homestatesandforeignstatesinwhichtheyprovide mortgagecredit. Thatmeansthatpolicyuncertaintyinonestatehasaspillovereffecttoother statesthroughfinancialinstitutionsservingmultiplestates. 4.3. Jumbo Mortgage Credit in Home States vs. Foreign States While our results highlight an important transmission mechanism through which policy uncertainty is passed on to households, one may wonder whether our results are driven by a decline in demand for mortgage loans rather than by a decline in banks’ credit supply. Our baselinespecificationincludesstate-timefixedeffects,whichhelpcontrolforthetime-varying demand for mortgage credit across states. In addition, because many banks in our sample operateinmultiplestates,weareabletocomparebanksexposedtosameeconomicconditions but different degrees of uncertainty: Those facing elections in their home state and those operating in the same state but headquartered elsewhere. However, we further investigate the question by comparing loans extended in banks’ home states and those in their foreign states. If the results are solely driven by a decline in demand, the reduction in loans should be concentrated in banks’ home states where uncertainty surrounding elections may depress demandformortgagecredit. Specifically, we introduce interaction terms between our quarterly election dummies and a home state dummy, which takes a value of one if the lending takes place in a bank’s home state. All regressions continue to include state-time fixed effects. Table 5 reports the results for the same dependent variables used in our baseline table 4: the volume and number of loans held, and the ratio of loans held to total assets. Across all three specifications, the quarterly election dummies remain negative and significant, suggesting that banks cut back lending outside of their home states as well. Meanwhile, the interaction terms are negative, 15

significantandlargeintheelectionquarterandthepost-electionquarter,despitebeingpositive and generally significant in the two previous quarters. Taken together, the interaction terms suggest that banks first start cutting back the credit supply more in foreign states, possibly tryingtomaintainbetterrelationshipwiththeirhomestate,butoncetheelectioncomesclose, cuttingcreditinthehomestatebecomesunavoidableaswell. These results provide additional support for our interpretation that the estimated lending cycles around elections are at least partly driven by changes in banks’ credit supply. Purely demand-driven changes around home states’ elections are unlikely to reduce the volume and number of mortgage loans to banks’ foreign states, where credit demand would remain stable onaverage. 4.4. Bank Characteristics and Sensitivity to Policy Uncertainty This section examines whether there is heterogeneity across banks in their sensitivity to electoraluncertainty. Inparticular,weconsidertwobankcharacteristics. First,wetestwhether state-chartered banks and nationally chartered banks headquartered in the same state respond differently to uncertainty surrounding the state’s gubernatorial election. We conjecture that state-chartered banks can be more sensitive to the change in their state’s political leadership. State banks are subject to both state and federal supervision as state and federal banking regulators alternate examinations of state banks while national banks are only subject to federal banking supervision. In addition, a state’s governor has a strong influence over the appointment of the head of the state banking regulators. The choice of state regulators is important to state banks as state regulators can implement identical rules differently than federal regulatorsduetodifferencesintheirinstitutionaldesignandincentivesandcancounteractfederal regulators’actionstosomedegree(Agarwal,Lucca,Seru,andTrebbi(2014)). However, changes in a state’s political landscape are not limited to bank regulation. They can affect both state and national banks headquartered in the state through various channels such as state taxes, subsidies, budget, and procurement. Liu and Ngo (2014) argue that government plays a broad and active role in the banking sector and that banks consider political 16

interference as a serious risk factor.15 Thus, it is possible that the differential effect of electionsonstatebanksareonlymarginal. Inaddition,legislationhasstrengthenedtheregulatory authority of the federal regulators relative to that of state regulators over time, potentially mitigatingthedifferentialeffect(LevertyandGrace(2016)). The second bank characteristic that we consider is banks’ risk-taking behavior. Banks’ risk-takingpatternhasbeendocumentedtobeassociatedwiththeprobabilityoftheirsurvival, especially during crises.16 Similarly, electoral uncertainty may matter more to risky banks because they are likely more vulnerable to changes in policy regimes. On the other hand, banks’ risk-taking tendency may persist over time, leading more risky banks to react less to the uncertainty surrounding elections. We construct three risk indicators based on each of the following three bank risk measures: z-score, equity ratio, and credit risk. Z-score estimates a bank’s capital and return buffers with respect to its return volatility to evaluate the bank’s distance to default. Equity ratio measures a bank’s leverage and is considered an important measure of a bank’s soundness and stability. Credit risk, measured as the ratio of risk-weighted assets to total assets, indicates how risky a bank’s asset combination is and is positivelyassociatedwithabank’sprobabilityofdefault. To test these hypotheses, we augment our baseline specification as follows to allow for interactionsbetweenbankcharacteristicsandelectionquartervariables: 1 1 Y = α +α + ∑ β Elect + ∑ γ Elect ·Z i,s,t i,s s,t k i,h,t+k k i,h,t+k i,h,t k=−2 k=−2 +δZ +X(cid:48)θ+ε , i,h,t i,s,t where Z is the bank characteristic variable of interest. For the state bank hypothesis, the bank characteristicvariableisStatebank,whichissettooneifthegivenbankisstate-charteredand zero if nationally chartered. For the risk-taking hypothesis, the bank characteristic variable is High risk, an indicator set to one if the value of a bank risk measure is in the top tercile of 15Related, Leverty and Grace (2017) and Kroszner and Strahan (1996) document government intervention in the U.S. insurance industry and thrift, respectively, and Dinc¸ (2005) and Brown and Dinc¸ (2005) document governmentinterventioninbanksindevelopingcountries. 16Forexample,seeBeltrattiandStulz(2012),ColeandWhite(2012),BergerandBouwman(2013),DeYoung andTorna(2013),andKaraandVojtech(2017). 17

thedistributionintermsoftheriskiness. Forz-scoreandequityratio,forwhichhighervalues indicatelessrisk,Highriskissettooneifthevalueoftheriskmeasureisinthebottomtercile of the distribution. For credit risk, for which higher value means higher risk, High risk is set tooneifthevalueisinthetoptercileofthedistribution. Weuseriskmeasureslaggedbyfour quarterstominimizeendogeneityconcerns. Table 6 reports the results. In the first column, we test whether state-chartered banks and national banks headquartered in the state respond differently to uncertainty surrounding the state’selections. Interactiontermsareallnegativeandtwoofthecoefficients,Elect ×State t−1 bankandElect ×Statebank,arestatisticallysignificantatthe5%and1%levels,respectively. t Itmeansthatstatebanksaremoresensitivetouncertaintysurroundinggubernatorialelections thannationalbanks. However,itdoesnotmeanthattheuncertaintycomingfromgubernatorial elections is limited to risks associated with state-level banking supervision. National banks also cut jumbo mortgage lending around elections as indicated by negative and significant pre-election quarter variables. One caveat is that state banks can choose to switch to national banksandviceversa. However,itisaveryrareeventandisunlikelytoaffecttheresults.17 The next three columns interact high-risk indicators with election quarter variables. Election quarter variables are all negative and mostly significant, indicating that less risky banks cut jumbo mortgage lending around elections. Turning to interaction terms, we observe that some election quarter variables interacted with high-risk indicators have negative and significant loadings, implying that more risky banks react a bit more to electoral uncertainty than less risky banks. When High risk constructed based on z-score values is used in column (2), High risk ×Elect has a negative and significant loading. When equity ratio and credit risk t−2 valuesareusedtoconstructHighriskindicatorsincolumns(3)and(4),respectively,Highrisk indicatorsinteractedwithElect andElect havenegativeandsignificantloadings. These t−2 t−1 results suggest that, earlier in the election year, more risky banks tend to reduce the supply of jumbo mortgage credit more than less risk banks. When elections are near, however, risky banks reduce lending at about the same pace as less risky banks. Results are qualitatively the 176%ofbanksinourjumbo-loansampleswitchedbetweenstateandnationalchartersonceduringoursample periodand0.5%switchedtwice. 18

samewhenwereconstructHighriskindicatorsusingdifferentcutoffvaluesofunderlyingrisk measures(unreported). 5. Election Characteristics and Sensitivity to Policy Uncertainty This section exploits various election characteristics to further examine whether the documented lending cycle is indeed driven by the uncertainty generated by elections. If the reduction in lending was driven by uncertainty, the effect would likely be higher when there is a higher degree of uncertainty over future policy. In some cases, election outcomes are predicted with a great deal of confidence prior to the election date. However, other elections are characterizedbyverycloseracesinwhichtheoutcomeishighlyuncertainuntilthedayofthe election. We investigate variation in electoral uncertainty by using vote margins as a proxy for the degree of uncertainty. We construct a dummy variable, Close, which is set to one if the vote margin in an election is less than 5%, and zero otherwise, where the vote margin is defined as the difference between the proportion of the votes garnered by the winner and the proportionreceivedbytherunner-up. Wealsoconstructanindicatorvariable,Wide,tocapture elections with wide victory margins, which are likely to be associated with less uncertainty. Wide is set to one if the vote margin is more than 15% and zero otherwise. Among elections in our sample, 26% are classified as close elections and 42% as wide-margin elections (table 2). In addition, we examine whether an incumbent governor faces a term limit imposed by the state’s electoral rules. Previous studies document that the advantage of incumbency is an important predictor of the election outcome: If an incumbent governor faces a term limit and, thus,cannotrunforre-election,competitionsurroundingtheelectionislikelymorefierceand the uncertainty about the election outcome are likely higher. To capture the variation in the incumbency advantage across elections, we define an indicator variable, Term limited, which issettooneifanincumbentfacesatermlimitandzerootherwise. Inoursample,incumbents facetermlimitsinabout25%ofelections(table2). 19

Weaugmentthebaselinespecificationasfollowstoallowforinteractionsbetweenelection characteristicsandelectionquartervariables: 1 1 Y = α +α + ∑ β Elect + ∑ γ Elect ·Z i,s,t i,s s,t k i,h,t+k k i,h,t+k i,h,t k=−2 k=−2 +X(cid:48)θ+ε , i,s,t where Z is the election characteristics variable of interest including Close, Wide, and Term limited. Table 7 reports the results. Only the election quarter variables and their interaction terms are reported in the table to save space. Column (1) usesClose as the election characteristics variable. All election quarter variables have negative and significant coefficients. In addition, interactiontermsareallnegativeand,inparticular,thecoefficientofElect ×Close,-0.107, t−1 is large in magnitude and statistically significant. The coefficient suggests that banks lower the volume of jumbo loans they either originate and hold or purchase and hold in that quarter by 11% more in close elections than in other elections. This finding suggests that the effect of electoral uncertainty is more pronounced in close election races where uncertainty about election outcome tends to be higher. Turning to column (2), we see that all election quarter variablesremainnegativeandsignificant. Consistentwithourprediction,interactiontermsare generally positive and significant, implying that cycles in mortgage lending around elections are less pronounced when races are highly predictable. Finally, column (3) interacts Term limited with election quarter variables. As predicted, interaction terms have negative and statistically significant coefficients. This means that banks cut credit supply more when an incumbent governor cannot run for re-election due to term limits, likely because uncertainty ishigherinthoseelections. Overall, the results in this section are consistent with the interpretation that mortgage credit supply declines more when uncertainty about the election outcome is higher. That is, the pattern in the data are likely driven by uncertainty surrounding elections. We also note that these results are consistent with the view described in the previous section that, after an election, there is some lingering uncertainty about the elected governor’s administration and 20

agenda. The dampening election effect is slower to go away after close elections and after elections where the governor faces a term-limit. The interaction terms, Elect ×Close and t+1 Elect ×Term limited are both negative and significant. The post-election negative effect t+1 isstrongerforterm-limitedelections,wherelingeringuncertaintyislikelyhighersinceanew governor replaces the incumbent regardless of the election outcome. Meanwhile, the election effectisnearlygonefollowingawide-marginelection: ThesumofthecoefficientofElect t+1 (-0.182) and that of Elect ×Wide (0.182) is zero. That is, close elections and term-limited t+1 elections appear to be highly contested and have more unresolved uncertainty even after the electionoutcomeisrevealed. 6. Additional Tests 6.1. Jumbo Loan Origination Theanalysessofarhaveexaminedthevolumeandnumberofjumbomortgageloansbanks either originated and held or purchased and held in their balance sheets. In this section, we consider alternative measures of jumbo mortgage credit–the volume and number of jumbo mortgageloansbanksoriginateineachstateandeachquarterregardlessofwhethertheyhold or sell the loans. These measures also exclude loans purchased rather than originated. These origination variables include loans that are sold soon after origination, a relatively more reversible form of investment. In the models of investment under uncertainty, irreversibility increasestheinformationvalueofwaitingtoinvest. Thus,theinvestment-uncertaintyrelation is likely more negative for more irreversible assets. To the extent that origination variables capture investment that is relatively less costly to reverse, the mortgage credit cycle may be lesspronouncedthanwhenloansheldwereusedintable4. Ontheotherhand,theresultsmay be similar because jumbo mortgages are often held by the original lender rather than being solduponorigination. Table 8 estimates the baseline specification using three origination variables: (1) log(1+ Volume originated), where Volume originated is the volume of jumbo loans bank i originates 21

in state s in quarter t, (2) log(1+Number originated), and (3) the ratio of Volume originated to the bank’s assets from a year ago. The results are qualitatively the same as those in table 4 with all three pre-election quarter variables showing negative and significant coefficients. However,theeconomicmagnitudeisgenerallysmaller. Column(1)showsthatthecoefficients of pre-election quarter variables range between -0.079 and -0.110, indicating a decline in the quarterly jumbo mortgage origination volume of about 8-12% relative to the volume in nonelection quarters. This is much smaller than a reduction of 13-25% in table 4 using volume held. This is well depicted in figure 1. Similarly, column (2) shows that the coefficients are smaller than the corresponding values in the baseline results. Column (3) shows that the ratio of Volume originated to lagged assets declined between 0.007% and 0.011%, slightly more thanthedeclineof0.006%and0.009%intable4usingvolumeheld/lag(assets). Thisislikely because the origination volume, which is on average larger than the volume held, declines moreintermsofthedollaramountandhencemoreasafractionofassetswhiledecliningless asafractionofpreviousvolumethanthevolumeheld. 6.2. All Loans In this section, we explore a broader sample of loans to include both jumbo and nonjumboloans. Becausenon-jumboloansaremostlyconformingloansthatcanbesoldtoGSEs, they can be viewed as relatively more reversible investment than jumbo loans. However, conforming loans are also, to some extent, costly to reverse. Seasoned loans need to meet various requirements to be sold to GSEs. Even the loans that are sold upon origination carry some non-balance-sheet risks such as put back risk.18 We test whether the mortgage credit cycle is still present once non-jumbo loans are added to the sample. For consistency, we construct an all-loan sample in the same way as we did our jumbo-loan sample by following the data-cleaning procedures described in section 2. As we did with jumbo loans, for each 4yearelectioncycle,weonlyconsiderbanksthateitheroriginateandholdorpurchaseandhold loansinatleastthreeoutoffourquartersintheyearbeforeanelection. Thefinaldatacontain 18For more detail, see Tarullo (2010), which describes former Federal Reserve Governor Daniel Tarullo’s testimonybeforetheU.S.SenateCommitteeonBanking,Housing,andUrbanAffairs. 22

457,005 observations at the bank/state/quarter-level. Note that the all-loan sample contains morebanksasitincludesbanksthatdonotoperateinthejumbo-loanmarket. Table 9 repeats the regressions in table 4 using the new sample. Similar to jumbo-loan regression results, all election quarter variables have negative and significant loadings. This finding suggests that the mortgage credit cycle around elections is generally present in the mortgage loan market, not just in the jumbo loan market. The election effect on the loan volume appears less pronounced than in the baseline results using jumbo loans: Column (1) showsthatthecoefficientsofpre-electionquartersrangebetween-0.069and-0.158,compared withtherangeof-0.122and-0.225intable4. Thisimpliesthatthequarterlyvolumeofloans, inclusive of both jumbo and non-jumbo, that banks either originate and hold or purchase and holddropsbyabout7–17%comparedwiththevolumeinnon-electionquarters,controllingfor variousbankcharacteristics. Theelectioneffectonthenumberofallloans,ontheotherhand, seems somewhat more pronounced. Column (2) shows that the coefficients range between -0.065 and -0.088 while the corresponding coefficients in the baseline result range between -0.042 and -0.062. Note that larger reduction in the number of loans does not necessarily translate into larger reduction in the volume because non-jumbo loans are much smaller in size. 6.3. Robustness Checks In this section, we perform a few robustness checks. We use log(1+Volume held) as the dependentvariablefortheseregressions. InthefirstcolumnofTable10,werepeatthebaseline regression using pseudo election dates, which are constructed by, for each state, randomly selecting a year in which a state does not hold an election and treating the year and every four years after the year as the election years for the state. If our results are indeed driven by electoral uncertainty, the credit cycle documented in earlier sections should not be present in pseudo election years. The results in column (1) show that the volume of jumbo mortgage loansupplydoesnotdeclineinthepseudoelectionyears,consistentwithourprediction. 23

Next, we address the concern that the pattern in the data might be driven by uncertainty surroundingpresidentialelectionsasthetimingofsomegubernatorialelectionscoincideswith that of presidential elections. We repeat our baseline regression excluding states for which gubernatorialelectionstakeplaceinthesameyearaspresidentialelections. Thatis,allbanks headquartered in these states are excluded from the sample. Column (2) reports the result: Election quarter variables remain negative and significant, suggesting that the documented creditcycleispresentoutsidepresidential-electionyearsaswell. Finally, we examine whether the result changes when we exclude three large states (New York,California,andFlorida). Ifourresultwasdrivenbyanidiosyncraticpatternthatmaybe present in only a handful of large states, then the result is not likely to hold when these states withlargeobservationsareremovedfromthesample. Weexcludealljumboloansextendedto these three states and estimate our baseline specification. Column (3) shows that the election quartervariableshavenegativeandsignificantcoefficients,similartoearlierfindings. 7. Conclusion We examine the relationship between banks’ supply of jumbo mortgage credit and policy uncertaintyusingthetimingofU.S.gubernatorialelectionsasasourceofplausiblyexogenous variation in policy uncertainty. We document that when banks face gubernatorial elections in their home states, they reduce the volume and number of jumbo loans that they either originate and hold or purchase and hold each quarter relative to non-election quarters. Reduction inlendingisobservedbothinthestateinwhichbanksareheadquarteredandinforeignstates. The result has two important implications. First, policy uncertainty matters for banks’ mortgage lending decisions. Second, policy uncertainty in one state has a spillover effect to other statesthroughlendingbyfinancialinstitutionsservingmultiplestates. Thedocumentedeffect isunlikelytobedrivenbychangesindemand. Allregressionsincludestate-timefixedeffects, which help control for the time-varying demand for mortgage credit across states. Furthermore, the estimated mortgage credit cycle around elections is present in banks’ foreign states aswell. 24

The jumbo mortgage credit cycle around elections is more pronounced when there is a higher degree of uncertainty over the election outcome, as measured by vote margins and incumbent governors’ term limits. We also document that some banks are more sensitive to policy uncertainty than others: State banks and risky banks cut jumbo mortgage supply more likely because they are more vulnerable to increased policy uncertainty. The results remain basically unchanged to various robustness checks. The cycle is also present when origination variables are considered and when a sample inclusive of both jumbo and non-jumbo loans is employed. Overall, the results show that policy uncertainty has a real effect on residential housingmarketsthroughbanks’mortgagecreditdecisions. 25

References Adelino, Manuel, Antoinette Schoar, and Felipe Severino. 2014. Credit Supply and House Prices: EvidencefromMortgageMarketSegmentation. WorkingPaper. Adelino, Manuel, Antoinette Schoar, and Felipe Severino. 2016. Loan Originations and Defaults in the Mortgage Crisis: The Role of the Middle Class. The Review of Financial Studies 29.7: 1635-1670. Agarwal, Sumit, David Lucca, Amit Seru, and Francesco Trebbi. 2014. Inconsistent Regulators: EvidencefromBanking. TheQuarterlyJournalofEconomics129,889–938. Alessandri, Piergiorgio, and Margherita Bottero. 2017. Bank Lending in Uncertain Times. WorkingPaper. Ambrose, Brent W., Michael LaCour-Little, and Anthony B. Sanders. 2004. The Effect of Conforming Loan Status on Mortgage Yield Spreads: A Loan Level Analysis. Real Estate Economics32. 541–569. Atanassov,Julian,BrandonJulioandTiechengLeng. 2016. TheBrightSideofPolicyUncertainty: TheCaseofR&D.WorkingPaper. Avery, Robert B., Kenneth P. Brevoort, and Glenn B. Canner. 2007. Opportunities and IssuesinUsingHMDAData. JournalofRealEstateResearch29,351–379. Baker, Scott R, Nicholas Bloom, and Steven J Davis. 2016. Measuring Economic Policy Uncertainty. QuarterlyJournalofEconomics131,1593–1636. Beltratti, Andrea and Rene` Stulz. 2012. The Credit Crisis around the Globe: Why Did Some BanksPerformBetter? JournalofFinancialEconomics105,1–17. Berger, Allen N. and Christa H. S. Bouwman. 2013. How Does Capital Affect Bank PerformanceduringFinancialCrises? JournalofFinancialEconomics109,146–176. Berger, Allen N., Omrane Guedhami, Hugh Hoikwang Kim, and Xinming Li. 2018. EconomicPolicyUncertaintyandBankLiquidityHoarding. AvailableatSSRN3030489. Bernanke,BenS.1983. Irreversibility,Uncertainty,andCyclicalInvestment. QuarterlyJournalofEconomics98,85106. Berrospide, Jose M., Lamont K. Black, and William R. Keeton. 2016. The Cross-Market Spillover of Economic Shocks through Multimarket Banks. Journal of Money, Credit and 26

Banking48,957–988. Besley, Timothy, and Anne Case. 1995. Does Electoral Accountability Affect Economic Policy Choices? Evidence from Gubernatorial Term Limits. Quarterly Journal of Economics 110,769–798. Bialkowski, Jedrzej, Katrin Gottschalk, and Tomasz Piotr Wisniewski. 2008. Stock MarketVolatilityaroundNationalElections. JournalofBankingandFinance32,1941–1953. Bloom, Nicholas, Stephen Bond, and John Van Reenen. 2007. Uncertainty and Investment Dynamics. ReviewofEconomicStudies74,391–415. Bordo, Michael D., John V. Duca, and Christoffer Koch. 2016. Economic Policy Uncertainty and the Credit Channel: Aggregate and Bank Level U.S. Evidence over Several Decades. JournalofFinancialStability26,90–116. Boutchkova, Maria, Hitesh Doshi, Art Durnev, and Alexander Molchanov. 2011. PrecariousPoliticsandReturnVolatility. ReviewofFinancialStudies25,1111–1154. Brown, Craig O. and Serdar Dinc¸. 2005. The Politics of Bank Failures: Evidence from EmergingMarkets. QuarterlyJournalofEconomics120:1413–44. Canes-Wrone, Brandice, and Jee-Kwang Park. 2014. Elections, Uncertainty, and Irreversible Investment. BritishJournalofPoliticalScience44,83–106. Cetorelli, Nicola, and Linda Goldberg. 2012. Liquidity Management of U.S. Global Banks: Internal Capital Markets and the Great Recession. Journal of International Economics 86, 299–311. Colak, Gonul, Art Durnev and Yiming Qian. 2017. Policy Uncertainty and IPO Activity: Evidence from US Gubernatorial Elections. Journal of Financial and Quantitative Analysis 52,2523–2564. Cole, Rebel A., and Lawrence J. White. 2012. Deja Vu All Over Aagain: The Causes of US Commercial Bank Failures This Time Around. 2012. Journal of Financial Services Research42.1-2,5-29. De Haas, Ralph, and Neeltje Van Horen. 2012. International Shock Transmission after the Lehman Brothers Collapse: Evidence from Syndicated Lending. American Economic Review: PapersandProceedings,102,225–30. DeYoung, Robert and Gokhan Torna. 2013. Nontraditional Banking Activities and Bank FailuresduringtheFinancialCrisis. JournalofFinancialIntermediation22,397-421. 27

Dinc¸, I. Serdar. 2005. Politicians and Banks: Political Influences on Government-owned BanksinEmergingMarkets. JournalofFinancialEconomics77,no. 2: 453-479. Fannie Mae. 2014. Selling Guide: Fannie Mae Single Family. Washington: Fannie Mae, April,https://www.fanniemae.com/content/guide/sel041514.pdf. Favilukis, Jack, Sydney C. Ludvigson, and Stijn Van Nieuwerburgh. 2017. The Macroeconomic Effects of Housing Wealth, Housing Finance, and Limited Risk Sharing in General Equilibrium. JournalofPoliticalEconomy125,140–223. Freddie Mac. 2016. Single-Family Seller/Servicer Guide. McLean, Va.: Freddie Mac, June, http://www.freddiemac.com/singlefamily/guide/bulletins/pdf/062916Guide.pdf. Gao, Pengjie and Yaxuan Qi, 2013. Policy Uncertainty and Public Financing Costs: EvidencefromUSGubernatorialElectionsandMunicipalBondMarket. WorkingPaper. Giannetti, Mariassunta, and Luc Laeven. 2012. Flight Home, Flight Abroad, and InternationalCreditCycles. AmericanEconomicReview: PapersandProceedings,102,219–24. Gissler, Stefan, Jeremy Oldfather, and Doriana Ruffino. 2016. Lending On Hold: RegulatoryUncertaintyandBankLendingStandards. JournalofMonetaryEconomics81: 89–101. Jens, Candace E. 2017. Policy Uncertainty and Investment: Causal Evidence from U.S. GubernatorialElections. JournalofFinancialEconomics124,563–579. Jime´nez, Gabriel, Steven Ongena, Jose´-Luis Peydro´, and Jesu´s Saurin. 2012. Credit Supply and Monetary Policy: Identifying the Bank Balance-Sheet Channel with Loan Applications. AmericanEconomicReview102,no. 5: 2301–26. Jime´nez, Gabriel, Steven Ongena, Jose´-Luis Peydro´, and Jesu´s Saurina. 2014. Hazardous Times for Monetary Policy: What Do Twenty-Three Million Bank Loans Say About the EffectsofMonetaryPolicyonCreditRisk-Taking?. Econometrica82,no. 2: 463–505. Julio, Brandon and Youngsuk Yook. 2012. Policy Uncertainty and Corporate Investment Cycles,TheJournalofFinance67,45–84. Julio, Brandon, and Youngsuk Yook. 2016. Policy Uncertainty, Irreversibility, and Cross- BorderFlowsofCapital,JournalofInternationalEconomics103,13–26. Kalb,Deborah,ed. 2015. GuidetoUSElections. CQPress. Kara, Gazi I., and Cindy Vojtech. 2017. Bank Failures, Capital Buffers, and Exposure to 28

the Housing Market Bubble. Finance and Economics Discussion Series 2017-115. Washington: BoardofGovernorsoftheFederalReserveSystem. Kelly, Bryan, L’ubos Pa´stor, and Pietro Veronesi. 2016. The Price of Policy Uncertainty: TheoryandEvidencefromtheOptionMarket. JournalofFinance71,2417–2480. Khwaja, Asim Ijaz, and Atif Mian. 2008. Tracing the Impact of Bank Liquidity Shocks: EvidencefromanEmergingMarket. AmericanEconomicReview98,no4: 1413–42. Kim, Olivia, 2017. Does Policy Uncertainty Increase External Financing Costs? MeasuringtheElectoralPremiuminSyndicatedLending. WorkingPaper. Kroszner, Randall S. and Philip E. Strahan 1996. Regulatory Incentives and the Thrift Crisis: Dividends, Mutual-to-Stock Conversions, and Financial Distress. Journal of Finance 51, 1285–1319 Labonte, Marc. 2017. Who Regulates Whom? An Overview of the U.S. Financial RegulatoryFramework. CongressionalResearchService. Leverty, J. Tyler and Martin F. Grace. 2017. Do Elections Delay Regulatory Action? Journal ofFinancialEconomics,Forthcoming. Liu,Wai-Man,andPhongT.H.Ngo. 2014. Elections,PoliticalCompetitionandBankFailure. JournalofFinancialEconomics112,251–268. Loutskina, Elena, and Philip E. Strahan. 2009. Securitization and the Declining Impact of Bank Finance on Loan Supply: Evidence from Mortgage Originations. The Journal of Finance64.2,861–889. Mian,AtifandAmirSufi. 2009. TheConsequencesofMortgageCreditExpansion: Evidence fromtheU.S.MortgageDefaultCrisis. QuarterlyJournalofEconomics124,1449–1496. Peek,Joseph,andEricRosengren. 1997. TheInternationalTransmissionofFinancialShocks: TheCaseofJapan. AmericanEconomicReview87,495–505. Peek, Joseph, and Eric Rosengren. 2000. Collateral Damage: Effects of the Japanese Bank CrisisonRealActivityintheUnitedStates. AmericanEconomicReview90,30–45. Peltzman, S. 1987. Economic Conditions and Gubernatorial Elections. American Economic Review77,293–297. Saiz, Hector Perez and Aggey Semenov. 2014. The Effect of Campaign Contributions on StateBankingRegulationandBankExpansioninU.S.WorkingPaper. 29

Schnabl, Philipp. 2012. Financial Globalization and the Transmission of Bank Liquidity Shocks: EvidencefromanEmergingMarket. JournalofFinance67,897–932. Tarullo, Daniel K. 2010. Problems in Mortgage Servicing, statement before the CommitteeonBanking,Housing,andUrbanAffairs,U.S.Senate,December1, https://www.federalreserve.gov/newsevents/testimony/tarullo20101201a.htm. 30

Appendix: Variable Descriptions Variable Description DependentVariables Volumeheld Thevolumeofjumboloansbankieitheroriginatesandholdsorpurchasesandholds i,s,t instatesinquartert. Numberheld Thenumberofjumboloansbankieitheroriginatesandholdsorpurchasesandholds i,s,t instatesinquartert. Volumeoriginated Thevolumeofjumboloansbankioriginatesinstatesinquartert. i,s,t Numberoriginated Thenumberofjumboloansbankioriginatesinstatesinquartert. i,s,t ElectionVariables Elect Elect takesavalueofoneifabank’shomestateholdsagubernatorialelection t+k t+k inquartert+k,andzerootherwise,wherethequarterleadinguptoanelection(Elect) t isdefinedasthethree-monthperiodfromSeptembertoNovember. Close Anindicatorvariablesetequaltooneifthevotedifferenceinanelectionis lessthan5%,andzerootherwise,wherevotedifferenceisdefinedasthe differencebetweentheproportionofthevotesgarneredbythewinnerand thatreceivedbytherunner-up. Wide Anindicatorvariablesetequaltooneifthevotedifferenceinanelectionis morethan15%,andzerootherwise Newgovernor Anindicatorvariablesetto1ifanewgovernoriselectedinanelectionand zeroifanincumbentisre-elected. Termlimited Termlimited isequaltooneifanincumbentgovernorfacesabindingterm limitandcannotrunforre-election,andzerootherwise. OtherVariables Size Thelogarithmofabank’stotalassets. Homemortgages Thesumoffirstlienandjuniorlienresidentialrealestateloansandhome i,t equityloansasafractionoftotalassets. (cont’dinthenextpage) 31

Variable Description Coredeposits Thesumoftransactiondeposits,savings,andsmalltimedepositsdividedbytotal assets. Returnonequity Netincomedividedbyaverageequity. i,t ROA × totalequityi,t Z-score i,t totalassetsi,t ,whereROA ,isabank’sreturnonassetsaveragedover i,t i,t sd(ROA ) i,t 8quartersbetweent andt−7. Similarly,sd(ROA )isstandarddeviationofabank’s i,t returnonassetscalculatedover8quartersbetweent andt−7. Equityratio Theratiooftotalequitytototalassets. i,t Creditrisk Theratioofrisk-weightedassetstototalassets. i,t 32

Figure1. ConditionalMeanJumboMortgageCreditAroundElections Thisfiguredepictsthevolumeofjumbomortgagecreditsupplyaroundgubernatorialelectionsusingtheregressioncoefficientsoftheelectiontimingdummyvariablesreportedincolumn(1)oftables4and8. x-axiscaptures the quarters around elections where quarter 0 indicates the last quarter leading up to a gubernatorial election measuredbyElect . y-axisshowslog(1+Volumeheld)andlog(1+Volumeoriginated),whereVolumeheldisthe t volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quarter 0, and Volumeoriginatedisthevolumeofjumboloansbankioriginatesinstatesinquarter0. 0.10 log(1 + Volume Originated) 0.05 log(1 + Volume Held) 0.00 -0.05 -0.10 -0.15 -0.20 -0.25 -3 -2 -1 0 1 2 Quarter 33

Table1 SummaryStatistics This table summarizes our loan variables and various bank characteristics at the bank-quarter level. All dollar valuesareshowninthe2010:Q1value. Bank-levelcontrolvariablesarelaggedbyonequarterforregressions. SeeAppendixforvariabledefinitions. N Mean Median Std. Dev. LoanVariables Volumeofjumboloansheld (unit: $M) 49,597 11.14 1.04 45.92 i,t Numberofjumboloansheld 49,597 17.26 2 68.64 i,t Volumeofjumboloansheld /Totalassets (%) 49,366 0.28 0.11 0.49 i,t i,t−4 Volumeofjumboloansoriginated (unit: $M) 49,597 14.82 1.28 61.15 i,t Numberofjumboloansoriginated 49,597 24.88 2 101.05 i,t Volumeofjumboloansoriginated /Totalassets (%) 49,366 0.37 0.13 0.71 i,t i,t−4 OtherVariables Totalassets (unit: $B) 49,597 6.84 0.88 22.33 i,t−1 Coredeposits 49,597 0.69 0.71 0.13 i,t−1 ROE 49,597 0.03 0.03 0.02 i,t−1 Homemortgages 49,597 0.21 0.19 0.11 i,t Statebank 49,597 0.59 1.00 0.49 i Z-score 48,200 196.00 153.46 165.92 i,t−4 Equityratio 49,366 0.09 0.08 0.03 i,t−4 Creditrisk 48,914 0.69 0.70 0.12 i,t−4 Elect 49,597 0.24 0 0.43 t 34

Table2 ElectionCharacteristics The table reports summary statistics for 323 gubernatorial elections held between 1990 and 2014 in 48 U.S. states. Allstatesinoursamplehavegubernatorialelectionsevery4years. NewHampshireandVermont,which haveelectionseverytwoyears,areexcludedfromthesample. SeeAppendixforvariabledefinitions. Electionvariables N I=1 Mean Median Std. Dev. VoteMargin(%) 323 15.84 12.67 13.40 Close 323 83 0.26 0 0.44 Wide 323 137 0.42 0 0.49 Termlimited 323 80 0.25 0 0.43 Newgovernor 323 172 0.53 1 0.50 35

Table3 JumboMortgageLendingaroundGubernatorialElections: Bank–levelAnalysis Thistablepresentsestimationresultsofthefollowingspecification: 1 Y = α +α + ∑ β Elect +X(cid:48)θ+ε , i,t i t k i,h,t+k i,t k=−2 wheredependentvariablesare(1)log(1+Volumeheld),(2)log(1+Numberheld),and(3)Volumeheld/lag(assets), where lag(assets) are banks’ assets from a year ago. Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in quarter t. Number held is the number of such loans. Elect (k=−2,−1,0,1) are set to one if a bank i’s home state h holds a gubernatorial election in quarter i,h,t+k t+k,andzerootherwise. Elect isthequarterleadinguptoanelection,thethree-monthperiodfromSeptember t through November of the election year. X is a set of time-varying bank-level control variables including size, home mortgages, core deposits, and return on equity. Note that all control variables are lagged by a quarter. Thespecificationincludesbankfixedeffectsaswellastimefixedeffects. Standarderrorsclusteredatthebank level are reported in brackets. *, **, and *** represent statistical significance at the 10%, 5%, and 1% level, respectively. SeeAppendixforvariabledefinitions. (1) (2) (3) log(1+Volumeheld) log(1+Numberheld) Volumeheld/lag(assets) Elect -0.078 -0.035* -0.006 t−2 [0.080] [0.021] [0.011] Elect -0.265*** -0.061*** -0.023** t−1 [0.080] [0.021] [0.011] Elect -0.445*** -0.109*** -0.034*** t [0.080] [0.021] [0.011] Elect -0.564*** -0.129*** -0.041*** t+1 [0.080] [0.021] [0.011] Size 0.787*** 0.428*** -0.119*** [0.040] [0.010] [0.006] Homemortgages 3.273*** 1.728*** 0.534*** [0.252] [0.066] [0.035] Coredeposits -0.608*** -0.197*** -0.137*** [0.219] [0.057] [0.031] Returnonequity 2.109*** 0.553*** -0.366*** [0.697] [0.182] [0.098] BankFixedEffects Yes Yes Yes TimeFixedEffects Yes Yes Yes Observations 49,597 49,597 49,365 R2 0.470 0.747 0.469 36

Table4 JumboMortgageLendingaroundGubernatorialElections: Bank/State–levelAnalysis Thistablepresentsestimationresultsofthefollowingspecification: 1 Y = α +α + ∑ β Elect +X(cid:48)θ+ε , i,s,t i,s s,t k i,h,t+k i,s,t k=−2 wheredependentvariablesare(1)log(1+Volumeheld),(2)log(1+Numberheld),and(3)Volumeheld/lag(assets), where lag(assets) are banks’ assets from a year ago. Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quartert. Number held is the number of such loans. Elect (k=−2,−1,0,1) are set to one if a bank i’s home state h holds a gubernatorial election in quarter i,h,t+k t+k,andzerootherwise. Elect isthequarterleadinguptoanelection,thethree-monthperiodfromSeptember t through November of the election year. X is a set of time-varying bank-level control variables including size, home mortgages, core deposits, and return on equity. Note that all control variables are lagged by a quarter. The specification includes bank×state fixed effects as well as state×time fixed effects. Standard errors double clusteredatthebank×statelevelarereportedinbrackets. *,**,and***representstatisticalsignificanceatthe 10%,5%,and1%level,respectively. SeeAppendixforvariabledefinitions. (1) (2) (3) Variables log(1+Volumeheld) log(1+Numberheld) Volumeheld/lag(assets) Elect -0.122*** -0.047*** -0.007*** t−2 [0.036] [0.008] [0.001] Elect -0.215*** -0.042*** -0.009*** t−1 [0.034] [0.008] [0.001] Elect -0.225*** -0.062*** -0.006*** t [0.036] [0.008] [0.001] Elect -0.113*** -0.043*** -0.009*** t+1 [0.037] [0.008] [0.001] Size 0.550*** 0.242*** -0.025*** [0.044] [0.015] [0.002] Homemortgages 2.876*** 0.902*** 0.048*** [0.228] [0.077] [0.013] Coredeposits 0.355 0.193*** 0.032*** [0.236] [0.073] [0.008] Returnonequity -0.259 -0.134 -0.049** [0.413] [0.117] [0.022] Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 207,535 207,535 206,544 R2 0.574 0.677 0.585 37

Table5 JumboMortgageLendinginHomeStatesvs. ForeignStates Thistablereportsestimationresultsofthefollowingspecification: 1 1 Y = α +α + ∑ β Elect + ∑ γ Elect ·Z i,s,t i,s s,t k i,h,t+k k i,h,t+k i,h,t k=−2 k=−2 +δZ +X(cid:48)θ+ε , i,h,t i,s,t wheredependentvariablesare(1)log(1+Volumeheld),(2)log(1+Numberheld),and(3)Volumeheld/lag(assets), where lag(assets) are banks’ assets from a year ago. Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quartert. Number held is the number of such loans. Z is a dummy variable equal to one for lending that was conducted in the bank’s home state and 0 otherwise. Elect (k=−2,−1,0,1)aresettooneifabanki’shomestatehholdsagubernatorialelectioninquartert+k, i,h,t+k andzerootherwise. X isasetoftime-varying,bank-levelcontrolvariablesincludingsize,homemortgages,core deposits,andreturnonequity. Allcontrolvariablesarelaggedbyaquarter. Notethatonlyelectionvariablesand their interaction terms are reported. The specification includes bank×state fixed effects as well as state×time fixed effects. Standard errors double clustered at the bank×state level are reported in brackets. *, **, and *** represent statistical significance at the 10%, 5%, and 1% level, respectively. See Appendix for variable definitions. (1) (2) (3) Variables log(1+Volumeheld) log(1+Numberheld) Volumeheld/lag(assets) Elect -0.146*** -0.059*** -0.006*** t−2 [0.037] [0.008] [0.001] Elect -0.215*** -0.054*** -0.010*** t−1 [0.035] [0.008] [0.001] Elect -0.124*** -0.043*** -0.000 t [0.037] [0.009] [0.001] Elect 0.085** 0.003 0.005*** t+1 [0.039] [0.009] [0.001] Elect ×Homestate 0.163** 0.076*** -0.003 t−2 [0.063] [0.014] [0.004] Elect ×Homestate 0.020 0.074*** 0.011*** t−1 [0.062] [0.015] [0.004] Elect ×Homestate -0.576*** -0.108*** -0.036*** t [0.067] [0.016] [0.004] Elect ×Homestate -1.105*** -0.261*** -0.078*** t+1 [0.073] [0.016] [0.004] Bank-levelcontrols Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 207,535 207,535 206,544 R2 0.575 0.678 0.587 38

Table6 BankCharacteristicsandSensitivitytoPolicyUncertainty Thistablereportsestimationresultsofthefollowingspecification: 1 1 Y = α +α + ∑ β Elect + ∑ γ Elect ·Z i,s,t i,s s,t k i,h,t+k k i,h,t+k i,h,t k=−2 k=−2 +δZ +X(cid:48)θ+ε , i,h,t i,s,t where the dependent variable is log(1+Volume held). Volume held is the volume of jumbo loans bank i either originatesandholdsorpurchasesandholdsinstatesinquartert. Zisthebankcharacteristicvariableofinterest. Incolumn(1),thebankcharacteristicvariableisStatebank,whichissettooneifthegivenbankisstate-chartered andzeroifnationallychartered. Thebankcharacteristicvariableusedincolumns(2)through(4)isHighrisk, whichissettooneifthevalueofariskmeasureisinthetoptercileofthedistributionintermsoftheriskiness. Three risk measures are employed: z-score, equity ratio, and credit risk. For z-score in column (2) and equity ratioincolumn(3),forwhichhighervaluesindicatelessrisk,Highriskissettooneifthevalueisinthebottom tercile of the distribution. For credit risk in column (4), for which higher value means higher risk, High risk is set to one if the value is in the top tercile of the distribution. All three risk measures are lagged by four quarters. Elect (k=−2,−1,0,1)aresettooneifabanki’shomestatehholdsagubernatorialelectionin i,h,t+k quartert+k, andzerootherwise. X isasetoftime-varying, bank-levelcontrolvariablesincludingsize, home mortgages, core deposits, and return on equity. All control variables are lagged by a quarter. Note that only electionvariablesandtheirinteractiontermsarereported. Thespecificationincludesbank×statefixedeffectsas wellasstate×timefixedeffects.Standarderrorsdoubleclusteredatthebank×statelevelarereportedinbrackets. *, **, and *** represent statistical significance at the 10%, 5%, and 1% level, respectively. See Appendix for variabledefinitions. 39

(1) (2) (3) (4) Variables Statebanks Z-score Equityratio Creditrisk Elect -0.120*** -0.066 -0.082** -0.012 t−2 [0.041] [0.041] [0.039] [0.045] Elect -0.164*** -0.211*** -0.178*** -0.159*** t−1 [0.039] [0.040] [0.038] [0.041] Elect -0.134*** -0.216*** -0.208*** -0.179*** t [0.041] [0.040] [0.040] [0.043] Elect -0.072 -0.151*** -0.101** -0.143*** t+1 [0.044] [0.043] [0.043] [0.045] Elect ×Statebank -0.001 t−2 [0.046] Elect ×Statebank -0.115** t−1 [0.045] Elect ×Statebank -0.204*** t [0.048] Elect ×Statebank -0.091* t+1 [0.053] Elect ×Highrisk -0.142*** -0.104** -0.192*** t−2 [0.053] [0.050] [0.047] Elect ×Highrisk 0.001 -0.115** -0.099** t−1 [0.052] [0.048] [0.045] Elect ×Highrisk -0.025 -0.054 -0.070 t [0.052] [0.051] [0.047] Elect ×Highrisk 0.100* -0.039 0.087* t+1 [0.055] [0.055] [0.053] Bank-levelcontrols Yes Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Yes Observations 207,535 202,131 206,544 205,303 R2 0.574 0.575 0.574 0.575 40

Table7 ElectionCharacteristicsandSensitivitytoPolicyUncertainty Thistablepresentsestimationresultsofthefollowingspecification: 1 1 Y = α +α + ∑ β Elect + ∑ γ Elect ·Z i,s,t i,s s,t k i,h,t+k k i,h,t+k i,h,t k=−2 k=−2 +X(cid:48)θ+ε , i,s,t where the dependent variable is log(1+Volume held), where Volume held is the volume of jumbo loans bank i either originates and holds or purchases and holds in state s in quarter t. Z is the election characteristics variable includingClose,Wide, and Term Limited. Close is an indicator variable set equal to one if the vote differenceinanelectionislessthan5%,andzerootherwise,wherevotedifferenceisdefinedasthedifference betweentheproportionofthevotesgarneredbythewinnerandthatreceivedbytherunner-up. Similarly,Wide issettooneifthevotedifferenceinanelectionismorethan15%, andzerootherwise. TermLimited isequal to one if an incumbent governor faces a binding term limit and cannot run for re-election, and zero otherwise. Elect (k=−2,−1,0,1)aresettooneifabanki’shomestatehholdsagubernatorialelectioninquartert+k, i,h,t+k andzerootherwise. X isasetoftime-varyingbank-levelcontrolvariablesincludingsize,homemortgages,core deposits,andreturnonequity. Allcontrolvariablesarelaggedbyaquarter. Notethatonlyelectionvariablesand their interaction terms are reported. The specification includes bank×state fixed effects as well as state×time fixed effects. Standard errors double clustered at the bank×state level are reported in brackets. *, **, and *** represent statistical significance at the 10%, 5%, and 1% level, respectively. See Appendix for variable definitions. 41

(1) (2) (3) Variables Close Widemargin Termlimited Elect -0.122*** -0.122*** -0.060 t−2 [0.037] [0.041] [0.041] Elect -0.190*** -0.256*** -0.104*** t−1 [0.036] [0.038] [0.040] Elect -0.221*** -0.268*** -0.142*** t [0.038] [0.040] [0.042] Elect -0.082** -0.182*** -0.052 t+1 [0.040] [0.043] [0.043] Elect ×Close -0.002 t−2 [0.059] Elect ×Close -0.107* t−1 [0.059] Elect ×Close -0.017 t [0.061] Elect ×Close -0.131* t+1 [0.071] Elect ×Wide 0.000 t−2 [0.052] Elect ×Wide 0.105** t−1 [0.050] Elect ×Wide 0.113** t [0.054] Elect ×Wide 0.182*** t+1 [0.059] Elect ×TermLimited -0.180*** t−2 [0.058] Elect ×TermLimited -0.317*** t−1 [0.058] Elect ×TermLimited -0.236*** t [0.060] Elect ×TermLimited -0.176*** t+1 [0.063] Bank-levelcontrols Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 207,535 207,535 207,535 R2 0.574 0.574 0.574 42

Table8 JumboLoanOrigination This table reports estimation results of the baseline specification (specification (1)) using alternative measures of loan variables. The dependent variables are (1) log(1+Volume originated), where Volume originated is the volume of jumbo loans bank i originates in state s in quartert, (2) log(1+Number originated), where Number originatedisthenumberofsuchloans,and(3)Volumeoriginated/lag(assets),thevolumeofsuchloansscaledby thebank’sassetsfromayearago. Elect (k=−2,−1,0,1)aresettooneifabanki’shomestatehholdsa i,h,t+k gubernatorialelectioninquartert+k,andzerootherwise. X isasetoftime-varyingbank-levelcontrolvariables includingsize,homemortgages,coredeposits,andreturnonequity.Allcontrolvariablesarelaggedbyaquarter. Notethatonlyelectionquartervariablesarereported. Thespecificationincludesbank×statefixedeffectsaswell as state×time fixed effects. Standard errors double clustered at the bank×state level are reported in brackets. *, **, and *** represent statistical significance at the 10%, 5%, and 1% level, respectively. See Appendix for variabledefinitions. (1) (2) (3) Variables log(1+Volumeoriginated) log(1+Numberoriginated) Volumeoriginated/lag(assets) Elect -0.079** -0.031*** -0.010*** t−2 [0.036] [0.008] [0.002] Elect -0.106*** -0.031*** -0.011*** t−1 [0.036] [0.008] [0.002] Elect -0.110*** -0.041*** -0.007*** t [0.035] [0.008] [0.002] Elect -0.019 -0.017** -0.010*** t+1 [0.036] [0.008] [0.002] Bank-levelcontrols Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 207,535 207,535 206,544 R-squared 0.606 0.725 0.644 43

Table9 AllLoans This table estimates the baseline specification (specification (1)) using a sample of all loans inclusive of both jumbo loans and non-jumbo loans. The dependent variables are (1) log(1+Volume held), where Volume held is the volume of loans, both jumbo and non-jumbo, bank i either originates and holds or purchases and holds in state s in quarter t, (2) log(1+Number held), where Number held is the number of such loans, and (3) Volume held/lag(assets), the volume of such loans scaled by the bank’s assets from a year ago. Elect (k=−2,−1,0,1) are set to one if a bank i’s home state h holds a gubernatorial election in quarter i,h,t+k t+k, and zero otherwise. X is a set of time-varying bank-level control variables including size, home mortgages,coredeposits,andreturnonequity. Allcontrolvariablesarelaggedbyaquarter. Notethatonlyelection quarter variables are reported. The specification includes bank×state fixed effects as well as state×time fixed effects.Standarderrorsdoubleclusteredatthebank×statelevelarereportedinbrackets.*,**,and***represent statisticalsignificanceatthe10%,5%,and1%level,respectively. SeeAppendixforvariabledefinitions. (1) (2) (3) Variables log(1+Volumeheld) log(1+Numberheld) Volumeheld/lag(assets) Elect -0.069*** -0.065*** -0.017*** t−2 [0.023] [0.008] [0.002] Elect -0.141*** -0.067*** -0.019*** t−1 [0.022] [0.008] [0.002] Elect -0.158*** -0.088*** -0.014*** t [0.022] [0.009] [0.002] Elect -0.106*** -0.061*** -0.018*** t+1 [0.024] [0.009] [0.002] Bank-levelcontrols Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 457,005 457,005 455,253 R2 0.628 0.711 0.558 44

Table10 RobustnessChecks Thistablereportsvariousrobustnesstestresults. Column(1)repeatsourbaselineregression(column(1)oftable 4)usingpseudo-electiondateswheretheelectionyearisrandomlyselectedforeachstatewitha4-yearinterval excluding the actual election year. Column (2) repeats our baseline regression excluding states which hold gubernatorialelectionsinthesameyearaspresidentialelections. Column(3)repeatsthebaselinespecification excluding loans extended to three large states, California, New York, and Florida. The dependent variable is log(1+Volume held), where Volume held is the volume of jumbo loans bank i either originates and holds or purchasesandholdsinstatesinquartert. Onlyelectionquartervariablesarereportedtosavespace. Standard errors double clustered at the bank×state level are reported in brackets. *, **, and *** represent statistical significanceatthe10%,5%,and1%level,respectively. SeeAppendixforvariabledefinitions. (1) (2) (3) Pseudo-election Excludingstatescoinciding Excluding Variables dates withpres. elections largestates Elect 0.025 -0.140** -0.124*** t−2 [0.030] [0.064] [0.037] Elect 0.111*** -0.268*** -0.205*** t−1 [0.029] [0.061] [0.035] Elect 0.031 -0.313*** -0.227*** t [0.029] [0.062] [0.038] Elect 0.010 -0.150** -0.121*** t+1 [0.029] [0.067] [0.038] Bank-levelcontrols Yes Yes Yes Bank-StateFixedEffects Yes Yes Yes State-TimeFixedEffects Yes Yes Yes Observations 207,535 170,536 184,842 R2 0.574 0.570 0.565 45