vaso 3008 Contents lists available at ScienceDirect Review of Economic Dynamics ELSEVIER wurm.elseviercomlocateired De-leveraging or de-risking? How banks cope with loss * Rhys M. Bidder®, John R. Krainer, Adam Hale Shapiro unter of Cambie, United Kingdom ° del Reserve Bank fan Fania Board of Coens, United state of Aner ARTICLE INFO ABSTRACT eae ‘We use variation in banks Toan exposure to industries adversely affected by the ol price eceived 27 Febuaty 2019 decines of 2014 to explore how they tespond to a net worth shock, Using granular data eee in revise for 26 May 2020, ceed in obtained under the Fed's stess testing programs we show that exposed banks tightened ‘edit on corporat lending and on mortgages that they would ultimately hold on thei Fidei TTC see However they expanded credit for mortgages to be securiized, patcuatly i ‘hore that ae governmentebscked. Thus, banks re-balance their portfolio so asco lower at their average esk weigh, rather than scaling back the sie f their balance sheet, as looking a at on-blance-sheet corporate or residential lending alone woud suggest, these result os Show the importance of taking a crose-balance sheet perspective when examining bank behavior in accition. in terme of the ultimate ‘credit channel to fms an household, epwons we show precisely how borrowers substitute to alternative financing when bank they Bank ending chanel inally botrow from tighten cet In shoving that sere was ultimately a minimal impact Ged hae fon borrowers overall funding, we provide a benchmark for crisis-peiod studies, which Banking typically find a powerful credit channel effect (© 2020 Elsevier Inc. Al rights reserved 1. Introduction Banks solve complex optimization problems when choosing their portfolio. Consequently, when a bank suffers a shock its response can be expected to be multifaceted. Nevertheless, many canonical models of bank behavior simplify the portalio problem to draw out broad intuitions. Perhaps the most prevalent intuition is that when a bank is damaged it will generally scale back its operations (Holmstrom and Tirole (1987)). Empitical studies inspired by this intuition have focused on a particular component of banks’ loan books (typically corporate lending) and show that damaged banks cut credit supply. ‘We investigate banks’ responses to credit losses induced by the precipitous decline in oil prices of mid-2014 and show that the standard intuition above is incomplete. We take a cToss-balance sheet approach, whereas much of the previous Dichine: The views expetsed io this acumen and alerts and enstons shouldbe regarded those ofthe authors and not necesaly those of che Federal Reserve Bank of San Fano the Peete Restve Bow of Goverbars ofthe Feel Keserve System Ths paper bas been seen to ‘tol asssane, We abo task sega pains a Carl Universi Warwick Unies, Bak f Eland. EC, Federal Reserve Bat l Coven ‘niverty College Landen, bay area Rance Werke Fed Sytem Mire Confrence erp. Mitchel Beli}. Chicago Ted UC Dav, ES Ata Hone Rong). {Computing i Eoramee and Fnsne (NC), CEBEA (er Viera Noger #3 Faroe lito, Fe Stern ery Conference esp. Crear Koch te Ermer rmisodcamacsk (RML Bide) johns krsine ogo UR Keine), earshot hong (AM, Shai ¥ ee kai ame Man (2098 Nashin and senate (2010), mene eal 2012) nd Chedoron-e (20) 1084-2025/0 2020 Elsevier ine. All ights reserved ese ce this ate in pes a: Bidder, RM, fa Deeveraging or essing? How banks cape with ls, Review of fzonomie Dynamic (2020) pang 01016) 264202006014,vyrepr009 a NE idler ea evi of roomie Byram one nee) oes ane literature examines the ‘bank lending channel’ within a given asset class* We find that banks do not uniformly reduce credit, supply fllowing 2 damaging shock. Instead, while some types of credit are contracted, others are exponded. In particular, corporate loans and on-balance-sheet mortgages are reduced, while mortgages to be securitized and shifted off balance sheet are expanded. This pattern of adjustment entails shift from heavily risk-weighted assets, against which banks must hold capital at a high rate to lower risk-weighted assets. The effect on total lending, total size of the balance sheet, and the degree of leverage appears ambiguous, What is unambiguous, however, is a pattern of ‘de-rsking’. We observe banks ‘making substantial reductions in their average risk weight rather than reducing their overall quantity of investments. We corroborate these empirical findings with survey evidence from the Senior Loan Officer Opinion Survey (SLOOS), which suggests that banks tightened terms of credit for the types of loans they wished to reduce (portfolio loans) and loosened ‘edit for those they wished to attract (securitizable loans). Interestingly, we find litte evidence that the de-tisking across asset classes was mirrored by de-risking within an asset class ‘he intuition that a bank might de-risk when damaged follows from basic portfolio theory. A shock to net worth may change the banks effective risk tolerance as i€ is shifted closer to some type of funding constraint (see Froot and Stein (1995) In this sense, 2 portfolio rebalancing reflects an updated solution to the risk-return tradeoff faced by the bank. (One way a damaged bank can pull back is in terms of loan quantities—the traditional focus of the bank lending chenne] literature. De-risking can be thought of as an alternative form of pulling back viewed from the broader portfolio perspective. We show that one can put the more traditional bank lending channel studies in a broader portfolc-level context and obtain important additional insights. ‘An additional contribution we make isto assess the ultimate impact of the ol shock on borrowers, We find no evidence of an operational ‘credit channel’ the term typically given to the indirect effect of a shock on borrowers, via their banks. Specifically, borrowers who were mote exposed to damaged banks do not appear to have made significant changes to their total Ioan balances or total assets ater the shock. Since our data provide the ientity of the borrowers, we can trace borrowers across banks, helping us to figure out why the effect is limited. The data reveal that borrowers facing worsening terms from exposed banks simply switched to other less-exposed banks within our sample. ‘The finding of a limited credit channel is in contrast to much of the literature that suggests such borrowers should be relatively harmed (see Khwaja and Mian (2008), Chodorow-Reich (2014), Acharya etal. (2014), Huber (2018), Murfin (2012) for example). Related work has examined the credit channel during periods of stress for the broader financial system.” While itis especialy important to assess how the banking system functions in such crisis periods (where welfare losses ‘may be more intense and where the role of policy is enhanced) it may also be the case that the strength or even the presence of the credit channel could depend on the crisis context. The availabilty of alternative financing from public debt and equity markets and other banks is likely very different from the normal period that we analyze. This study helps ‘quantify the extent to which this isthe case, providing a benchmark for papers that examine the effect of shocks in crisis periods.” ‘We use granular data ftom the FR Y-14 flings (Y14) obtained from bank holding companies as part of the Federal Reserve's Comprehensive Capital Analysis and Review (CCAR). This dataset includes a broader array of borrowers and loan types than those typically used in the banking literature, such as the Shared National Credits (SNC) data or Dealscan and spans a wider range of asset classes. Information about specific loans held on the Danks’ balance sheets allows us to construct the exposure ofthe banks’ corporate lending to firms in the oll and gas (ORG) sector prior to the sudden decline in oll prices in 2014. Our treatment stems from exploiting variation in this vatiable across banks, which implicitly induced variation in the impact ofthe price decline on net worth. For a variety of loan-types. we isolate the credit supply effects by using borrower fxed effects as in Khwaja and Mian (2008) which strips out possible credit demand effects. Any remaining endogeneity bias would arise ftom supply factors - specifically, factors associated with the banks’ decision to hold O&G loans. Our knowledge of the identities of the borrowers and banks allows us to include a wide range of controls at the bank and bank-borrower level to address these concerns ‘These observable bank characteristics are not correlated with ORG expasure, indicating that high and low exposed banks appear similar on average, Additionally, we instrument our exposure variable with banks’ branch locations from several ‘years prior to the shock (similar to Gilje et al. (2015)) with litle change in our results. Overall, our results are robust to a variety of alternative exposure measures, standard ertor constructions, and permutation exercises 2, Relation to literature Banks’ risk management as been studied both theoretically and empirically. root and Stein (1998) provide a theoretical framework motivating tisk-management practices. In applied work, researchers have examined how banks adjust their risk profile in response to shocks. For example De Jonghe et al. (2016) show that Belgian banks reallocated credit to less risky 7 This tern wa anally coined inthe cone f monetary policy tanemision (Bernanke ané Blner (9%) Kashyap ané Stein (92) and Sein Insights othe US sys * Eee Pp a, ML a DE Taga a7 SEU HW Wane Gap Wah TH RUE a zoe DE 20RD tip idtovgiaa0¥6jueazoz0 05014vaso 3008 Lider ea ei of eam Dynamic one (vee) ovens 2 firms during the financial criss. Liberti and Sturgess (2018) and Ongena et al (2013) also suggest some tendency for corporate loan substitution after a shock, towards borrowers that are less risky in some sense. However, the key difference ‘between our paper and these existing studies is that the latter show evidence of de-risking within asset classes.” In contrast, ‘we find that banks shift risk between asset classes, rotating their entire balance sheet away from high risk-welgited asset clases, In fat, we do not see much evidence of risk reallocation within the asset classes we consider.” ‘Our study touches on an existing literature that examines cross-balance sheet patterns, Using aggregate data Bernanke and Blinder (1992) show that banks shift from loans into securities after a shock. den Haan etal. (2007) also use aggregate data to assess responses to monetary policy shocks across banks’ portfolios. Abbass et a. (2016) use micro data on German banks to show that trading banks increase their holdings of securities whose prices had fallen during the financial crisis However, they do not assess overall bank balance-shet effects. ‘More closely related to our work are Chakraborty etal. (2018) and Chakraborty et al. (2019) (CGM). The former study looks at shocks that render mortgage lending more attractive and shaw that banks better placed to take advantage of this ‘opportunity were induced to make additional mortgage loans. Where the banks were more constrained in their ability to fund these loans, they freed up resources by making fewer commercial loans to fms. Thus, their paper documents an inter~ testing form of ‘crowding out that suggests that banks face real constraints on raising funds to support new lending. In the latter study, the authors observe that banks particularly exposed to Fed policy increased mortgage origination following the Fed's MBS purchases, while also reducing their CAI lending, The crowding aut effect however, was not observed following ‘Treasury purchases, suggesting a novel ‘origination channel’ (as they term it) rather than the typical ‘bank lending channel. While these studies feature cross-balance sheet effects they differ from ours in important regards and our main contti- butions remain novel. Our paper considers a shock tha, in the first instance, is co the bank whereas their impulses operate primarily as shocks fo (the relative desirabity ofan aser (mortgages). Ultimately both shocks have effects across the balance sheet but they are very different in nature and interpretation.” The analysis im our study assesses what happens when a ‘bank is damaged, taking as given the general attractiveness of assets. In this case, the cross balance sheet co-movements are not ‘crowding out, but an attempt to de-risk. Indeed, the de-risking insights we obtain arise from the granular nature ‘of our data and the fact that we observe off-balance sheet lending. ‘Our paper is also related to a number of studies that have assessed how banks change their asset liquidity following a shock. Comett et al. (2017) find that banks that had relatively liquid asset holdings at the start of the crisis were more prone to increase liquidity and reduce lending in response. Peek and Wilcox (2003) use data aggregated to the bank level to show the option to secuntize mortgages helps stabilize mortgage lending over the business cycle. This is consistent with the ‘work of Loutskina (201 1) and Loutskina and Strahan (2009) who emphasize the increased liquidity of loans via securitization in recent years. Holding more liquid loans, itis suggested, attenuates the lending response to a shock uncer the standard bank lending channel and allows the banks to shift away from traditionally more liquid assets. Our results show that the net ‘worth shock induced a smaller effect on balance-sheet liquidity than it did on overall riskoweighted assets.” The negligible effect on liquidity may be due to the fact that we are assessing a net worth shock as opposed to a liquidity shock, as i the ‘ase in these other studies, ‘We note that de-risking is not completely @ priori obvious, despite it being consistent with theoretical models such a5 Froot and Stein (1998). Indeed, following the work of Jensen and Meckling (1975) (see also Stiglitz and Weiss (1981), Diamond (1989) and Acharya and Viswanathan (2011) it is well understood that agency problems between creditors and Docrowers can lead to ‘isk shifting’ behavior, whereby borrowers are incentivized shift activities towards visky pursuits This could particularly be the case after an adverse shock that drives a bank closer to default. Qur paper confirms that, as ‘one might have expected, at the comfortably capitalized levels of banks in 2014 our shack. while painful, did not cause sufficient damage for the risk-sifting effect to predominate.”® Rampini et al. (2017) also note that degradation of net ‘oulinatonal bank. AS escssee below out use of éata frm the especially important ~ ané Klesynarae~ US. econoryfbankang sstem. posi, ins weld whee, lotessngy ew iperan to ilerentate betwen rk weighted and total ses fr negultry pupees, our teu provide important inghs. Even outsde ext haneng or regultory Irate, the imparance of stinging ask weigted ane teal sss als Mow ‘dered ee Fier (02) apd Da a a (08 far example) See alee Lane eal (2 reer sta (26) an Tela apd Kat (2017) Tor "he Fe implements ite porhares though the f-be-announce (TEA) mack ro tat aniston af mortace wae particulary inetd by the ircases = 40 eerste ted Wy 4 pacar ca of ae bring made re aarti, cola wo enter es #745 CGM themselves pu i the sok ote Banks ota pel negate seks cepa Brae aps shee the Banks the lendingspporanies that eee abstr ey rom commer! ane OF ete, fe paeble tht tere ae seene-aine net woth eects relecting pect pereption {he new lending eppatunes ~ ths Hey ave dea of Tune stk to assets ane otto the babk inte Ast sane Fu ‘te Trearnypureates ms chalrabory ea (015) somewhat srgnly seem t hve ite pat trough 3 pe net wort carne, nthe absence tte “ns mead bythe kd ta, wher iid tine ashing, US. Tau US government any et nd agency bcd MS Dut hardest in the ecen css to rece thet risk profile. meting the contrast with tacibana theories of ambling for resurrection. ese ce this ate in pes a: Bidder, RM, fa Deeveraging or essing? How banks cape with ls, Review of fzonomie Dynamic (2020) pang 01016) 264202006014,vyrepr009 4 A idee Review of room Byam. tons Yo aa wed balance 1070 Worth can also hinder risk reductions by intermediaries by limiting their ability to hedge if hedging is also subject to financial constraints and Laeven and Levine (2008) note the effect of eapital requirements on risk taking isin some contexts ambiguous and may depend on bank corporate structure, ‘ur results on banks’ changing of ther risk (weight) profile in response to the shock provide an important extension to the traditional empirical literature testing for the importance of net worth, as predicted by financial frictions models. While the cross-balance sheet aspect of our analysis takes us beyond the standard ‘bank lending channel’ Iiterature, our work does include analyses of bank lending. As such, for our corporate and mortgage loan analysis. we implement the fxed-effects ‘identification approach taken by Khwaja and Mian (2008) (KM). Recent papers that use the KM technique include: Schnabl (2012), Chodorowekeich (2014), Jimenez etal (2014), Iyer et al. (2014), Acharya et al. (20%), Cingano etal, (2015), Bottero ft al. (2016), De Jonghe et al. (2016) and Heider et al, (2017). 3. Data ‘We use data from the quarterly PR Y-140 and monthly FR Y-14M flings required of bank holding companies (BHC) with more than $50 billion in assets, The ¥14 has been fled since 2012 and contains detailed data on the banks’ balance sheet exposures. capital components and categories of pre-provision net revenue. The primary purpose of the filing is to help assess the capital adequacy of banks, in support of supervisory stress testing programs that the Federal Reserve is requited to run under the Dodé-Frank Act. Our samples of commercial (residential) loans consist of loans originated by 28 (25) banks and in Table 2 we illustrate their characteristics prior to the decline in oil prices. These variables will be used as bank controts in our commercial and residential loan analysis below, ‘The commercial loan data are from the quarverly corporate loan schedule in the FR Y-14Q. This schedule provides a dit register’ - loan-level information identifying borrower and lender and many characteristics of the loans themselves Since this paper Is one of the first ro employ the Y14 data, we provide some perspective on how much of the market for comporate lending we are able to capture. Table 1 shows that in 2014:Q2, just prior to oll prices’ decline. the commercial loans included in the Y14 totaled $1.17 trillion, of neatly 70 percent of the $1.69 trillion in commercial and industrial loans extended by the entire BHC population that files a FR Y-9C (Y9) report. For perspective, the total amount of debt ‘owed by U.S. nonfinancial corporations at that time was $2.39 trillion. Compared to other bank loan data sets used in the literature (eg, syndicated loans data from the Shared National Crecits Program (SNC) or DealScan}, the Y14 also captures loans that are typically smaller in size (down to $1m) and to smaller-scale borrowers. The FR ¥-14M filings contain loan- level monthly data on banks’ mortgage positions, including characteristics (but not the identity) of the borrower and the ‘mortgage. including their geographical location to fine level of detall From the perspective of our work, there are several especially important aspects to the data Ftst, we are able to look across multiple asset classes. Within the corporate loan sample we are able to distinguish credit lines and term lending. and within mortgages we ate able to distinguish those that ate securitzable (government loans or conforming to Fannie and Freddie's requirements). Second, for mortgages we know whether the loan is actually held on the balance sheet (a “portfolio loan’) or whether it has been sold off (typically to a GSE) yet remains in our dataset because the bank continues to service the loan. As we will show, if one looks simply at on-balance sheet data, such as in the commonly used Y9 reports, one can be led to believe there is a relative decline in mortgage lending. when in fact there has been an offsetting inerease in ‘mortgage origination for securitization, Third, our detailed knowledge of borrowers allows us to construct a measure of bank TT Tis poverlehniqe ees onthe (emanding requirement hat ane has acess to a ceitreies ope ast fore KM, eal ests pial ‘Wea {102) Peek anéRevengren (985) Stein and Rei (2000), Avera [2005), Aa (2006} lem eta (2012) Heweve, a aeted by Kowa nd an (2008 such sues suet fom the etc of separating ee deand a supply eet 6 shoks co Arms and Dunks. ely, one would ‘wish to evamine the effect fan belated shock te bank bale sets ut these shack ten ol silanes nfuenee frm: derand for ced (1987), atv 2008 vain and Sehr (2010), Chava and Purananga (211), creentene et (2014, tere elated cei demand and supply Sts mig cenfune eines othe ening Channe, "other papers tht ave wed hese novel ceglalry data ate Abdymamaoy (2012) petal ese), Black et ab 2015, Joba and Sarama {205 Epoube and Hal 200), Clem and Suma (2016) (al ete) and making Und use af the ceporste lene Ga ease the ise a pve Invrmaton by syne leaders. alarobcamaryan a (01>) Fr contemparanees work making eof ony the cps Jan da abe the (sng on banks acorn ote MBS bois * " HEE Pi ae ML a DE Tagg we ACER Wn Cape Wah TREE a ERE DRI DODO, ps ideevgt0a016jseaanan05.c14vaso 3008 Lider ea ei of eam Dynamic one (vee) ovens s ‘exposure to oil and gas drillers, which will be the treatment in our study. In addition, since we can track the same borrower across multiple banks, we can exploit within borrower variation that is orthogonal to possibly confounding credit demand effects." Knowing the borrower also allows us to track their substitution of alternative bank lending when switching banks Finally, the corporate loan data isnot restricted simply to syndicated lending (such as the SNC and Dealscan, typically used in the literature) and extends down to loan values of $tm."* In addition to regulatory filings, we use the Federal Reserve's Senior Loan Officer Opinion Survey (SL00S) to provide insight on banks’ lending standards. The SLOOS is a quarterly survey where banks report how various terms and conditions have changed since the previous survey. Importantly, rather than employ the economy-wide aggregate SLOOS responses, ‘we make use of the confidential bank-specific survey responses.'” We also use the Federal Deposit Insurance Corporation (FDIC) Summary of Deposits data in constructing an instrument for our exposure variable based on branch location, to be discussed further below. 4. The oil price dectine of 2014 Fig. |, shows that the oil price fell suddenly and significantly in the middle of 2014 Furthermore, oil futures prices clearly indicate that although a moderate pullback in prices might have been expected, the speed and extent of the decline ‘was a surprise, Indeed, much commentaty in the eatly part of 2014 was bullish on oil (see Hamilton (2014), for example) We note that our analysis does not require that the oll price be exogenous in the broadest sense. Of course it is possible (though pethaps not likely in this case) that there ate feedbacks from the state of the US. banking system to the broader ‘economy and (indirectly) to the oil price."* But that does not undermine our analysis since we exploit variation across the ‘banks in our sample, within the US. banking system. Performance of loans to firms in the O&G industry depend critically on the market value of oll. As shown in panel A of Fig. 1, the rate of loans identified as either past due, charged-off, or in non-acerual status (what we refer to a5 ‘problem loans} spiked in the O&G industry following the oil price decline. We define firms as being in the O&G industry ifthe bank reported NAICS codes 211, 213111, and 213112 (oil and gas extraction, drilling oil and gas wells, and support activities for oil and gas operations).” Fig. 2 shows that these NAICS codes are obvious outliers in terms of being ‘net makers’ of oil= related products, based on the ‘make’ and ‘use’ tables provided by the Bureau of Economic Analyss (BEA). The faction of ‘ONG loans that were in problem status climbed from 0.6 percent in 2014:Q2 to 10.4 percent in 2016:03. By contrast, taking all sectors other than O8G we observe essentially no tend, Restricting the sample to net users of O8G products (defined as firms in industries with net trade shares less than —0.1 percent - depicted on the left hand side in Fig. 2) we again observe no bvious pattern. Thus, there was apparently little scope for passive hedging of the shock within the loan book. Loans are debt contracts and given the already low rates of default expected on the typical loan, the upside from having lent to industries whose costs may have declined with the oil price do not offset the downside ftom exposure to ORC. In panel B of Fig. 1 we investigate whether difficulties in the O&G sector led to a liquidity drain for banks - a topic of recent interest given the important role played by losses in liquidity arising from backup facilities and lines of credit in the last crisis (see Cornett etal, (201)). We plot the average utilization rate af credit lines among the same set of sectors as in panel A. While there does appear to be some increase in the utilization rate of credit lines in ORG, relative to the other sectors, the difference does not seem to be particularly dramatic until peshaps the final periods. At the least, the alignment of the timing of any deterioration with the oil price decline seems much less clear than in the ease of loan performance. ‘Another concern might be the explicit hedging of borrowers’ credit risk using derivatives (such a8 CDS). However, Minton cet al. (2009) suggest that derivatives postions were largely held by banks for their dealer activities, rather than to protect against the credit risk of their oan book. In addition, we consulted the FR-Y9C and calculated that all but one of our banks had purchased CDS protection of notional value of around 1 pescent or lower of their total Ioan book and the number was approximately 3 percent for the remaining bank Thus, hedging of credit exposures via derivatives appeats to be limited, 4 theories of adverse and selection and moral hazard might lead us to expect in contexts in which banks have superior information on the underlying loans relative to the broader market (see Duffee and Zhou (2001) for discussions of banks ‘our atset snot a superset of syed Leng ata athugh the ¥14 oes peri information that lows one tient i ans are syodeaee ‘neal rca ofan nthe Desteanestbace eed by MurBy (27) and Chograeeh (0 ate fr bse fates atu Se coverage 7 he 08 may net be used or superisey prpases abd the canal Bales suvey fesponses may ely be used by Federal Reserve System ‘of lower pices by Sau abi ae carnmaniy-cied reasons for the ell ree deine. Te sae ofthe US ecenamy and bank health ge no ypally feature "Same aks repet SC oF GCS codes, bur the vat marty reer NAICS codes. We map the SIC an GCS te the NACS indus. See the eine data or for NAICS code 213112. For eath industry Wwe onstucte he Det tade with Industries 324 (pele refineries). 21 (and gas extraction}, 213171 ol and ling) a sare of aids’ ttl ue Seca, he at ade sare fr indy egal to A ue ese ce this ate in pes a: Bidder, RM, fa Deeveraging or essing? How banks cape with ls, Review of fzonomie Dynamic (2020) pang 01016) 264202006014,Lee 5 Eider eta Review of esac Dyamie on (os m0 A: Problem Loan Rates 5 : 086 Firms Non O8G Firms (086 NetusingFiems poe i Percentage ours Ps Ove, Cte Orin Nn cr Satie badaaadaacabhh ' to B: Credit-Line Utilization Rates wos ” O&G Firms “ oo i wnt 60. & i wl a t ry won PLES SL LS. PP LL PS. Note: ‘he re ashe tine presets the pie per bare fede Wast Tas Inert oi lls to the quartry level. Panel A: Blue bart represent the percentage of lana (Fe, oeadingwndrawn com- rites) tool and eas (OMG) frm (defined by NAICS codes 211, 219111, and 21112) that are either ast du, charged of or im nom acerual sats (refered to at “problems lune). Gray hate represent the perceatage of loans to all non-O1 frme that ae problem lane. Orange bars represent the perentage of Toana to all et ters ofthe O&G industy—defined as Sr in industries wih I than 010 fn et trade (Ge, make sinus tee wih the OG industry, ae dein by the 2007 BEA rake and we ables—that ae problem loans, Pasel Blue bare epevent the average ization rate on ine of ere (ined tla zed amount divided by total commited amcust) to OG firs. Gray bars ard orange bars represent ‘he sverage ulation rate for non-O4G ab net ur of the OL industry * oe "ig. 10 ces, Poblem tous, and Ciedt-LneUliation Rates (er nerputaton af the coos in the Hes} se seade leet the web version is ace}RM idler ea Seview of eoomic Byam oa ote) erase 2 GULL" KE ALE PELE GG OG OL Ge Noes: Bare represent net trade (that $8, make minus use) of ac industey with NAICS §24, 211, 213111 (petroleum refineries, ol and gas extraction, and driling ol and gas wells, reepectively), based on the 2007 Bureau of Beonomic Analysis (BEA) make end use Lables (heepa://ew bes. gov/sduatry/sa_ sansa? ata). BEA provides tnake and use raatrix tables for 389 industies, where each industry is mapped into & ‘hee, four, of five digit NAICS cose, NAICS eode 215112 (apport activities for oil an ga operations), inched in our meaeur of ol and gas exposure, is not separately included in the BEA input-output table ‘We omit vie (many) industries for which thie net potion e neghgible 2 et Tae wit NAICS 324, 213, and 2199173 a Shte of Total Oupu by Industry, informational advantage), In addition, recent work by Caglio et al. (2017) using Y14 and DTCC data, suggests that if anything, Danks appear to be net seles of protection on a substantial fraction of their credit portfolio. Based on these observations, our analysis will use the banks’ loan book exposures to OSG as our treatment variable. Variation in this exposure implies that some banks experienced larger shocks to their loan books than others when oi prices fell. We calculate ORG exposure as the share of each bank’s total committed exposure (including term loans as well both drawn and undrawn lines of credit) reported on the Y14 that is accounted for by lending to firms from the O&G industry over the period 2012:03-2013:04. We start our sample in 2012:Q3, the date at which the banks that experienced the oil shock period are all present in the Y14 data, We leave a butler before the period of the price decline to guard against picking up simple reversion to the mean and other concerns of endogeneity ‘The average O&G exposure is 5.9 percent and there is considerable variation across banks, implying a standard deviation ‘of 49 percent, Roughly speaking, a one standard deviation increase in pre-shack O8G exposure represents a 50 basis point increase in the overall problem-Ioan rate for the ‘average’ bank over the course of the post period.” is cleat, at least from the market's perspective, that O&G exposure was important. In Fig. 2, we group banks into ‘quartile bins based on theit pre-shock O&G exposute. Banks that were more exposed to the OSG industry, defined here as "7 The facen of O86 uss problem tas urease by 9.8 percent plats beoween 201402 and 2016.5. A one sandad deviation cease ‘exposure (0048) would cause 20.049 #98 =0.5 pecetae point Ieeae Inthe eel ate ease te this anil in pes a ide Re aL Deeveagng or desishng? How Banks ape wit loss. Reviw of Ecanemnie Dynamics (2030), pang 01016) 264202006014,vyrepr009 * A idee Review of room Byam. A: Stock Price 8: Market Capitalization fe - ° pat Bo 5 / ae 3 ED 2 Notes: This figure lutates the iapact of bank’ ol and gas extraction (OG) exposure to bank ries (panel A) and market capitalization (pane! B). For eath bank in each month, we alelate the deviation oft stock price or market capitalization froma 2014:M6, ‘The rd dashed line represats the average deviation for Vanks in the upper quarter in tera of exposure tothe OG sector, the blue dashed ine represents the Fie 5. stock rice ané atk capitation - dependence on ol expsute those in the top quartile, underperformed in terms of their stock prices and market capitalization following the oil price decline. The timing of the separation in performance is notable and is indicative ofthe repeated guidance given to investors at the time, Further confirmation of the importance of the impact of the oil price decline on bank condition comes from ‘media and survey evidence. The price declines of 2014 drew comment in many quarters and the implications for banks performance were broadly discussed (Alloway (2014), Jenkins (2014), Noonan and MeLannahan (2014). In fact, in bank surveys the importance of the effects was also emphasized: From the SNC 2016( 1 review ~ ‘The high level of credit risk stems {from large share of risky leveraged finance loans underwritten based on weak practices, and the significant decline in oil prices since ‘mid-2014 thet has reduced the repayment capacity of ebligors in the oil and gas (O8G) sector’. Als, ftom the January 2015 SLOOS Some survey respondents specifically noted their concerns about the oil and fas sector resulting from the sharp dectine in he price of oil as a reason tht they had tightened their lending polices. It is important to point out that the committed exposutes reported by the banks include both on and off balance sheet amounts, In particular, undrawn of partially drawn credit ines are reflected in the Y14 as loans with ulilized exposures lower than the committed exposure. The off-balance-sheet commitments can sum to particularly large amounts. Indeed, committed O8G loans make up a non-trivial fraction of both total on-balance-sheet loans and total assets, averaging 46 and 2.3 percent over the pre-period, respectively. In addition. if one alternatively expresses exposure as a fraction of equity Capital, the mean and standard deviation across banks are 19.1 and 21.0 percent, respectively, emphasizing that multiple banks had significant net worth riding on the performance of the O8G sector and, by implication, oil prices” 5. De-leveraging or de-risking? ‘We begin our results with a high-level analysis based on publicly available bank-level data on important components of the balance sheet - corporate and residential lending and agency MBS holdings, While the insights from these series are suggestive we will then probe deeper. making use of the more granular éata from the ¥14, augmented with the responses from the sLoos, 51 Descriptive analysis: FRY9C data In Fig. 4 we show how different components of bank balance sheets behaved around the oil price shock, using informa- tion from the FR-Y9C. This aggregate bank-level analysis is meant to highlight the direction and timing of Uhe impact, but ddoes not control for time-varying bank-specifie covariates or demand factors, For each bank, we regress the logarithm of the bank-level variable on a linear time trend over the pre-2014:02 sample. We then collect the residuals and normalize them by taking the log difference relative to the 2014:02 value. The unweighted average for banks in the upper (lower) quartile of expostte aze given by the red (blue) dashed line. below using loan exposure a a ration of equity and in sma resus : HEE Pi ae ML a DE Tagg we ACER Wn Cape Wah TREE a ERE DRI DODO, ps ideevgt0a016jseaanan05.c14‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens 4 cos Leans 5: Residential Loans Agency mB. OP a ‘dustrial June (CLI), portfolio renidetil loans, ad ageney-beked MBS. Al dat ae res th re-2014@2 sample We ele the residuals and then normaie them Dy tlang the diference relative to the 201402, We ther ort banks nt two groupeLanks in the upper quartile (rd-dashed lee) and bottoms ig. Bank balance set vals - dependence on i exposure Panels A and B show that the more exposed banks exhibited lower commercial and residential lending following the shock, in terms of total on-balance sheet loan exposures in these businesses. The timing, again. suggests 2 noticeable break around the period of the oil shock, While we make no attempt to control for demand factors heze, these patterns are consistent with the existence of a bank lending channel to the extent that the reductions in lending reflect 2 tightening of credit supply. Panel C shows the analogous plot for agency mortgage backed securities (MBS) and is essentially « mitrot image of panel 8 ‘We will delve into the connections between the banks’ lending responses and the propensity to securitize, but for now ‘we note that the ¥9 data relates to on-balance-sheet exposures such that any mortgages that are securtized will not show up in the balances used to calculate the series in panel B. Instead, they would manifest in MBS holdings. The patterns in these diagrams are therefore suggestive of a more complex set of responses than are normally attributed to the bank lending channel, Finally, we also note that movements out of corporate lending and (to a lesser extent) residential lending. towards MBS signifies @ substantial portfolio shift from high risk weighted assets to low - a form of de-isking 5.2, Econometric specifications The results of the previous section are only suggestive of a tightening of bank credit. In particular, they could entangle credit supply and credit demand effects and omit possibly influential controls. We account for such potential bias in the more granular analysis that follows. S21 Regression analysis ‘We emplay a fixed effects regression of the following form (Khwaja and Mian (2008) AY) = By + BiZi + BX +8 o ‘here Yi; is 2 loan amount between bank i and borrower j from the pre- to the post-shock petiod. 8 is a borrower fixed effect, Z; isthe bank's exposure to oil (the share of its loans accounted for by O8G firms in the pre-shock period), and X, are bank controls. We abstract from interactions of Z; with borrower characteristics although they will be used below. ‘As in Khwaja and Mian (2008) we define a particular loan concept for our empirical analysis. Our concept of a loan is 42 bank-borrower pair based on the underlying raw loan facility data so that, in examining the intensive margin, AYjy is the change in the mean indivicual loan committed balance within the pair, The mean is a simple average across quarters in the relevant subperiods and we require that there be a continuous bank-firm relationship starting before or during the pre-shock period and extending into the post-shock period, We focus on term loans as opposed to lines of credit to avoid complexities related to the decision of whether or when to draw upon the line, With term loans we have a clear picture of ‘hat funds were being loaned and when these funds were made available, However, the behavior and presence of existing, lines of credit will play an important role in our analysis below, so that we do sot abandon their information content completely. Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,vyrepr009 te A idee Review of room Byam. Bank Summary sass - average values peal shock Notperoring lon a (commercial las on? 1007 Eeuryfeal ses 3 me ae ‘Ombalance-shet tea loneTtal ets ms sa 19 Toll eharge-offeaalon-alance-shect loans 005 06 005, ag A og commercl loans (201203 to 201402) 52 - as lo residential loans (2012705 vo 201402) 35201 123 log fms (2012.03 to 291400 is 300 {Comte O46 leansTotalor-balance-shet leans 46a? 45 oat ‘over Daa are fam she VOC an ae akan she sverage fom the 207203 fo 2014) pened Lag oy Commerc oan Taga log vest lan, and Ia og MBS ae contd he ehange between 2012-92 snd 2914.2. Commaed ORC lean ate commer loan commitient mae tims in RACS codes 211, 213111, apd 219112. ake tom the ¥14 ta “omnites O46 leai/ot commited commercial loss Cemmied OR loarlal on-balnc-thet aan "Commitee (066 laneTetal sets” snd“Commited OW league alte athe uutery erage eer he 201203 fo 20108 Period, “Coniited O4G Toue/eal conned cetera lanl theta expec varable sed thoughout te ape For the extensive margin we track the existence of the bankfirm pair in the pre- and post-shock periods, allowing us to apply a linear probability model where AY, is replaced by an indicator capturing an exit (a relationship that had been present in the pre-shock period but was not present or came to an end in the post-shock petiod)** “The importance of the fixed effect in equation (1) is that it absorbs the component of the AY, that is common actoss In particular, it s intended to capture any general shift in the firm's eredit demand, common across banks. The fxed effect alleviates concems that the oil shock could induce both credit demand and supply shocks, This would lead to endogeneity bias in the regression and confound our measurement ofthe credit supply effect, captured in P.This isthe main coefcient of interest as it captures the identified effect on the lending relationship ofthe credit supply shock arising from the bank's relative exposure. ‘The interpretation of f is somewhat subtle. Consider, for example, the intensive margin regressions. The coefficient will indicate the association between a bank’ oll exposure and the typical change in the amount borrowed from that bank by the firm in question - the response of lending to the identified credit supply shock. However, one should not interpret the coeficient as indicating the amount by which a bank reduces its desired lending. ceteris paribus. Terms of credit are ‘multidimensional so that even if a bank is in some sense still willing to lend the same dollar quantity, it may nevertheless tighten credit along some other margin, such as price ot covenants. This could (and we argue below, did) induce a response from firms, such as seeking alternative financing, that would be encoded in any reduction in lending captured by A 5.22, Identification ‘The inclusion of firm fixed effects controls for any factors due to general shifts in 2 borrower's demand.” However, endogeneity bias could stil arise from other supply-side factors, The ideal treatment would entail randomly assigning O&G exposure actoss banks. Since our banks chose their exposure to the ORG sector. it is important to address potential endo- geneity First, note that we observe an extensive amount of information about the banks in our sample from the FR Y-9 data as shown in Table 2. This includes broad characteristics of the bank (total assets, recent trend in lending, and whether the bank ‘teal Gemand eet systemic supply elects 0: eet on lending co particular firms that ae cnmon ats banks. Thus By tle tothe arena eect fore lacton a boreners ip O86 intensive ae. " " HEE Pi ae ML a DE Tagg we ACER Wn Cape Wah TREE a ERE DRI DODO, ps ideevgt0a016jseaanan05.c14‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens 1 egesslon samples ~ fr an lous arace uber of ms bon ast) 78028 eet (lo naa 179 Ey te cess to exe finance " a " Number a bank teltioshipe pr-sece pig % x ve lean vanabler bt in pstsodk period oss oat (048) (048) Tots tse ae means abd sandard deviations pteatheses of arabes used lathe ala commercial fan {amples Alaa’ it dened ars baer pat The pre and por dt ate serged ever 201203 fo 20342 ‘ha 201s 01 to 201503, respecte, Accent exenal Bounce = danny To whether the enh {CUSIP ota tke: share feet hes lized = ft utlized corel loan eposue Inthe pre-petod {eora:e) te 201402; die by rol commnsted commercial lan exparce aver the pei is foreign or domestic). but also characteristics of the banks’ risk taking (return on assets, the non-performing loan rate and the charge-off rate) and capital positions (leverage ratio and tier-1 capital ratio)" We also abserve the banks’ balance sheet structure (loan share of assets, share of residential and commercial loans in the loan book, and the deposit ratto) which accounts for the banks’ preferences for certain asset classes,”” Given these observable bank characteristics, the remaining identification assumption in our empirical analysis is that a bank’s O&G exposure is not correlated with any unobserved factors influencing how it responds to a shock. That is, we are concerned about factors, unobserved fo the econometrcian that may be correlated with both the bank’s decision to hold O8G loans and its decision to change sts balance sheet in response to the shock ‘We address this potential endogeneity issue in two ways, ist, we examine whether O8G exposure is correlated with any observable characteristics If exposure is correlated with any observable bank characteristics then itis plausibly more likely toe cortelated with unobserved characteristics. Regressing ORG exposure on each of the bank variables discussed above indicates no statistically significant relationship with any of these observable vatiables. This gives some credence to the idea that O&G exposure is also not correlated with any unobservable factors ‘Second, we implement an instrumental variable specification using the banks’ 2008 share of branch locations in “O&G ‘counties’ as an instrument for O&G exposure (see Gilje et al, (2016) for a similar approach). Specialy, the instrument is the share of bank branches in 2009 that were in counties reporting any income from the O8G sector. This instrument will be strong insofar as certain banks were well placed geographically to originate loans to ORG and valid to the extent that the location of branches in 2008 is independent to how 2 bank would respond to a net worth shock in 2074. The ‘exclusion restriction relies on the assumption that branch location isan independent historical decision which subsequently ‘caused its corporate loan book to be more concentrated in certain industries. We cannot fully rule out that banks choose ‘branch location for the sole reason of obtaining corporate loan exposure in certain industries, im which case the instrument is no more valid than OG exposure. However, this seems unlikely given that business loans are just one of a multitude of Danking services offered at local branches (for example. automatic tellers, consumer loans, and savings accounts). Moreover, it seems even less likely that banks would choose branch location to gain mortgage exposure to households employed in certain industries. Yet we find that using branch structure in an instrumental variables regression yields roughly the same ‘conclusions for both corporate loans and mortgages that are retained in the bank portfolios, 2 ‘hese ae counties Whee the Hatt of nonfarm expeyee compensation attibutable tthe OMG Igasty by COUN. Kom 2013 BEA Table ABN pension of empleses by nest, wat greser tan er Rests wre rit fo mate sringnt deinen fe, tare > Sx ar» 10) The bateh current organizational frm . - " * Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,53. Corporate lending A idee Review of room Byam. Cae Bp ona 0586 ase) cern ommerci esnfoa leans 2576 ‘S04 (0230) 169 Foreign Bank Dummy o21e usr (0083) (cose) Fed eects Fim Firm controls Frm cone ‘Notes: The dependent variable is the change in lg termloan size. A quartery data for a given loan were claret singe pe snd port peri defined ae 20123 0201402 ane 201501 to 201513 respecinely. We removed om the spe lane ef the ol td ny and aa ws tested 6 ren ous for conseeatve ques houghout bth the pe ald post eid {he_ no voting and recenerne) O4G exposure censucted as ft conve las to O86 ree O46 exposure is ssttumented by the sac of the BHC bance: Io 2009 ib eounes ‘with any O8G sneome.Calumns 1 ape 2 ate run on te sample of fms tat Baro ten ans ‘ontals Al bank conirals sted are measure inthe pre-shock pried. Frm concals in clurans 3 ‘om mulple banks standard eves in parentheses are wo-way clusere@ atthe bank (28 banks vyrepr009 ‘The frst element of the balance sheet we examine is the corporate lozn book. In this component, at least, we find the basic message from the aggregate level ¥9 data to be broadly confirmed using the greater detail of che Y14. 5.31. Main results In Table 4 we show regression results on the intensive margin. The coefficient of interest, 1 (in the first row), is estimated to be negative and significant at the 1 percent level for both the fxed-effects instrumental variables (FEV, column 1) and fixed-effects regression (FE, column 2)” The point estimates under IV and FE are quite similar, indicating that unobservable supply factors ate likely not affecting the FE estimates. The IV-FE and FE coeificients imply that a one standard deviation increase in ORG exposure (5 percent from Table 2) is associated with about a 4 percent decline on the intensive margin of a 2 silar results ate obtained when Wwe weight observations bythe eof an te gta ou sus ae oot Beng den bythe eo the ean. randomly selected loan.” HEE Pi ae ML a DE Tagg we ACER Wn Cape Wah TREE a ERE DRI DODO, ps ideevgt0a016jseaanan05.c14‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens 2 ‘The bank lending anne commer loans, excessive mtg, Cae Spam os nae ani form) (0732) (aasd aan) lag Loz Commerc ans 0076" = 0062058 ote) (1988) 1324) : _ (0150) {0as4) 060) Commercial Leena ans O51" O46 ona (0168 (0161) (0.094) Foteg Bank Duny Soin -aoor 02 (oo) (oose 021) aed eects Fim Fim‘ Cancels contol je i columns 1-4, “ex a idan whether the An nt Re (ac industry. OC exposes contuted a Intl commited lune (o O46 fs died By toe {al commited cenumeral fans tn 2012 sod 2013. Coun 1s estinted under NV wbete ONG ‘xporare ir nstrimented bythe site a he BIC branes in 2009 in countes with any O8G ties-reion fed effects (aro 522 industies-3-cet NAICS codes, ad 9 resons—the ist cit Columns 3 and 4 of Table 4 show OLS results under two samples: mult-bank firms (ie. the FE sample) and the entire Y¥-14 sample, which includes single-bank firms. By comparing the OLS and FE estimators on the same sample we can assess the size of any bias induced by frm credit demand. If anything, the covtelation appears negative: rms that experience positive demand shocks appear more concentrated among banks that exhibit negative supply shocks. Thus, endogeneity bias from correlated demand, if there is any, may be towards attenuating the measured effect. In this case, ané assuming any correlation between credit supply and demand effects is similar inthe larger sample, we may expand our sample ané ‘ueat the estimated OLS coefficients with reasonable confidence as a lower bound (in magnitude) on the credit supply elect. In the larger sample, we again obtain a statistically significant coefficient on ORG exposire, although smaller in magnitude ‘than the FE sample (0.32) Table 5 shows results on the extensive margin, Similar to the results on the intensive margin, the point estimates ofthe FELIV and FE results are similar A 1 percent increase in O&G exposure is associated with about a 1 percent increase in the probability of ext (columns 1 and 2) with a similar coefficient using OLS (column 2). When (fourth column) we expand the sample to include single bank firms the effect is lower in magnitude and insignificant. 53.2. Exploring the mechanism - corporate lending The decline in the magnitude of the estimates when including single-bank firms in the sample suggests that alterna- tive financing opportunities may influence the results. To explove this. we run a specification with three interactions: a ‘dummy that fags whether of not the firm has access to external finance (has a CUSIP of a stock ticker), the pre-period utilization rate that captures the ability of the firm to draw on existing lines of credit, and the number of pre-existing bank relationships (in the Y14 sample), reflecting the costs of switching from an established banking relationship Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,vyrepr009 4 A idee Review of room Byam. ‘The bunk lendg haael~ cones loans. nteracton eects ae Bip ne" 07 1 . ato) (036) (aoe xt) (0131 (0206 reper # of Rank Rlsionshipe on8" (aso) fosesy ——(oaasy (0337 haa for agen ten fous were clase tase pe and post sock ped dened a 201203 to 201402 nd 201501 to 2015°5.respectvay. We remeved fom the ape last isin the OG ines. Data Joans to ell and gas frms éivded by teal commited commercal loans in 2012 ané 2013. Columns 1 and 3 are ‘gion Ged elles (aos 322 lndusies~3-dgt NAICS Codes abl 9 eles Lt dg of Bs ap coc {'¢ éumny indcaong wheter the fm borews fom maliple banks. External rane Ben» dur fot ‘epee i the prepeig (2012.2 to 201403) died by ttl commited cores loan expsute ort the pre-pend:pre-peed # of banks = numberof banks the fy asa as ane lan within Se preshock [ied enh of tionsip- yeas between tte of fit oriznatin and 201402 an antl nee pre {hock pesod ROA equlytoa ase NP sue et) explalfskeweghed sues ttl lous ase, ‘Garg ett euts, commercial iano leas teen usftl late depen labs, Ia “hange in logcommecl aan eefnee between 2012°Q3né 2914'Q0 36a freien bank dary Standad ‘Table 6 shows that most of these interactions are statistically insignificant.” However, consistent with the decline in ‘magnitude ofthe estimate for single-bank firms, there is a statistically significant postive coefficient on the number of pre- existing banking relationships in the exit regression. Tis suggests that firms with existing alternative banking relationships in the pre-period were more apt to leave their relationship with exposed banks. Since we know the identity ofthe firms, we can examine ifthe exiting frm formed a new relationship with another, perhaps less exposed, bank in the Y14 sample. To do 0, we run the fxed-eflects entry regressions ofthe type we ran in section 52 above, but on different subsamples of firms. In Table 6, we group the firms according to whether they experienced any relationship exit (column 1), did not experience any exit Column 2), experienced any decline in term-loan borrowing from a BHC (column 2), of did not experience any decline in borrowing from a BHC (column 4), Colurans 1 and 3 demonstrate that firms that experienced any reduction in lending (by any bank, for any reason) were more likely to strike up new relationships with less-expased banks, Next, we dig deeper by assessing whether the default risk ofthe frm also played a role in the probability of exit due to 8G exposure. Fig. 5 shows the implied coeficients of exit on ORG exposure by default risk. Specifically, we divide all firms in the ¥-14 sample into quartiles based on their probability of default (PD), Dased on the internal credit rating and, where available, PDs provided by banks in the YI4. We run this specification on two samples: the sample of multi-bank fms and single-bank firms"® As in the previous results, the figures show that the probability of ext due to exposure was larger for ul it shows thal this exit effect within multibank firms was mainly stemming from the lower-rsk firms. This result i likely due to the fact that low-risk firms have an easier time finding alternative financing than high-risk firms. n contrast to the evidence at the broader balance sheet level (showing that more exposed banks de- the tae, with mre We wee inducing these interatins ebe-y-one. However, bey ie the same gultave resus as pooling al 2 Spay we neat OSG expe withthe Gunes, nag he 2 ough Ah (FD quale cme catego beng eI gare y= B+ Bus 2s 4 Paz APD = PO) < FDP + 8,42, e1PD «PD, PO) — fad, oP = P0,) = H:X ey. We then we he faa metho to Const Standard ers or he impli ees onthe 2b (fi...) din Bi and th» Pe! guae HEE Pi ae ML a DE Tagg we ACER Wn Cape Wah TREE a ERE DRI DODO, ps ideevgt0a016jseaanan05.c14‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens s ee cane en repens, iene Milam tab tqy oe RgOKC Epa OnE Ow als aa 020 (013) (ais) (a6) (0767) (0500) rave 08 tor oa Oe oat borrowing (column 1. the change log tra-lean borrowing (olurn 2. he change in Joan browsing includes ll term loans aross all banks inthe ¥14 sample. Teton bar- ‘fs. quarter dat for ven loan were cllpseé tossing pe snd por 3 ‘od: eefne 2512 fo 2014103 and 20150) 0205-03, reper Frm conte Incdesndvtysimesregion Sued elects (eros 322 snder—Seigt NAICS cages fn 9 vegionswihe fst dig of asap cage) abd a dummy inating whether the em bores frm mulepe banks. Ag-expesue ihe leans weed O4G egeste {cos al ans by Ihe fr, Bak arly nue pre-sosk ped lenewehteg we Se lees af KOA, eeu set, NPL aes, Ue captalscweigte ast {otal ousttl ase, charge offal leans. comet lastea los, resent Secwen 201203 and 20142) and oe bank damn, Habit standard ees i purenieses 7p <010,"p-=-005, p< 00 Tables eg 086 Exposure os fed He Firm firm Rm rm fem Canele ‘wing frm fxed eet. Column 1 is cun onthe sample of fms that ented from atleast fected in soe. bank cones inne pe-shck ered HOA, egal sss, NP lasses, et capa RWA toa loaal ase charge ait lot, comme Toast loans, resent Lanta eas, deposit abit, lag ange i log Conmeral loans (eine betwen 20125 and 201402), ané a feegn bank eur banking relconsip between the pe- abd postsedk period. Han cones ae la total) and fim lev! for FE Eegressins and robust standard esos repertd far the OLS risked), this figure shows no evidence that the more exposed banks de-risked al the corporate loan book level. 1 anything more exposed banks were increasing their corporate loan risk due to safer loans leaving the loan book. These results imply that firms with low-switching costs (ie, low-risk firms with multiple banking partners) switched from more exposed to less exposed banks. The question than arises: what was causing these firms to flee? A plausible hypothesis is that exposed banks varied their tems of lending. inducing those firms to switch. To examine this hypothesis, Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,6 AE Bide eal eviw of esomic Dye one (eee) as = outtcient ° ‘Quartile of Default Probability [SFE (mutt:bank tems) ‘OLS (mult-bans firms) [OS (singie-bank tems) Noter: This gute pot estimates from the following regresion: Beity = 6+ Daa Zi + Buata PB" < PD, < PD + 8552, 01(PD < PD, < PD") + Hyd UPD < PD) + 8:%, +65. FE (eauicbank rms)” are estimates with fred eects over the sample of mult-bank fms “OLS (zalti- ‘bank firms)" reports estimates wiere with industryregion control on the sample of rul-bank fms, "OLS (esl bagh Sem” report stimates with ndsty:region contol onthe semple of sng bank rt. Tho fest quartile reports fi, the second quartile reports fa ~ a the thind quartile reports i+ ys, and ‘he fourth quartile reports +4) quartile fos, where standard cers ate constructed using the delta saethod Fi 5. InpedCoeces on OG Exporie By Haroner Def Rsk we look at the Senior Loan Officer Opinion Survey ($1008). The SLOOS is qualitative: respondents indicate whether lending standards have tightened or eased (‘somewihat” or ‘considerably’, or remained about the same. Importantly, the survey questions ask about lending standards for new loan applications, and thus can be interpreted as the willingness of the banks to supply credit. In Fig. 6 we plot an index of tightening on loan rates and covenant requirements for commercial and industrial (CB) loans.” We again average individual responses of banks in the top and bottom quartiles of exposure ‘The index shows average cumulative tightening, so a decreasing index depicts banks that have eased terms since the last survey, an increasing index depicts tightening since the last survey. ‘The banking sector as a whole was generally easing lending standards aver this period of time. However, on average, those with relatively high exposure to the oil sector (red lines in Fig. 5) stopped easing in late 2014 and began to tighten the terms of required loan rates and loan covenants.”* The results fram the SLOOS suggest that exposed banks shifted their «edit supply function in price relative to less exposed banks, 54, Residential oan analysis Our results using the corporate foan book give us interesting insights into how banks respond to shocks and are broadly consistent with the consensus in the literature on the bank lending channel: damage to banks leads to a tightening of credit A significant advantage of having access to other components of banks’ balance sheets is that we can also examine other types of lending. As noted in section 51, panel B of Fig. 4 indicates that residential lending among more exposed. banks lagged behind that of the less exposeé, In this section, we tease out the underlying drivers of this relationship, showing that this latter conclusion isin some ways misleading, 541. Mortgage regression results In the case of mortgage lending, we can no longer apply the borrower fixed effects approach at the borrower-bank level because we cannot identify a given household across different banks. However, this ine a level of grandlarty is likely unnecessary to avoid the aforementioned concerns with confounding credit demand and supply shocks for household 5 The BACs in he indenes are ones where we could merge the ban iD inthe ¥i4 data t he SLOOS daa Almost every ¥4 Dank was ase lected a the sioos 2 fot recet wed on the cle of covenants in the mechanisms by which bas ten lending in the sya lane rae see Chadeowe Rok and fac (2017. The ¥04 bas very gana information of loan quanties an lan pefermance but des fot contin interation on covenant ease ce thi are in pes a: Bidder, EM, eal Deleveraging a dein? Haw banks cope wth los Review a eonanie Dynami (2020), ps ideevgt0a016jseaanan05.c14eS Rider ea ei of eam Byram. ” ‘A: Loan Covenants xonur Ave Inde of Crna Teening elite to 20432 st Foreeeie B: Loan Rates pure Ave Inge of Comite Thong Relive 20182 ‘Notes: This figure ilusteates the inspact of « banks oll and gas extraction (OG) exposure to banks lending ‘andards for commercial loans. All data are from the Senior Loan Offer Opinion Survey (SLOOS). Ror ach bank we measured the cumulative amount of survey seeponses in which the response was a tightening of the epecied variable loan covenants (panel A) or loan rates (panel B)—relative to 2016Q2. The red dotted line represents the average for banks in the upper-quarter in terms of exposure to the O&G sector, ‘he Blue dotted line represents the average for Danke in the bottom quartie. ‘ig 6 Bank Lending Sandard for Crt Lan dependence il exposure.vyrepr009 te A idee Review of room Byam. The Bank lending chanel - sient las Nig) __Al__unbo Rename Al Nensumbe ane County Fed ets ~ (0716) (0338 osz0) (101s) 553) (tts) (190) t) Fred tes Couny County Couny County County County County County Panel B-o1s jams) (asal@stt) (tas) fos) ay (Lia) 88) Nurber af Observations 19828 wot 338882 sO os Fee Hee Sut Sure Sate stite Site Ste Sete Sate ‘ere The dependent arabe the change fn Tog redetl lane by bankecoiy Al quae data for lane war agreraed by Ba funky at thes eollaped toate pre-and pstsock peed. defined se 201203 t 201402 and 201501 to 20150, respecte. (GG exposes constructed stot come fut oo and gus ts dived by tal corte comune! ous in 2012 and 2073 {Coun is estiated uncer W. where OM expesure i isumente by the sar ofthe Bs branches in 2009 in cOUNes With ay O8G FIA and VA loans}. We iclude only loans erginated by the BHC (.e, we vereved matgages with loan source marked as cevtespondent, Servicing rights purchase. ane blk purchase). Bank cones include peshok period ROA, equyttal asses NPL otal we Ge 1 G- lag change i log residential loan defined between 2012-05 abd 201402} lag change it log MBS (defined between 2012-Q5 and 2014:02) ‘nea oer bank army County conel inde pe-petied leg poplaen. eg population gens. pevent veterans. og Nowig Ses, Sundaes in patentbses ae twoonay cused at the bank (28 bake i ol 04 ouy lee “pc010,"p © 008, ""P 0.01, borrowing, Since our mortgage data include borrower location data (zip codes) we aggregate to the county-bank level before applying the fixed effects approach to absorb county-level mortgage demand effects In Table § we show the results of our regression analyses. We pasttion residential loans into those originated and retained on the banks’ portfolios (portfolio loans’) and those originated but either sold of secusitized with only servicing rights retained (1ion-portflio loans). We use only those loans originated by the BHCs themselves, exchuding those obtained from correspondent relationships.” The different categories of loans are meant roughly to mimic the broad categories Included in SLOOS, For non-government mortgage originations that were retained in the banks’ portfolios, the banks are exposed to the credit risk associated with default. The results generally suggest that banks with greater exposure to the oil shock reduced their balance sheet exposure to such mortgages. For both the fixed eflects regression in panel A and OLS regression with county controls in panel B of Table ©, the estimated coefficient of -22 implies that a one standard deviation increase in oil exposure leads to a roughly -11.0 percent decline in portfolio lending. We estimate the balance sheet pullback to be stronger in the noa-jumbo category (column 3) than inthe jumbo category (column 2). Conversely. we see the opposite effect for the non-portfolio loans that wete originated and then securitized by the banks ln this case, banks exposed to the oil shock in the corporate portfolio increased their origination and securitization activity (columns 4 and 5), The effect is particularly pronounced when we limit the mortgages to non-jumbo loans, which are ‘mote likely to be guaranteed by the GSEs, The coefficient estimate of 3.0 smplies that a one standard deviation change in (O&G exposure on the corporate balance sheet is associated with a nearly 14 percent increase in securitization, Similarly, in column 6 of the table we see a significant postive response of changes in loan growth for FHA and VA (Le. ‘government’ ‘mortgages-loans likely to be securitized by Ginnie Mae. The coefficient of 8.4 implies a one standard deviation increase in (O&G exposure is associated with 2 41 percent increase in these types of loans ‘apeuring tends in the banks" merteate lending and MBS boléings (See Table fr a description of the controls. Results ae essencialy the same if the 2007 althaugh our breaer finding ofa pul-bick in parole mertzageleaing(s somewhat in tension. ° Eee Pp a, ML a DE Taga a7 SEU HW Wane Gap Wah TH RUE a zoe DE 20RD tip idtovgiaa0¥6jueazoz0 05014‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens " These results are consistent with the patterns observed in Fig. 4, Recall, in that figure (panel C) we observe a substantial increase in aggregate MBS for banks according to their oil exposure. We find no statistically significant change in overall residential loans (column 7). That is. the decline in portfolio loans (again consistent with 4, panel B) was offset by the increase in non-portfolio loans, implying no discernible change in total residential assets. The ultimate effect was a rotation of their balance sheet away from portfolio loans towards non-portfolia loans and, in turn, agency MBS, ‘5.42. Exploring the mechanism - residential ending Fig, 7 shows how the evolution of credit terms depended on banks’ exposure after the ail price shock” In panels A and B of Fig. 7 we show the cumulative tightening of terms for GSE-eligible and Government mortgages. We observe that there was a greater tendency on average to loosen terms among the more exposed banks. In contrast, in panels C and D we repeat the analysis for jumbo and non-jumbo, non-CSE lending and observe thatthe pattern is lipped-on average, the more ‘exposed banks seem to have tightened relative to the less exposed.”® The (relative) tightening of terms by the more exposed Danks is consistent with a decline in their lending as borrowers are less attracted to their loan products. GSE-eligble and ‘government mortgages are largely intended to be securitized and shifted off the banks’ balance sheets. Loosening of terms for these products is consistent with an increase in thei off-balance-sheet portfolio, 535, Portfolio implications Traditionally, studies on the bank lending channel give the impression that a shock to banks induces a uniform reduction in credit. However, we have shown that the bank’s overall response is more complex. More exposed banks pulled back lending, but this pullback was concentrated among loans against which they would be requited to hold regulatory capital at a relatively high rate.” In contrast, securitizable lending that was expanded - with relaxation in terms allowing, this ‘expansion, as reflected in the SLOOS. This rebalancing is consistent with the ¥9 data in that the MBS ané residential lending diagrams were essentially mistor images in Fig. 4 This offsetting pattern adds an important subtlety that might be missed ‘were one only looking at the residential lending data from the Y8. Banks are constantly solving complex optimization problems in their lending behaviors and there is no reason to expect that a pullback in lending will be uniform across all products, To get a sense of the degree of de-risking we run a simple exercise. We combine our estimated y coefficients from the fixed-elfects C&l and mortgage regressions with the pre-shock period loan balances and O&G exposure by bank. This yields an estimated impact of the oil shock arising from the banks’ relative exposure." We find that for the banks in the upper ‘quartile of exposure there isa predicted change of about -15 percentage points in the average risk weight among the asset ‘lasses considered in our regressions {in comparison with a pre-shock average risk weight of 71%), whereas for the lower quartile banks the predicted change is about -8 percentage points.” In terms of loan balances, the equivalent figures are 3 and +011 percentage points, respectively, reflecting the reallocation rather than a wholesale reduction in credit that ot ‘estimates imply, The above exercise assesses the degree of de-tisking within a portion of the banks’ balance sheet: commercial and residential loans. Of course, one would like to extend this analysis to include all elements of the balance sheet, though this would be beyond the scope of this paper. Nevertheless, we can obtain evidence that the patterns we have uncovered refect a general strategy by the banks. Thus, we regress the logarithm of bank-level variables on time-dummies interacted 75 Pots bein 2015-01 becase the 1005 quetins on mortgages changed a tat date. % ‘he GSB-eiple cacegory of residential mortgages neues loans that meet the uncewrtng suede. nlg lean mit amounts. ofthe GSES Fannie Me and fredeie Mae. The government eserory of redenal torgater inde las hat ate inva by the Feder! Howsne Admintaton ‘ualanteed by the Depart of etry Als or egiblee under govecment program, indading the US. Deparesen af Agele baie lah Drogas. The Quaid Martgage (QM) Nonjunbe.nensoseelgble cee of resdensal mortage Ince: lens That sty the standards fot 2 [uaiee mortage and have lan Dslaces tat sr blow te asn limit smunts set by the Cts bt ether da not mete whe SE anderen {Eudsines The QM jumbo clear of serena mortgages include Ions ta ay the andar fr guard ergs thar oan bleees Se above the fan Lint about se bythe CSE "Depending onthe LTV and the ater the prevailing risk weights ander the Bael standardize approach ovr the sample period ranged frm 35 The preci eal charg spre-shack ending» expsute» f 2» Note tha these cleans ate rested L the pss af the alte see for Which we hae etinate fy cet We apply can et fs se 3 cpa standards The assured sk weights for asset Gass kate as flows. Corea lens cae 2 TODE weigh, port orteates CY 2 50: wight, CSt-eigie onprl raigges cary 20% weigh ad gavertmeninsredmatgages ety 4 OF weigh Wil hee sumptions, {erage ek weiht marie fr bank ir defied 9y Tvatal The nee Bank's oan exposure in asst category ke Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,20 AE Bide eal eviw of esomic Dye one (eee) A: GSE Eligible Inet cme Teherng abet 2082 Non bo, Non 5 el : ouambe i 4 me i ae an Te ., ae Scorn | i- i Notes: This Sure illustrates the impact ofa bank's of and gat extraction (O4eG) exporure to bank lenng standards for mortgages, All dats ate fom the Senior Loan Officer Opinion Survey (SLOOS) questions about sqaliying mortgages A positive slope indicates tightening. For each bank we measured the curslative amount of survey reeponsos in which the response was a tightening of the specified variable relative to 2015.Q1, the fist date avilable, The GSE-eligile category of residential mortgages includes loans that rneet the underwriting pidelines,tncding Ioan Int amounts, of the GSE*Pannie Mae and Freddie Mac. The government category of residential mortgages includes Loans that are insured by the Federal Housing Administration, guaranteed by the Department of Veterans Aa, or originated under government programas, sacuing the U.S. Department of Agriculture home losn programs ‘The non-jambo, non-GSE- sligibe category of residential mortgage includes loans that satafy the standards for a qualified mortgage sd have loan balance that are below the Joan limit amounts art hy the GSB but otherwize do not mest ‘he GSE underwriting guldelines, The jumbo category of residential mortgage includes loons that stily the tandarde far a qulied mortgage but have loan balances that are above the loan limit amount ect by ‘the GSB, Tho red dashed Line represents the average for banks i the upper-quarter in terms of exposure to the 06 sector, the hue dashed line sepeesents the average fr banks ia the bottom quartile ig 7. bank Lending stanésés fr Mortgage dependence on oi exposure with 0&6 exposure. wwends.” me dummies interacted with bank controls, ne dummies, Dank fixed effects, and bank-specfic time 7 ed pesids in he regression are 2012. and 2014.2. mphingbank-speiic tine wends age pre-sock tends Hank cones inde pre shoes lar-eigied average eves of KOA quit) assets. NP ests, See captalskweighted asst total onsale. hag ofa loess, commer lossttal nas resident! lost losns depots abies and oeign bank dummy OG expovre was nace ta have iat stad deviation. The eskeweghed aes fate dened as tll tskeweigled ses dvded by itl wie Te leu ase alo ele a gue asets vcd by coral asset, wheel 555 ae deed as ash Rlengs- US. eases, US. government ageny de, nd ageny-bahee MES ease ce thi are in pes a: Bidder, EM, eal Deleveraging a dein? Haw banks cope wth los Review a eonanie Dynami (2020), ps ideevgt0a016jseaanan05.c14‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens a ‘As Ris Weighted Asset Ratio B Liquid Asset Ratio, SE OO OPW {oP POM ooo He C:teverage Ratio Diag Total Assets TS ee Sa PPP MP Pe er ee PPP POM PP Pw oP Pp “Thi gute lteter the impact of «bank's oil and gas extraction (OK) exposure to the tne sighted asst ratio (tral rake weighted atte divided by Sotl as), the quid asset ratio (defined a Tigui assole vided by total sees) the leverage rato (Ltal aes divided by equity and log total ase Liga atts ave defined och holdings, U.S. Treaturie, U.S. government ageney de, abd ageney-backed MBB. A data ate fom the BRCM. We rene the Iga interacted with OG expoure, time dummin interacted with bank controle, time dummig, bank fa feta, ad bs ape tine trends, Shown ate the coficets and 95 percest confidence bands of the toeficeatson the time dune interacted with OG 2012.98 an 2014.02, implying bank epeic time tens capture preshock period Srends, Bank eonole luce preshock period loe-welghted average levee uf ROA, es /ttal acts, NPL total ate, ter] ae etaard deviation. i sctmalied to havea mean of coding change in re weight implessente b) for 2015.1 fo enc hank wa spiced with te 202404 vale dard ceo ae hs (QL, the nskweightd ase ratio (panel ro by bank. To erect fr Bose I in ig 8 Response of Aveage Rsk Wright guy Ral, Leverage Rat, ad Tet Ase 90 Shack, In Fig. 8 we show coefficients and 95-percent confidence bands of the coefficients on the time dummies interacted with O8G exposure, where the dependent variable is the risk-weighed asset ratio (Panel A) the liquid assets ratio (Panel 8) the leverage ratio (Panel), and log total assets (Panel D). Panel A shows that there is a statistically significant movement downwards in the average risk weight across the balance sheet among banks who were more exposed to the oil shock. A ‘one standard deviation increase in exposure amounts to about 4 200 basis point decrease in the risk-weighed asset ratio, interpretable as the average risk weight, The pattern is consistent with our more granular regression evidence. In contrast, the relative behavior of total assets and leverage show no obvious or statistically significant pattern. Thus, we find evidence of de-risking, but litle evidence of de-leveraging in the traditional sense, By looking for coordinated actions across the balance sheet, shows that the bank lending channel is a much more subtle ‘concept than is typically allowed, Since the financial crisis there has been much demand for analyses of how banks behave ‘when damaged and for there to be a greater understanding of the detailed interlinkages within their balance sheets and change nk weight knplerented by Basel Ib 2015.0, he iskrighed ass! (pael Do 2015: fo each bak was sled wih Ws 200408 ‘al Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,vyrepr009 2 A idee Review of room Byam. their role as intermediaries - fo diferent sectors and classes of borrowers, Our de-tsking results speak to these important debates 6, (Weak) credit channel ‘The ability to substitute alternative funding sources is an important determinant of the ultimate impact of the oll shock fon borrowers. Our main focus above has been to tell primarily a ‘bank-level’ story regarding the banks’ de-risking behavior. ur second main contribution comes from analyzing how the banks’ decisions ultimately affected borrowers ie. the ‘edit channel. To assess the credit channel, we run a specification similar to (1), but aggregated to the level of the borrower Ay, BE + BIZ) + BSW) + BER, + Where AY, is the change (from pre- to post-oil shock) in an attribute ¥ of firm j% ZF is the weighted average O&G exposure of the banks from whom firm j borrowed in the pre-shock period. The variable is meant to capture how exposed the firm was to the credit supply effects induced by the oll price shock, via the banks with whom it was associated. W ate firm controls and X; are weighted averages of the chatactristics of the banks. 6! indicates whether ot not firms were systematically affected by being exposed indirectly to the oil shock. If we find that oil exposure typically induces a tightening of credit supply then one might expect a non-zero coefficient if firms are unable to substitute alternative Snancing. However, ifthe frm can substitute for any reduced financing or away from lending on worsened terms, then one ‘might not expect any effect. In Table 7 we show results for various dependent variables. The first two columns of the table relate to the change in firms! total term loan borrowing and total borrowing from Yi4 banks, respectively. These regressions show a statistically insignificant relationship with the aggregate exposure variable, implying that firms that were borrowing from more exposed banks inte pre-shock period had no net change in their term loans due to the credit supply shock. Ths result implies that firms were able to substitute simply ftom other Y14 banks. In columns 3-5 we move beyond the Y14 borrowing to look at additional balance sheet objects from Compustat. Here we see no statistically significant estimates, implying essentially no relationship between changes in total liabilities (or asses, or equity) and the aggregate exposure variable. Overall. in spite of the significant adjustments that exposed BHCS were making in the wake of the oil price shock, borrowing firms (at least in the Compustat sample) were apparently able to smooth out the effects of the shift in credit supply using other means." To pin down a particular channel for substitution, we isolate those firms that switched banks (among the Y4 filers). Returning to Table 8 (columm 5), we run our firm-level regressions with the dependent variable being a dummy variable equal to one if the firm exited atleast one banking relationship and entered atleast one banking relationship, Te coefficient on aggregate ORG exposure is positive and statistically significant, indicating that those firms more exposed to the credit supply shock were more likely to switch banks." ‘We also consider the ultimate effect on mortgage borrowing at the county level, Table 10 contains the aggregate regres sions. where we relate changes in county-level mortgage balances to the counties’ aggregate exposure (analogous to the corporate case). We assess three different categories: portfolio loans, all Y14 loans, and all lozns from the Black Knight- “MeDash Analytics database, which samples loans from the universe of lenders and not just those in the Y14, For all three levels of aggregation there is no statistically significant effect on loan quantity, again suggesting a limited broad credit channel and suggestive of 2 large degree of substitution to other lenders by households in exposed counties ‘There are of course a couple of caveats to our result of a limited credit channel. Fist, itis possible that firms who were unable or unwilling to switch banks and who chose fo maintain theit same scale of borrowing may possibly have suffered in other dimensions due to the worse terms they may have faced. Second, responses to a larger shock duting a systemic tightening might plausibly differ from our findings, as suggested by Chodorow-Reich (2014) and other crisis era papers. Nevertheless, it is interesting that in our case, which may be thought representative of a ‘normal shock in notmal times’ the degree of substitutability of credit seems substantial. Thus we provide a benchmack for studies focusing on mote turbulent times, 7 Sice raising capa requements ae in some sense comparable co 2 shack to et wrth our ress as eae te ascusions alc tole Flng n Keley (1s) Leven and Line [0% ane mare ent Lanéeigt ane Hepes 201). "2 Gr county for he matage resus traded ene etal. (20) suggest a bas adjustment (described the oalive descpove appengix) tothe fir level estiatedceficent based ot vera there appeate tobe an effec for firms with relatively large shares oftheir Drzowing capacity untappe (ess ace avalable on request) Eee Pp a, ML a DE Taga a7 SEU HW Wane Gap Wah TH RUE a zoe DE 20RD tip idtovgiaa0¥6jueazoz0 05014‘aEoe09 Lider ea ei of eam Dynamic one (vee) ovens B ee OR eT ORS Fed eter Site State State Tits The dependent arabe the change Tog ede leans By unt. Al qoute data for ns were agregatod iy county and then Celaiee 10a sgl pre-and pestshock pelioe. defined a5 2012.03 fo 2016:@2 and 20180) fo 2015:0. respecte In clit 1, we nce ony pete Tous, snd in eolran al V4 mortgage. Th eg-exposre i he loan weighted O4G expesre actos ll Danks ib level) ROA, ROR, NPL/total assets ier-1 capitalfRWA, total loa nan (toed between 2012-02 and 201803) ag change i log MBS (Getned between 2012.03 and 201402), anda focegn bask dans. ‘County contls neue pe-pesed ng poptstion lo ppultondes- Gian ouse vale log median total vale atin of Rausing with 3 we ‘her indvdval pint poverty. snemplayoet rte. pee. theses *p e010. p00, “p< 001 7, Robustness While anecdotal reports and the banks’ answers in lending surveys give support to our regressions and our overall story, inthis section we carty out a set of robustness checks. Fist, we consider the implications of defining our exposure variable 5a share of total on-balance-sheet loans, of total assets, and of equity capital (al as reported in the FR Y-SC) rather than 25 a share of total committed commercial loan exposure, The motivation for this (though Chodorow-Reich (201) also scales his exposure by loans, for example) is that the sizeof the loan book and capital postion of banks may diffe, implying noise nd possibly bias in our index. Table 12 shows that these changes do not substantially alter our findings. ‘A possible concern is that our regression specification contains too few clusters, leading to either “overfiting’ of the ‘luster-tobust variance matrix estimate or over-rejection of the critical values (Cameron and Miller (2015). There is no definitive measure of too few, but the general consensus is around 30 clusters ~ close to the number of banks in our sample. To address this issue, we re-assess our standard ervors using the wild cluster bootstrap recommended by Cameron and Miller (2015) Table 12 shows the p-values on for all of our firm-bank and county-bank pair regressions, under the wild cluster bootstrap. We also include in the table p-values under the amended definition of bank exposure, it which ‘we scale committed balances by bank equity rather than total committed C&l loan balances, The results coincide with our main results above using two-way clustering, showing strong statistical significance for the C8l, portfolio residential, ané -overnment residential lending, but noisier results on non-portolio non-government lending.” There is also a potential concern that the predictive power of our exposure variable was somehow a coincidence. To address this possibility we conducted a placebo test. Formally, we run a type of permutation test using each bank’s loan ‘exposure to other industries as placebos. That is, we test whether exposure to any other industry delivers similar results to expasure to the O8G sector. Analogous to the ORG exposure variable used throughout this paper, for each of the most prevalent other 49 industries (defined by 3-digit NAICS) we create a bank-specific exposure variable analogous to our O&G ‘exposure variable. We then run our main fixed-etfects specification 50 times - once replicating our original specification, and 49 subsequent regressions using the alternative industry exposures. To be conservative. we collected the associated ‘statistic based upon the standard errors obtained from the wild cluster bootstrap. In Fig. 9 we show the distribution of testatistcs obtained from our intensive and extensive corporate lending regressions. Cleary, sampling variability associated ‘with this test means that, even if our ONG exposure variable were the ‘tue’ exposure variable, it would not necessarily ‘emerge as having the largest t-statistic, However, we would at least hope for it to be in the tai. In fact, the results are "F Aiough catering should guard guint an individual San rving aur ress we aso an ur commer itesne and extensive Bae fe. (verge and standand Gevition of ~0.81 and 0.06, respectively, fr inensive and 0.84 and 0.08 fr the ext rezessions). " Pease te this ale n pres ddr Ri ea. De-eveazng of dersking? How banks cape wit los. Review of Eeanomie Dyas (2020), pang 01016) 264202006014,a AE Bide eal eviw of esomic Dye one (eee) Notee: This figure shows histograms of wiléchuser olstrap tates foreach of the 50 most common snduateie (deine by Slit NAICS code) in Uh commercial loan Yi dala. Analogous othe O&-G exposure ‘vate used throughout tht paper, foreach of the other 40 indie we create a banlcepeie exponure ‘rable, We then rahe ised fixed.eectsspecifeation 50 times once as ou ovigial specication, and 49 subsequent regressions, each time replacing Ok exposure with one ofthe 49 other industry exposures. atsties or cach regzesian were contrete ing the wld chter bolsrap zeaampling metho (cusered ‘by bark) ae deere i Casezon and Mile (2013) Ti sequence a tps as perfarned wie: producing ‘wo histograms, one forthe intnsio margin (Pane A) and one for tho extensive margin (Pane B). The ld chaser bootstrap Lettie forthe Ole exposire variable ie marked in each pane i. Wd ster Bootsy tate Femur Test by Indust. extremely stiong. In the intensive regressions, our exposure variable is a clear outlier in the left tail (where the left tai is ‘most relevant given the limited upside to a debt contract) and in the right tal in the extensive regressions Roughly contemporaneous to the oil price shock, US, banking regulators were in the process of phasing in new balance sheet liquidity requirements for many of the banks in our sample. The new regulations stipulated that banks maintain a Liquidity Coverage Ratio (LCR) whereby liquid assets could be used to offset possible funding outflows in a period of stress. The rule may have influenced the demané for liquid assets for certain banks, which include agency MBS. In the US. BHCS with assets over $250 billion needed to meet 80 percent of their liquidity requirements in 2015 (Le. in our post shock window). BHCs with assets above $50 billion but below $250 billion Were subject to a modified and less severe LCR requitement with a phase-in petiod in 2016 (Le. outside our post-shock window). In response. we construct an indicator variable to identify banks with assets over $250 billion and subject to the LCR phase-in in 2015, and then interact this vatiable with a proxy for exposure to the LCK inthe pre eil-shock period. This proxy Consists of a measure of high-quality liquid assets (Treasury Securities + Agency MBS) which is then scaled by a measure of possible funding outflows (Total Liabilities - Deposits). Thus, BHCs with large values for the proxy have ample liquidity, fr are more reliant on deposits (or both), and thus would be less constrained by the LCR. When we include these variables among the bank controls in our regressions we find only small quantitative differences in the estimates of the sensitivity of Tending to O&G exposure and no differences in statistical significance. Indeed, the coefficients on the LCR controls are not statistically significant in any of the mortgage regressions. Finally although itis less a robustness test than a useful re-expression, we recast our shock in terms of its relative Impact on bank valuations. To do so, we tun a two-stage specification of our model, fist regressing the bank’s change in log market capitalization (between the pre and post period) on the bank’s O&G exposure and then using this fitted change in valuation to explain the various changes in loans discussed above. In this way, we obtain an elasticity of lending with respect to a change in market capitalization attributable to the oil price shock. This allows us to map the oil shock into a common scale that might be transferrable to other studies, rather than leaving it in terms of exposure to a particular industry In Table 11 we repeat our main regressions under this two stage approach. We lose some power throug this procedure but qualitatively our results are confirmed and in most cases we retain statistical significance, Our results imply that a 1 ease ce thi are in pes a: Bidder, EM, eal Deleveraging a dein? Haw banks cope wth los Review a eonanie Dynami (2020), ps ideevgt0a016jseaanan05.c14vaso 3008 Lider ea ei of eam Dynamic one (vee) ovens 3 ato) (0209) Number af Observations yon Wee? " aio) (08) (oa Required (seconestage) 024024 nk Concle ve Ye ‘Notes: Al regresions were ran under two-stage least squares using ONG exposure as an log average query msde cptalzaon inthe pstpeied (2015 1 apd 25°03) ane pre-pecid (201203 ané 201402). The depencet variables are the Change log {atlas te tem oan was made forthe Ht ne ia the pose (ous Spane Aland the changes log reenial lone (ote by Phe BAC by coon {nd typeof loa panel BY Bank online pre-hack ped ROK equal = Set, NP ase ez eps. tll lost seb, eg sata lus, Commerc asf leane. resent last! Ian, dost labile ne Tesdetl lan dened between 2012.03 snd 201402) and lag ehatge in log NES (Gennea benwen 201203 and 2014.2) 25 cones, andthe comune an 1 Sens inca the ag change in log commercal fas defined between 2012 03 on SD14 2} acon Standard vest prntbves re two-way steed a the ak nd tun eve sp-=0.10,""p = 005. "~>p “00 percent hit to net worth leads to 2 0.90 percent decline in corporate lending on the intensive margin and 3 0.95 percent increase in the probability of exit. In terms of mortgage lending. the response is a 18 percent decline for non-government Portfolio Joans, 0.6 percent increase for non-government non-portfolio, and a 3.5 percent increase for government loans. Our aggregate regressions again suggest that the ultimate impact on exposed borrowers was minimal, relative to those borrowing from less exposed banks 8. Conclusions Banks damaged by exposure to industries adversely affected by the oil price decline made significant adjustments to their balance sheets. Banks de-risked their balance sheet by undertaking actions that reduced their average risk weight. We document a significant pull-back in commercial lending by exposed banks, both on the intensive and the extensive margins. Importantly, however. the balance sheet adjustments by banks exposed to the oll price shock were not confined to the corporate loan portfolio where the shock might be expected to have its most direct impact. Exposed banks tightened credit supply for those mortgeges that they would have to hold sn their own portfolios, while expanding credit for loans that were government-backeé and could be securitized. Overall, it appears there was a tendency of the damaged banks to de-leverage by de-risking, This is a new insight relative to the traditional “bank lending channel’ view which emphasizes 3 general e-leveraging and pull back in lending. ese ce this ate in pes a: Bidder, RM, fa Deeveraging or essing? How banks cape with ls, Review of fzonomie Dynamic (2020) pang 01016) 264202006014,vyrepr009 26 NE idler eta ew of roomie Byram one (nee) oan Fed fects Fem Pim fm rm Pim Residential Lane Tontehe ans New-Fonfobe tos ai ion RL Neri esque ox ozs 027 028028 oat 028 (a) ei ats (aaa) ees) aT Commitee 0sGiequty 065" 0342-69" nar nI6A™ amo Bank controls we we ves ee ve xe ed ets wry Coumy County ___ County County __ouny Covey joss eed Oy teal on-balnce shee loans, (2) commited ORG commer fans vided by tea sets and (3) commited (4G commercial leans vided by tal ery capt. The dependent arabes are te ebang leg cermelean sie (clumas 1 (costs 3 ad & panel 8) an the change lg testes! ous (opted Wythe BEC) by caus ad peo loa publ 1) Bank contol ince pre-stack period KOA equal ase, NPL sss, eT capa RWA, teal lotsa es, ‘dma. Residential loan segressions in include the ag change in log residential loan (defined between 2012.03 abd 2014.Q2) fn log change i log MBS dened Decween 2012.Q3 and 20%4 02) a cons, ad te comer loan repens incuge {he lag chang on log commercial loans ened between 2012-Q3 spd 201409) a cael tard err in pretest ‘Sstred tthe bank 28 bank In taal} and fim eel. "p01," < 808, ""9 001 Ultimately, the relative impact on borrowers who were particularly exposed to damaged banks appears to have been limited. The ability to substitute to alternative financing sources largely allowed borrowers to smooth through any credit tightening imposed by the exposed banks—although itis still possible that firms unable to switch banks easily may have sifered from worsened credit terms even as they maintained their quantity of borrowing, This likely reflects the health of the US. financial system in 2014 and the moderate size of the shock, Nevertheless, we provide a benchmark for studies of larger shocks in more turbulent periods, [Appendix A. Supplementary material Supplementary material related to this article can be found online at hitps|/éoi.org/10.1016))red 202008014. Eee Pp a, ML a DE Taga a7 SEU HW Wane Gap Wah TH RUE a zoe DE 20RD tip idtovgiaa0¥6jueazoz0 05014ees Rider ea ei of eam Byram. eS Robustness wld duster boststap peau. ‘ommned 086] Comme O67 commited Ct Laune Eau ‘toga size 005 ass ae 008 baat Residential ans onveroment Pri Laane ons 008 | Non-Gaverment Port Jabs) oats 500 Nea-Gevenment Poi Netejunbe) 074 bana Nonioveoment Non Porto ome ost Nonovermment Nen-Prtst(Nen-umbe) 0301 bias ‘oan 0x00 esis Tite Tis ble shows pralues of de ceecent on OMG exponte contacted wing te wild hse: baetstap resampling metho sere by ban) as esctbed In Caeton ane Mle (2975 ‘The dependent varable is repertee i column 1 Panel A shes pels fr the comer oan epson atthe bank level with fm fae eles, Panel 8 shows pote ar the tec {age reatesins atthe Diecut lel with Cou) Abed eects. We Lear seul unde to Fears of ONG exposure. commited O4G ans iid by oa comme Ca loans [or main ‘ipo variable, th coetiientsteperted i abe 5, and) at wel cated DL lst “Gedy tal equy feoeticent sepred ta Table 12) References ‘aha, os 2 ee gy ks cP we om ium of sh sat 204g er pt ae nM nen pied ae ‘a he wa oe puede . “ea elerpe de ane ind sc fr ea el th ef nny nein Te Anan Soe Re 19 9 Berrospde, [ML Edge. R. 2016. The Effects of Bank Capital Requisements on Bank Lending: What Can We Leaa from the Fos-cisis Regulatory Reforms. carb in a al ramp AD es fh np ofS Madge cas an AF 28 oy gg bon ren in ia stan an pane lah Bs cov ty, 20 My si adn, tl fo aan Gi tn 20 he hain lr need oa i Ea 98, 828 itm tana 3 Ps cn es om Rew sd le ekg Sl he ev De Jonghe, 0. Dewachter, H, Mule, K. Ongena, S. Sch ‘car Usted ye i ect 27 ak pa many niin ncn, Mey Ean 5 sate Da an a Gi oes mn mg a et tn 854 5 prt and seurzed commercial estate loans Journal of 6. 2016, Some Borowers ae More Equal than Others: Bank Funding Shocks and CredleLea 2 AE idler eta eiw of esac Dynamic one (see) onan pouhe 0. Hal A. 2016. Faymeat shock i (HELO) at te en af thew peta Journal of Economies and Busines 8, 131-147, Spee sue eulung Consume! Clee reat KA, tein, 1998, Rs management, cpl bude, and capil srctte pay or San nein an inerated approach Journal of Fong FT, Keeley Mi, 1968 Capt tegusion and bik cabin «nee Journal of Banking & Finance 13 (6) #82 {je EF. Laue, Sana, PE, 2015. Exprag bul: branch banking 2 fiance The Journal ef nance 7 (31138-1084 Grenson M. Mat A, Nae IL 2014 Bo rett Market Shot fet the Rel Economy? QasExpenmental [viene fe fe Great Recession nd alton, $0, 2014. The Changing Face ‘Od Markets. NBER Worlang Papers 20355. National Bureau of Economic Reseach, Inc. inca. leas, JA. 1953 Hae there heen 3 cpl enc in ankng? The eects on bank lending of rel ene mate canons ané bank apa ‘ori os af Housing Eeanemiee 3) 31-50. otnszom. 8, Tee, 1997 Pancalnermedation loanable fds. andthe eal ecto The Quartet Jural of Economics 112 (3) 653-591 Ishin Share 2010 bank lending dug the ancl sss of 2008 Jounal f Fnancal Ecorse 37 (04 319-298 Te 2007-8 Gane {ss esas om eiporte Sune Iyer. Petia, J. da Rochalapes 5, Scher A, 2014 Itrbnk uty rich ad the fr red ere evince from the 2007-2009 ee The Jenkins, 2, 2014, Falling ol price poses new threat to bunks. com (Onlin; posted 15-December-2014] Jessen MC, Meckling WH 197 Teoty of the fom: managerial Deva. agency costs abd ownership stucre Journal a Jeera, AR, Pv, Jo, Susi Salas, 2014 The Rel Efects af the Bak Lending Chanel: Unpusbed paper mene Gy Ongena Pet, I, Sutin |. 2012. Cree supply and monetary ply: Meaoyng the bank blancesbet chanel wi Jeheson KW. Seana 2015 Ea of he Line! Rehan of HELOC Rorrawers ating Payent Changs Werking Paper 201-072 Feral Reserve Board Fastyap, AK, Stan JC, 1953. Maney Poley aod Bask Lebding EK Working Papes 4317 Naina Bue of Econ Resear In aan, AK. sein JC, koe DM. 1992 Monetary peli nd cet con of eter) finince. The American conan Review 83 (178-88, wat AL a, A 2008 Tang the snpact of bak Lqudy socks: evidence Lieven, evneR, 2008 Buk governance, wegen and ck aking Jol of Lape rer D- Thesis, D201. anksExpsute fo neet Rae Rsk ane The ‘Sool of Eeonomie TSE LanevoigtTSegens) 2017 Finacial Regulation ina Quantatve Medel ofthe Moder fantng Stes Discusion pape. Uber Msg 206 The sanyo eet pp sha eee Kamo el el tet ole unl ad uave Loutlins E2017 Te leaf eurtizaton in bank iy and fing mangement Journal of Financial Eonamics 10 (3, 63-644 tuck $, Zimmermann, T, 2078 Employment Bfecs of Uncenventional Monetary Folic: Evidence from QF. Minton BA Suz K Wiareon, R200, How much do hanks une ce erates In hee lens ural of Finacial Services Reverch 35 (1). 1-3, Muti, 2612 The spplvade etacmnats foun contac sess The Jol of nance 7 (5). 1385-160 Noonan i, Melanoaban B, 2074 US banks hit by capo 6 Opec warns ef lng-erm low om (Onine: posted 23-Decemder 2015 Dnaena 5 ey fl van Haren N, 2018 Shoe Aes. Pain lame? bank rm Leve Gren an he international Tension of Fanci Shacks ara . 208 acl bank nana onsnts and facet eer Sane. The Journal of nance 63 (5) 2161-2182 Fee) Resenren #5. 1985 The capt! crneh ether 3 Borower ne lender be Jota of Mone. rend banking 27) 625-698, eek J Rsengen ES. 1987 The interstinal tanemiston of ancl shocks: the cave at Jopan Te American Seanomic Review #7 (4 485-65, eek) Wien JA, 203 secondly Margge luke, GSE, ad he Changing Cyeeay of Mstage Hows. Upablsee pape ere 5, 2073 Recent developments in monetary policy implementation. Remais 2 the Money Marktets of New York Unies, New Yar Cy. Schnabl P, 2012 The international rasmision of bank lgudity shocks: evidence (roms at emerging mathe. The Journal of Faance 67 (3), 887-952 Sten JC, 1996 An adversesleion noel ef bank sean Indy managemea wth unplaans fer the Wansmision of entry poy. The Kane oural ef Bonen 29 (3) 4ab-ate, sue JC Kasay, AK. 2000 Wha oa lie abervatins on bank ay about the tation of manetay poly? The Amtian Fea Review 90 Sige JE. Weiss. 881, reic aonng i markets with imperiec information. The American Eonamic Review 7 (3, 383-410,