How Costly were the Banking Panics of the Gilded Age?

Transcription

1 How Cosly were he Banking Panics of he Gilded Age? Benjamin Chabo Yale Universiy and NBER Sepember 25, 2008 Absrac: How cosly were he panics of he gilded age? I consider hypoheical insurance conracs based on observable New York Clearing House saemens. These hypoheical conracs would have allowed invesors o insure agains sudden deposi wihdraws. I esimae he cos of bank panics by esimaing he price of insurance implied by hisorical asse prices. Banking panics were cosly. The cross-secion of gilded-age sock reurns imply invesors would have been willing o pay a 4-4% annual premium o insure no more han a 3-7% deposi loss during banking panics.

2 How cosly were banking panics of he lae 9 h and early 20 h cenuries? A naural way o hink abou his quesion is o ask how much invesors would have paid o insure agains he consumpion loss caused by banking panics. Panic insurance did no exis bu i was possible o creae a real ime insurance conrac from he weekly balance shee saemens of New York Clearing House (NYCH) banks. I consruc hypoheical insurance conracs and use hisorical asse prices o draw inferences abou invesor marginal uiliy and esimae he equilibrium insurance premium had hese conracs exised. The resuls sugges invesors cared a grea deal abou banking panics and would have paid approximaely 4-4% per year o insure agains runs on NYCH banks. The Banking Panics of he Gilded Age The lae 9h and early 20h cenury business cycle was characerized by booming expansions puncuaed by financial panics and depression. In he era before deposi insurance deposiors raionally ran on banks whenever hey feared a sudden change in acual or perceived solvency. These runs combined wih asymmerical informaion abou he sae of individual banks ofen proved conagious and panic would emporarily rule he day. The NYCH aemped o minimize he informaion asymmery by requiring is member banks o publish weekly balance shee saemens. These saemens repored he average weekly and Friday closing values for each bank's loans, deposis, excess reserves, specie, legal enders, circulaion and clearings. The saemens were published in he Saurday morning New York Times, Wall Sree Journal and Commercial and Financial Chronicle. Bank saemens were carefully scruinized by invesors and Friedman Schwarz (963), Calomiris and Goron (99) and Wicker (2000) each provide excellen reviews of he facs and heory of lae 9h and early 20h cenury banking panics.

3 unexpeced changes in liquidiy could se off a sock marke rally or decline. Figures II-VI graph he NYCH loans, deposis, loan/deposi raio and NYSE minimum call money rae during he major panics of he gilded age 2. I include he minimum call rae because i is an excellen proxy for he marginal cos of excess reserves. Brokers and banks could lend or borrow agains securiy collaeral a he NYSE call money pos. Typically a borrower could borrow up o 80% of he marke value of he pledged collaeral. The rae of ineres charged varied wih he volailiy and liquidiy of he collaeral. The minimum call rae was always equal o he rae of ineres charged on loans wih long-erm governmen bonds as collaeral. As he name implies, call loans gave he lender he righ o call in he loan a any ime. The borrower of a call loan signed he pledged securiy ino he name of he lender. If he lender called he loan and he borrower was no forhrigh wih he money he lender could sell he collaeral o saisfy he obligaion. If he collaeral fell in value he lender could issue a margin call and demand he borrower raise his collaeral back o 80%. Thus lenders suffered parial defauls only when he borrower defauled and he collaeral declined by more han 20% in a single day wihou he lender being able o liquidae. Call loans on governmen bond collaeral were for all pracical purposes defaul-risk free. Despie he righ o call a any ime a call loan did commi he lender s money for a brief period. Even in he even of a collaeral sale he lender would no receive his cash unil he sale cleared 3 days afer he rade dae. The call loan rae herefore refleced he marginal cos of a bank holding excess reserves in heir vaul as a defense agains bank runs raher han loaning i riskfree for a minimum of 3 days. 2 There is no consensus on exacly wha consiues a gilded age banking panic. However, Sprague (90), Miron (986), McDill and Sheehan (2007), Calomiris and Goron (99) and Friedman and Schwarz (963) largely agree ha 873, 884, 890, 893 and 907 were years of major banking panics in NYC. See Table 3 in McDill and Sheehan (2007) for a summary of he agreemens and disagreemens on banking panic daes.

4 Figure I Call Rae on US Gov Collaeral a he NYSE /6/66 /6/69 /6/72 /6/75 /6/78 /6/8 /6/84 /6/87 /6/90 /6/93 /6/96 /6/99 /6/02 /6/05 /6/08 /6/ /6/4 /6/7 /6/20 /6/23 Figure I graphs he minimum call rae over our sample period. The call rae is generally quie low. I rises during periods of business expansion when banks wish o leverage heir balance shees and he marginal benefi of excess reserves is high. The call rae also spikes during panics when banks are desperae for reserves. Because i increases during boh panics and booms he call rae by iself is a poor candidae for an insurance conrac. However, knowledge abou he call rae and bank balance shees can be combined o consruc a real ime derivaive ha reflecs banking panics.

5 Panic of Loans per $00 Dep. 40 Deposis (Jan873=00) Loans (Jan 873=00) Call Rae (righ scale) Jan-73 Mar-73 May-73 Jul-73 Sep-73 Nov-73 Jan-74 Mar-74 May-74 Jul-74 Sep-74 Nov-74 Jan-75 Mar-75 May-75 Jul-75 Sep-75 Nov-75 The series breaks during and 5 weeks afer he panic of 873. This panic resuled in he closing of he NYSE and he suspension of reporing requiremens by he NYCH. Panic of Loans per $00 Dep Deposis (Jan884=00) Loans (Jan 884=00) Call Rae (righ scale) //84 3//84 5//84 7//84 9//84 //84 //85 3//85 5//85 7//85 9//85 //85 //86 3//86 5//86 7//86 9//86 //86

6 08 06 Panic of Loans per $00 Dep Deposis (Jan890=00) Loans (Jan 890=00) Call Rae (righ scale) //90 3//90 5//90 7//90 9//90 //90 //9 3//9 5//9 7//9 9//9 //9 35 Panic of 893 Loans per $00 Dep Deposis (Jan893=00) Loans (Jan 893=00) Call Rae (righ scale) Jan-93 Mar-93 May-93 Jul-93 Sep-93 Nov-93 Jan-94 Mar-94 May-94 Jul-94 Sep-94 Nov-94 Jan-95 Mar-95 May-95 Jul-95 Sep-95 Nov

7 Panic of Loans per $00 Dep Deposis (Jan907=00) Loans (Jan 907=00) Call Rae (righ scale) Jan-07 Mar-07 May-07 Jul-07 Sep-07 Nov-07 Jan-08 Mar-08 May-08 Jul-08 Sep-08 Nov-08 Jan-09 Mar-09 May-09 Jul-09 Sep-09 Nov-09 Afer a carefully examinaion of he balance shee series some common paerns emerge. The call rae increases and he level of loans and deposis decline sharply during panics. The loan o deposi raio ineviably spikes during panics and hen falls afer panics subside. These rais are no surprising. Banking panics are defined by sudden wihdraws of demand deposis. The loan o deposi raio iniially rises because banks are unable o conver illiquid loans ino reserves a he rae of deposi wihdraws. The raio hen falls as banks curail lending and build excess reserves unil fear subsides. Daa The figures sugges ha derivaives based on he change in deposis, loans or call raes may have allowed invesors o insure agains banking panics. To es his conjecure I consruc ime series of NYCH deposis and loans sampled every fourh Friday beween Jan 866 and December 925. The 28-day sampling frequency was seleced o correspond wih daes for which I have previously colleced he call rae of money and

8 price, shares ousanding and dividends of every NYSE sock. The sock daa allow me o compue he marke value and 28-day holding period reurn for each sock on he NYSE. Using Sock Reurns o Draw Inferences abou Marginal Uiliy and Banking Panics Sock reurns conain a grea deal of informaion abou invesor s consumpion. When producion (wealh) unexpecedly falls, risk-averse invesors wish o smooh consumpion by selling asses. If he producion (wealh) decline is idiosyncraic, some oher invesors wih unexpecedly high wealh will buy asses. Thus idiosyncraic risks can be shared wihou alering prices. If aggregae producion unexpecedly declines risk-averse invesors will ry o borrow and smooh consumpion by selling asses, bu in he aggregae we canno all borrow! If aggregae producion declines someone mus consume less. Sock prices mus herefore fall (and expeced reurns rise) unil invesors are willing o consume less. Sock reurns herefore reflec changes in aggregae consumpion. We can use his insigh o draw inference abou aggregae consumpion from sock reurns. Before we price he hypoheical gilded-age securiies i is useful o consider a simply discree asse ha pays $X p if a banking panic occurs nex period and $X np oherwise. The asse is an insurance conrac so X > $ X. If his securiy rades in a $ p np marke where invesors face he same price o buy or sell he price of he securiy mus saisfy P = E[mX ] or P = π m X + ( π ) m X () p p p p Where π p is he expeced probabiliy of a banking panic and m is he marginal uiliy of money in each sae. () is derived from he firs order condiion of invesors who purchase or sell he securiy unil he expeced marginal gain from buying E [mx ] equals he marginal cos P. Nex consider a nominally risk free asse ha pays $ in boh he panic and no panic saes. This asse will rade a P = E[m]. The gross risk-free rae is herefore equal np np

9 o R =. If we divide boh sides of () by P we can express he expeced excess f E[ m] reurn of he insurance conrac as a funcion of he covariance beween he insurance reurn and marginal uiliy. E[ R] R = f = E[ mr] = E[ m] E[ R] + cov( m, R) R f cov( m, R) (2) Insurance conracs pay high reurns when imes are bad and he marginal uiliy of money is high. cov( m, R) is herefore posiive and he expeced excess reurn of an insurance conrac is negaive. Equaion (2) provides a esable predicion abou he cos of banking panics. If R is he reurn of any variable posiively correlaed wih banking panics and banking panics were cosly in erms of uiliy, he expeced excess reurn of R should be negaive. Securiies based on changes in deposis, loans or call raes appear o be excellen candidaes for insurance conracs. An insurance conrac should pay a high rae of reurn in he saes of naure we wish o insure agains and a low reurn oherwise. Consider he following hypoheical securiies.. A series of 28-day cash-seled fuure conracs ha rade each observaion dae and pay $*(change in NYCH aggregae deposis). 2. A series of 28-day cash-seled fuure conracs ha rade each observaion dae and pay $*(change in NYCH aggregae loans). 3. A series of 28-day cash-seled fuure conracs ha rade each observaion dae and pay $*(change in NYSE call rae on governmen collaeral). An invesor would be able o insure agains changes in balance shees or call raes by buying or shoring hese conracs.

10 Figures II-VI sugges our hypoheical fuure conracs are correlaed wih banking panics. Were banking panics correlaed wih he marginal uiliy of gilded-age invesors? In oher words, were banking panics cosly in a uiliy sense cov( m, R) > 0, beneficial cov( m, R) < 0 or neiher cov( m, R) = 0? To answer his quesion we need a es of he null hypohesis ha cov( m, R) = 0. Where R is he reurn on one of our insurance conracs. If we could observe a ime series of m and R a naural es would be o esimae a regression of m on R m = α + β + ε (3) R The marginal uiliy of gilded-age invesors, m, is unobservable, however. In mos cases an unobservable LHS variable is a considerable burden when esimaing a regression! In he case of many asse reurns, however, we can esimae α and β from (3) and he momen resricions P = E[mX ] and he law of one price. The law of one price requires he same m price all asses. Therefore he unobservable m ha prices our hypoheical insurance conrac mus also price observable gilded age NYSE sock reurns. We can herefore esimae he regression of unobservable marginal uiliy on our hypoheical fuure conracs via GMM by choosing α and β o bes saisfy P = E[mX ] for observable asse reurns. I esimae (3) via GMM by choosing he regression coefficiens o bes price 5 NYSE call-rae bea and 5 size-sored sock porfolios. The size-sored porfolios were formed by assigning socks o quiniles based on marke value a he beginning of each 28-day period. Value-weighed reurns are compued and socks are reassigned each period based on updaed marke values. The call-rae bea porfolios were compued via he following wo sep procedure.. For each ime period esimae he following regression from he railing 3 years of 28-day daa:

11 R i callrae = i + β i R + β i2 α R + ε marke i 2. Wih bea esimaes in hand, assign socks o quinile value-weighed porfolios based on heir railing call-rae bea The resuling 0 size and call-rae sored porfolios should exhibi cross-secional differences in reurns and sensiiviies o banking panics and business cycles. Table I repors he regression coefficiens and -sas from GMM regressions of gildedage marginal uiliy on he deposi, loan and call-rae fuure conracs. GMM regression: Table I m fu = + β R α + ε Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) (3) Fuure Conrac: $*deposi growh $*Loan growh $*change in call rae bea sa Changes in deposi growh, loan growh and he call rae were no significanly correlaed wih he marginal uiliy implied by asse reurns. The regression coefficiens on loan and deposi growh have he expeced sign bu he effecs are saisically indisinguishable from zero. Were bank balance shee variables and consumpion uncorrelaed? Before we jump o ha conclusion a word of cauion is in order. Changes in marginal uiliy reflec unexpeced changes in consumpion. If changes in deposis or loan were predicable hese changes would already be refleced in invesor s consumpion decisions and asse prices. In fac, deposi and loan growh were predicably seasonal during he gilded-age 3. 3 Miron (986)

12 Roughly 30% of he ime series variaion in loan and deposi growh can be explained by a simple auoregression. This predicabiliy inroduces measuremen error in our independen variable. Table II repors regression coefficiens and -sas from GMM regressions of gilded-age marginal uiliy on he unexpeced change in NYCH deposi and loans. GMM regression: Table II m α β + ε fu = + R Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) Fuure Conrac: $*(Unexpeced deposi growh) $*(Unexpeced loan growh) bea sa ) unexpeced growh is he residual from an OLS auoregression wih 39 lags The unpredicable movemens in balance shees are more correlaed wih implied marginal uiliy bu he resul remains insignifican. In he case of loans his is no so surprising. Banks suffer panics because heir loan porfolios canno be quickly convered ino specie. Therefore loan growh and panic saes may be weakly correlaed. Furhermore, conracs based on unexpeced changes require knowledge abou invesor s expecaions. Opions Fuure conracs based on changes in bank balance shees and call raes are poor predicors of changes in implied marginal uiliy. Eiher, banking panics were relaively cosless in erms of uiliy or our hypoheical insurance conracs do a poor job of paying high reurns during panics and low reurns oherwise. Too many false posiives (high reurns in non-panic saes) can weaken he relaionship beween our fuure conracs and

13 marginal uiliy. An examinaion of he balance shee daa suggess a soluion. Panics are characerized by severe declines in deposis and a spike in he call rae. Table III repors GMM regression coefficiens from a regression of invesor marginal uiliy on dummy variables for differen magniudes of deposi declines. If changes in deposis are correlaed wih marginal uiliy we would expec he marginal uiliy o increase wih he severiy of he deposi wihdraw. Table III GMM regression: m = α + β Dum + ε Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) (3) Dum Var = if: Dep growh < 3% Dep growh < 5% Dep growh < 7% bea sa % of obs wih dum = Implied marginal uiliy increases wih he severiy of deposi wihdraw. This is consisen wih cosly banking panics. The dummy variables used o esimae he resuls in Table III have an insurance conrac inerpreaion. Each dummy is equivalen o a binary pu opion ha pays $ if deposi growh falls below a hreshold and $0 oherwise. We could price his opion and draw inference abou he uiliy cos of large declines in deposis. A binary opion does no include all available informaion, however. We know banking panics are characerized by sharp declines in deposis and increases in he marginal value of excess reserves. Table IV repors GMM regression coefficiens from a regression of invesor marginal uiliy on dummy variables for differen magniudes of deposi declines ineraced wih a dummy for increases in he call rae of money.

14 Table IV GMM regression: m = α + β Dum + ε Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) (3) Dum Var = if: change in call rae > 0 and Dep growh < 3% Dep growh < 5% Dep growh < 7% bea sa % of obs wih dum = The difference beween he coefficiens in Table III and IV reflec he change in esimaed marginal uiliy beween panic and non-panic saes when we re-code panic saes o include boh a large decline in deposis and an increase in he call rae. The coefficiens increase dramaically. Large declines in deposis corresponded wih high marginal uiliy if he marginal cos of liquidiy also increased. In he rare cases where deposis declined sharply bu he cos of excess reserves did no increase he marginal uiliy implied by asse prices remained low. The resuls in Table IV sugges an insurance conrac ha saisfies our condiions of correlaed wih banking panics and observable o he economerician. Consider a pu opion on deposi growh wih a knock-ou provision if he call rae does no increase. The knock-ou pu opion would have he following payous. If call rae increases: $*max{(srike - % decline in deposis),0} 2. If call rae does no increase: $0 Table V repors GMM regression coefficiens from a regression of invesor marginal uiliy on he payou from our knock-ou pu opion.

15 Table V GMM regression: m = α + β Pu + ε Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) (3) Srike: Dep Growh< 3% Dep growh < 5% Dep growh < 7% bea sa The pu opions pay high reurns in saes where implied marginal uiliy is high. The coefficiens increase as he srike price becomes furher ou-of-he-money. This is exacly wha we would expec if banking panics were cosly uiliy erms. The opions pay $.0 for each % decline in deposis below he srike price. The larger he decline in deposis he more invesors valued a marginal penny of wealh. Robusness Check: Are Really Measuring Sock Marke Risk? We've esablished ha call raes and deposi growh are correlaed wih banking panics and marginal uiliy. Before we place a price on his risk we need o be cerain ha we aren' simply measuring sock marke risk. The sock marke declines during banking panics and he fac ha he observable sock marke excess reurn E sm [ R ] R f is posiive suggess he sock marke is negaively correlaed wih marginal uiliy. When we exclude he sock marke from our esimaion of (3) we should worry ha our esimaed beas may be biased from an omied variable. To es if banking panics effec marginal uiliy holding he sock marke fixed we require a muliple regression of marginal uiliy on our hypoheical pu opion and he reurn on he sock marke m α β β + ε pu sm = + R + 2R (4) Again we esimae (4) via GMM by choosing he regression coefficiens o bes

16 price 5 NYSE call-rae bea and 5 size-sored sock porfolios. GMM regression: m Table VI α β β R + ε pu = + R + 2 Esimaed wih 5 size and 5 call-rae bea sored NYSE porfolios regression () (2) (3) sm Srike: Dep Growh< 3% Dep growh < 5% Dep growh < 7% Bea Pu sa Bea Sock Marke sa The beas decrease slighly in magniude bu remain significan. Thus we are confiden ha our hypoheical pu opions conain informaion abou marginal uiliy even afer conrolling for he sock marke declines ha so ofen coincided wih banking panics. Pricing he Pu Opions We have consruced opion conracs ha gilded age invesors could use o insure agains he uiliy loss of banking panics. The quesion remains, jus how cosly were hese panics? The regressions in Tables IV-VI provide srong evidence ha marginal uiliy was higher during imes when our opions expired in-he-money. Wih over 700 ime series observaions and porfolios comprised of more han 200,000 individual sock reurns even economically insignifican uiliy differences will be saisically significan. Before we draw conclusions abou he economic cos of banking panics we require a price of panics in erms of forgone consumpion. A naural way o hink abou he cos of bad oucomes is o ask, wha would one

17 pay o avoid hem? The pu opion payous increase during panics. If an invesor expeced his consumpion o fall due o a banking panic, he could insure agains his risk by purchasing opion conracs. This would eliminae he risk bu i would come a a cos if E[ X ] P = E[ mx ] >. Tha is, i would be cosly o insure if he expeced reurn o buying R f he conrac is lower hen he reurn of he risk-free asse. From (2) we know ha his is equivalen o saying i is cosly o insure if cov( m, X ) > 0. Our GMM regressions of m on our hypoheical opions ell us i is cosly o insure by buying pu opions. How cosly amouns o an empirical quesion of wha price would our hypoheical conracs rade for if hey were offered for sale during he gilded age? We can price he pu opions from he ime series of opion payous and he momen condiion price P = E[mX ]. The price will obviously depend upon he marginal uiliy m. Wha m should we use? An obvious choice is he marginal uiliy implied by our regression m α β β + ε pu sm = + R + 2R. Wih he realizaions of opion payous and esimaes of P = E[mX ] we can pu compue he opion risk premium E[ R ] R. Table VII repors he annual cos of insuring $ of bank deposis agains declines below he srike price during monhs ha he call rae increases (he expeced cos of buying a $ knock-ou pu opion). f Table VII Annual cos of insuring $ by purchasing he knock-ou pu opion Srike Price 3% O.T.M. 5% O.T.M. 7% O.T.M. Annual Cos of Insuring $ I would have cos beween % o insure bank deposis agains simulaneous increases in he call rae and deposi declines of 3-7% respecively. For comparison i

18 would cos 2.6%, 5.64% and.6% o insure each monh agains 3%,5% and 7% declines in he sock marke using plain vanilla pus priced via Black-Scholes a 20% implied volailiy. As 20% is approximaely he average implied volailiy on he S&P 500 over he pas 20 years we can conclude ha gilded age invesors feared bank panics a leas as much as modern invesors fear sock marke declines.

19 References Calomiris, Charles W. and Gary Goron. (99) "The Origins of Banking Panics: Models, Facs, and Bank Regulaion." In Financial Markes and Financial Crises, edied by R. Glenn Hubbard, Chicago: Universiy of Chicago Press. Cochrane John H. and Jesus Saa-Requejo. (200) "Beyond Arbirage: Good-Deal Asse Price Bounds in Incomplee Markes," Journal of Poliical Economy, Universiy of Chicago Press, vol. 08(), pages 79-9 Hansen LP, Jagannahan R. 99. Implicaions of securiy marke daa for models of dynamic economies. Journal of Poliical Economy 99: Miron, Jeffrey A. (986) Financial Panics, he Seasonaliy of he Nominal Ineres Rae, and he Founding of he Fed. American Economic Review 76: Friedman, Milon and Anna Schwarz. (963) A Moneary Hisory of he Unied Saes, Princeon: Princeon Universiy Press. Sprague, O. M. W. (90) Hisory of Crises under he Naional Banking Sysem. Naional Moneary Commission, 6s Cong., 2nd sess. Senae Documen 538. Washingon, DC: Governmen Prining Office. Wicker, Elmus (2000) Banking Panics of he Gilded Age. New York: Cambridge Universiy Press.