The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
On th e Way to a New M on etary Union: The E u ro p ean M on etary Union T he P -S tar Model in Five Sm all E co n o m ie s Is th e D iscou n t W indow N ecessary ? A Pen n C en tral P e rs p e c tiv e Can D eposit In s u ra n c e In c re a s e th e Risk of B ank F a ilu re ? Som e H isto rical Evid en ce TIIE FEDERAL A RESFKM J kBANK of Z st . m i is 1 F e d e ra l R e s e rv e B an k of St. L o u is R eview May/June 1994 In This Issue . . . 3 On th e W ay to a New M o n etary U nion: T h e E u ro p e a n M o n etary U nion Helmut Schlesinger In the last two centuries, only a handful of m onetary unions have been created successfully. Now, Europe has em barked on the creation of one of the most ambitious to date: the European M onetary Union, w hich will encom pass nearly 400 million people and have the highest gross domestic product in the world. In the Eighth Annual Homer Jones Memorial Lecture, Helmut Schlesinger, form er president of the Deutsche Bundesbank, briefly traces the econom ic and political history of the European Community, from its beginning in 1957 with six m em bers to its likely expansion to 16 m em bers in 1995. He also reflects on the cu rren t situation in light of the M aastricht Treaty of 1993, which established the fram ew ork for the m onetary union. Although the actual date of the European M onetary Union is in question, Schlesinger concludes that its birth is imm inent and that one of the most critical elem ents in determ ining its success will be the establishm ent of a single curren cy that is as stable as the best-perform ing national currencies within Europe. 11 T h e P -S tar Model in F iv e Sm all E co n o m ie s Clemens J.M. K ool and John A. Tatom A nation’s exchange rate regim e affects the link betw een its m onetary ag gregates and its general level of prices, according to Clemens J.M . Kool and John A. Tatom . They explain an em pirical specification of a quantity theory of money called the P-star model, which indicates that a country's price level depends principally on its own money stock. This theory, however, applies only to a closed econom y or one with a flexible ex change rate. In contrast, they argue, a small open econom y which pegs the value of its curren cy to another country’s cu rren cy also pegs its prices to that cou ntry’s money stock. Kool and Tatom explain how the P-star model can be adapted for use in small open econom ies with fixed exchange rates. The}' test their openeconom y P-star model on five countries bordering Germany: Austria, Bel gium, Denm ark, the Netherlands and Switzerland. These countries have pegged their curren cies to the deutsche m ark to varying degrees since the breakdow n of the Bretton Woods agreem ent. Kool and Tatom ’s evi dence supports the theoretical model, especially the principal hypothesis that, to the extent a country pegs the value of its cu rren cy to the Ger man m ark, its own inflationary developments are determ ined by m one tary conditions in German)'. MAY/JUNE 1994 2 31 Is th e D isco u n t W in d o w N e ce ssa ry ? A P e n n C en tral P e rs p e c tiv e Charles II . Calomiris Policym akers generally regard the discount window as an essential tool for preventing the spread of financial crises, but some critics have argued that it is an unnecessary—and costly—policy instrum ent. The argum ents against the discount window emphasize that it may unwisely postpone bank failures or underm ine the Fed’s control over the supply of reserves. Fu r therm ore, the critics argue, open m arket operations are a sufficient tool for policy objectives. Charles W. Calomiris argues that the discount window, properly ad ministered, can help the governm ent direct tem porary credit subsidies through the banking system to firm s suffering from a “panic” in a non bank financial m arket. He looks at the com m ercial paper crisis of mid-1970, w hich revolved around the failure of Penn Central, a w atershed event in the history of the com m ercial paper m arket. Penn Central, Calomiris con tends, is an example of a beneficial discount window intervention. He concludes that the discount window in the future could have a potential ly stabilizing effect on nonbank financial m arkets, including new, u ntest ed m arkets for credit and derivative instrum ents. 57 Can D eposit In s u ra n c e I n c re a s e th e R isk of B an k F a ilu re ? S om e H isto rica l E v id en ce David C. VVheeloch and Paul W. Wilson Many econom ists have argued that unless prem ium s are risk-based, deposit insurance will encourage banks to take g reater risks than they otherw ise would, thereby increasing the likelihood of failure. Because virtually all banks today are insured, isolating the effects of deposit insurance from other regulatory and econom ic conditions that affect bank perform ance is problem atic. David C. W heelock and Paul W. Wilson study the impact of deposit insur ance by drawing on historical evidence from a voluntary insurance sys tem that operated in Kansas betw een 1909 and 1929. Because insurance was optional in this system, com parison of insured and uninsured banks facing otherw ise similar regulations and econom ic conditions is possible. The authors find that insured banks held less capital and reserves than uninsured banks, and that banks with low capital and reserves, or a heavy reliance on borrow ed funds, w ere m ore likely to fail. In short, risky banks w ere m ore likely to fail, and m em bers of the state deposit insurance system tended to be riskier than nonm em bers. All non-confidential data and programs for the articles published in Review are now available to our readers. This information can be ob tained from three sources: 1. FR ED (F e d e ra l R e s e rv e E c o n o m ic Data), an e le c tr o n ic b u lle tin b o a rd s e rv ic e . You can access FRED by dialing 314-6211824 through your modem-equipped PC. Parameters should be set to: no parity, word length = 8 bits, 1 stop bit and the fastest baud rate your modem supports, up to 14,400 bps. Information will be in directory 11 under file name ST. LOUIS REVIEW DATA. For a free brochure on FRED, please call 314-444-8809. FEDERAL RESERVE BANK OF ST. LOUIS 2 . T h e F e d e ra l R e s e rv e R an k o f St. L o u is You can request data and programs on either disk or hard copy by writing to: Research and Public Information Division, Federal Reserve Bank of St. Louis, Post Office Box 442, St. Louis, MO 63166. Please include the author, title, issue date and page numbers with your request. 3. I n te r -u n iv e rs ity C o n s o rtiu m fo r P o l i t i c a l a n d S o c ia l R e s e a r c h (IC P S R ). M em ber institu tions can r e quest these data through th e CDNet O rder facility. N onm em bers should w rite to: ICPSR, In stitu te fo r Social R esearch, P.O. Box 1248, Ann A rbor, MI 4 8 1 0 6 , or call 3 1 3 -763-5010. 3 Helmut Schlesinger Helmut Schlesinger, former president of the Deutsche Bundes bank, is professor at the Post-Graduate School of Administra tive Sciences, Speyer, a guest professor at the University of Karlsruhe, and will teach in the fall at Princeton University. This article is taken from the Eighth Annual Homer Jones Memorial Lecture, delivered at the Federal Reserve Bank of St. Louis on April 14, 1994. The author’s comments do not necessarily reflect the views of the Federal Reserve Bank of St. Louis or the Federal Reserve System. On the Way to a New Monetary U n io n: The European Monetary U n io n jf jO O K IN G BACK IN HISTORY over the last two centuries, we will find only a few cases of a successful creation of a m onetary union but a far larger num ber of cases w hich failed. As far as the final goal is concerned, it m eans having only one single currency for all the nations in the European Union and replacing the individual national currencies; it is a very big undertaking. It means that one has to create a m onetary un ion ultimately encom passing nearly 400 million people and a region which would have the highest gross domestic product (GDP), the most extensive foreign trade, and so on, in the world. In oth er words: It is an ambitious goal which has not yet been reached, but we are on the way to reaching it. A MONETARY UNION: IN THE BEGINNING The goal of a m onetary union must be seen as one specific objective along the road from the European Community (EC) created by the 'IY'ea ly of Rome in 1957 and the European Union shaped by the Treaty of M aastricht which was ratified in 1993. The long way we have com e from the beginning to the present state of the European Union is impressive; this holds true not only of the progress made, but also of the time span which has been necessary to make it, and it has not been without a num ber of crises, set-backs and periods o f stagnation. From the beginning, the way went along the econom ic integration of the national economiesin direction of a single European economy. But the real target was a political one. This was very clear at the outset. The first step was the crea tion of the European Coal and Steel Community in 1952. At that time, three politicians were of particular im portance: Robert Schum an, Alcide de Gasperi and Konrad Adenauer. It is worth rem em bering that these three statesm en all cam e from bord er regions, betw een France and Germany or betw een Italy and German-speaking Austria. Schum an and de Gasperi were both bilingual, that is, they spoke French and Ger man, and Italian and German, respectively. Adenauer came from the w estern banks of the Rhine River and was, to a certain extent, French-oriented. These three men, who had all MAY/JUNE 1994 4 experienced the two big wars in Europe b e tw een the nations w hich actually belonged so close together, had the courage to create this European Coal and Steel Community as the first Community organization. To a certain extent, coal and steel w ere considered to be the most im portant strategic materials. In Germany at that time, the production of such materials was limited and under the control of the Allied Forces. Later, the production of these materials cam e into surplus and lost its particular strateg ic im portance. But the Coal and Steel Communi ty, which exists even today, fulfilled the role of a nucleus for the development of the European Common Market. The subsequent steps towards the European Community should be well-known. I m ean the creation of the Community of six nations, West Germ any France, Italy, the Netherlands, Belgium and Luxembourg, in 1957. The next important step was the extension of the Community to in clude the United Kingdom, Denm ark, Ireland and later Spain, Portugal and Greece. At the begin ning of 1993 the target to create a so-called sin gle m arket had been reached. Since that time, no governm ent b arriers have been in place b e tw een the m em ber countries as far as the trade in goods and services is concerned, as well as the movement of labor and capital. A fter the ex tension of the Community from six to 12 coun tries, the so-called deepening of the Community had reached an im portant point. Now, Europe is on the brink of doing both: an extension of the Community to include new m em bers com prising the Scandinavian countries and Austria, and the deepening on the way to a m onetary union. Even now, the prim ary target is a fu rth er strengthening of the econom ic in tegration. It means achieving a large, single m ar ket for the w estern and central European countries and, at the end, to use only one cu r rency in that market. But one should not forget that the target even now is political. It should bring the European countries with a rather stable dem ocratic consti tutions closer together and allow them to de velop into a political union. The end of the Cold War has given them m ore freedom to do so. Be fore that, Austria and Finland, in particular, had to take care not to provoke the Soviet Union. The development of free and dem ocratic socie ties in Poland, the Czech Republic and Hungary is another new elem ent which must be handled with care. FEDERAL RESERVE BANK OF ST. LOUIS CONFLICTS IN MONETARY POLICY: “MONETARISTS” VS. “ECONOMISTS” From the beginning of the EC there were differences in the attitudes in the various cou n tries, especially in France and Germany, as far as the role of m onetary policy in the econom ic integration was concerned. The dispute was seen as a conflict betw een "m onetarists” and “economists,” which m eans something different in St. Louis. The “m onetarists” believed that m onetary integration has to start first and that econom ic and political integration would follow. The “econom ists” believed that econom ic conver gence betw een the national econom ies m ust oc cur before any move into very close m onetary integration and a m onetary union. At the end of the 1960s and the beginning of the 1970s, the "m onetarists” gained stronger in fluence. An EC Community study, the so-called W erner Plan, described a step-by-step introduc tion of a m onetary and econom ic union. But the first step under this plan, that is, the obligation to have fixed but changeable exchange rates b e tween the national currencies, was rath er un successful. The tim es were characterized by the end of the Bretton Woods system, and later by the first oil price hike and a world-wide reces sion in 1975. Only the core of hard-currency countries w ere able to stay together without in terruption, specifically, Germany and the Bene lux countries. The real reason for the failure of this first at tempt to have a system of fixed but adjustable exchange rates for all countries was the strong deviation in the rates of inflation. Between 1973 and 1979, for instance, the annual rate of infla tion was 11 percent in France and 16 percent in Italy, but only 4.7 percent in Germany. For the Federal Republic of Germany, however, this was the highest rate of inflation in a medium-term average in four decades. The heavily engaged politicians in this field were following the doc trine of the so-called m onetarists: Giscard d'Estaing in France and Helmut Schmidt in Germany tried to base the integration in the m onetary field on a stronger institutional platform than before. They created the European M onetary System (EMS), which came into effect in March 1979 and is practically existing up to now. I do not think it is unfair to explain the common in terest of the French president and the German chancellor, d’Estaing and Schmidt, respectively, 5 apart from very im portant points, with one com m on motive: The dominance of the deutsche mark was of particular concern for Giscard and the dom inance of the Bundesbank seemed to be a big concern in the eyes of Helmut Schmidt, even though this was not actually true. The EMS was constructed as a system of fixed but adjustable exchange rates betw een the EC countries, w hich allowed fluctuations of these exchange rates only within a relatively narrow band. And which, secondly, established a large fram ew ork of partly unlimited credit facilities for the m em ber central banks. This credit mechanism was to make it possible to keep the exchange rates stable, even under strong pres sures on one particular currency, through the obligation of central banks to intervene in the foreign exchange markets, if necessary, with un limited amounts. THE INHERENT WEAKNESS OF THE EMS Looking back, the creation of the EMS was a very im portant step towards a European m one tary union at a later date. It worked in two directions. First, all countries learned what is possible under given conditions in Europe and— w here necessary—how these conditions had to be changed. Second, all m em ber countries learned that a fixed exchange rate can stimulate the integration of trade and other transactions across the borders of the m em ber countries of the Community. We observed that the m em ber countries felt a systemic pressure to try to orient their own domestic development, especial ly as far as the price developments were con cerned, to that of the best-perform ing countries. Taking the same countries which I have m en tioned before, for the period from 1979 to 1990, we can see that the annual increase of prices in France then was only 6.9 percent, com pared to 10.7 percent in Italy and 2.9 percent in Germany. The inflation rate differentials had diminished, and the average rate of inflation in the EC was lower. In the years after 1990, these differences were sm aller still and partly the oth er way around.1 But we also experienced that even those sm aller d ifferences—whenever they existed over a longer period of tim e—create a need for a change in the exchange rate structure. In other words, they result in the need for a realign ment. If one counts exactly, we have had 16 bigger and sm aller realignm ents since 1979, namely 11 betw een 1979 and 1987, but there w ere no general realignm ents up to Septem ber 1992, the rest later. This seems to be a relative ly good experience of a system of fixed ex change rates aimed at achieving a convergence of the economies. But even in the period from 1987 to 1992, price differentials accumulated, with the conse quence that those countries w hich kept their nominal exchange rates stable experienced an increase in their real exchange rates. Italy, for instance, recorded a real appreciation of its cu r rency vis-a-vis the deutsche m ark and other hard currencies in an amount of nearly 15 p er cent betw een 1987 and 1992, and the same was m ore or less true for the pound sterling. Such real appreciation leads to a loss of the com peti tive position of a country, and it depresses its exports and the overall domestic situation. The latter is aggravated by the fact that it becom es necessary to increase the interest rates in these countries far above the level in the hardcurrency countries. The solution to eliminate these tensions was not ideal: The United King dom and Italy withdrew from the ERM in au tum n 1992. These events show the real w eak flank of a fixed exchange rate system. To avoid these events, a realignm ent would have been n eces sary at an earlier stage, a devaluation of the m ore inflationary currencies vis-a-vis the rather stable currencies, that is, the deutsche m ark, the Dutch guilder, the French franc, and so on. But realignm ents are highly political affairs—at least the politicians have made them into that. Mv long-standing experience with this—reflected by Oscar Wilde, who wrote: “Experience is the sum of the failures”—is that realignments are made only under the strongest pressure in the foreign exchange market, not on the basis of any profound backward and forward-looking analysis. This was also the case in the Bretton Woods system and one of the reasons why it broke down. In oth er words, this is the inherent weakness of the EMS or, to be more precise, of any system with fixed but adjustable exchange rates. 1See Fischer (1994). MAY/JUNE 1994 6 One solution to ease the problem —having a change of the exchange rates but not n ecessari ly a political decision about a realignm ent—was found by widening the band of fluctuation of the exchange rates. This was done in August 1993 by widening this band to ±15 percent; in stead of ± 2 .2 5 percent and ± 6 percent, respec tively. In fact, since we have had this wider band, the exchange rates of the EU countries have been behaving rath er calmly. No country was forced, or prepared, to use the wider band for a stronger devaluation or revaluation. Each country now has a clear responsibility for its own currency and for the exchange rate of its currency. In my opinion, this is a rather good basis for fu rth er developments on the way to a European m onetary union. A fter a longer peri od, in which the exchange rate stru ctu re of the EMS currencies would have com e into a longerlasting equilibrium, the fluctuation band could be diminished somewhat. TH E CONCEPT OF THE ECONOM IC AND MONETARY UNION (EMU) The 'Iteaty of M aastricht, called the Treaty on European Union, the most im portant extension of the Treaty of Rome, is rather clear about the different steps needed for a m onetary and eco nomic union, but rath er vague concerning the elem ents of a political union, that is, a common foreign policy, defense policy, harm onization of laws or social policy. By the way, a new in ter governmental conference is to take place in 1996 to implement the political part of the Union. Coming back to my earlier rem arks, however, the Treaty of M aastricht contains more of the ideas of the “m onetarists” than of those of the "economists.” But this type of victory was easier to reach than anything in the purely po litical field. A m onetary union means that all m em ber countries have to give up their national sovereignty in this area. It m eans m onetary un ion first and political union later. A monetary union is a sacrifice of sovereignty for each country. And, as French President Mitterand said before the referendum in France in 1992, the biggest sacrifice is in Germany, the country with the anchor currency in the EMS and, therefore, the country with the m onetary policy that is most independent from the policies of the other m em ber countries. Having said this, however, I should add that Germany has never been completely independent from the others in the past and that it has usually had to take into FEDERAL RESERVE BANK OF ST. LOUIS consideration the consequences of its own poli cy decisions for the p artn er countries. But, nevertheless, this has been done on its own judgment. In the oth er political areas, the national sovereignty is not given up. T h ere are some limi tations, some common rules, some Community directives, but there is no direct interference, for instance, in foreign policy decisions. Even in the case of fiscal policy, the national govern ments are sovereign. They agreed to a process of m ultilateral surveillance and they agreed in the M aastricht H'eaty on some soft sanctions for misbehaving countries, but each state decides in full sovereignty on its own taxation system and on tax rates and public expenditures. Of course, some rules have been formulated, but only for the public sector deficit and for the level of governm ent debt. Agreem ent on comm on Euro pean foreign and defense policies is very difficult and can be seen in the role which the European Union played—or did not play—in the b reak down of the Yugoslav Republic and the conse quences that followed. Having said this, the m onetary union described by the M aastricht Treaty is not an illusion. It takes into account many of the experiences gained in the EMS period, and it proclaim s a path toward the EMU in three stages. Stage I means freedom for international capital move ments; this is given now. It also calls for m em bership in the EMS, preferably in the ERM, but—as I have said b efore—the latter is not yet given for all countries. For example, the United Kingdom, Italy and Greece are not in the ERM. Stage II started at the beginning of this year. It includes the establishm ent of a European M one tary Institute (EMI) as the foreru n n er of EMU and the European Central Bank. The Institute has started with its work and now definitely has its seat in Frankfurt, the financial cen ter o f Ger many and the place w here the Bundesbank has its headquarters. The EMI has to clarify the technical and operational questions for the com ing European Central Bank; it has to continue and intensify the coordination of the m onetary policies of the m em ber countries and it has to control the EMS and the development of the ECU. In the final stage, the heads of state of the Eu ropean Union will m eet and decide if EMU can start, in 1997 or 1999. This decision is to be made on the basis of which countries fulfill the 7 so-called entry criteria. This will be a very im portant exercise. The Treaty gives indications of how to proceed in this process of examination. They have to clarify w hich countries have enough and sustainable price stability have sound public finances, and have no strong ten sions as far as the exchange rates of their par ticular currencies are concerned .2 The purpose of this examination is to ensure that only those countries w hich fulfill these criteria can join the EMU at the beginning, thereby guaranteeing that only countries with a sound econom ic basis can be founding m em bers. This means also that, at the beginning, presum ably not every m em ber country would qualify for mem bership in EMU. In oth er words, the Tl'eaty opens the way for a two- or threespeed solution. Those countries w hich are not in EMU at the beginning can com e in when the)' fulfill the entry criteria. The United King dom and Denm ark, however, have the right to stay out under anv circum stances, even if they fulfill the criteria. I think the idea of entry criteria can be consi dered a “victory" on the side of the "economists,” on paper at any rate. One cannot forget that the politicians will have a certain degree of discre tion in their final judgem ent about w hich coun try should join the EMU and which should not. By the wav, the room for discretionary assess ment is strongly limited for the German govern ment, because the German Parliament as well as the Federal Constitutional Court are forcing the governm ent to be very strict in its judgements in this. Nevertheless, it is a highly political issue. Everyone knows it cannot be solved solely on the basis of statistics, and even the interpreta tion of the statistical criteria is now under dispute.3 In my opinion, the Tl’eaty o f M aastricht is a rath er well-balanced document. The question is w h eth er this concept can be filled with real life and w hether this life could satisfy the wishes and hopes which many are connecting with it. If I w ere to try to give an answ er as far as the whole content of this new European Union, the political Union, is concerned, I would have to be very vague. I would have to start with the ques tion of w hether the nations are prepared to give up their identities. Are they prepared to forget the experience which they have had with each other during for the past hundreds of years? Can they give up w hatever judgm ents and prejudgments they may have made and— especially—can they forget all the injustices they have inflicted on each other? O ther open points are the consequences of widening the European Union from 12 to 16 states, and what consequences would be for opening for m em bership for the central Europe an states, that is, Hungary (which has already applied for m embership), Poland, the Czech Bepublic, and so on. All these are im portant questions and not really new. More than 35 years ago in Germ any Ludwig Erhard, the Minister of Economics who contributed so much to the "German revival” (Henry Wallich, 1955), was strongly engaged in the European debate. Even then, he was against a Europe of only six countries. He published an advertisement in newspapers, writing only: 6 + 7 + 5 = 1. He m eant that six alone would not comprise one Europe; the seven EFTA countries and the rest of the non-comm unist countries in Europe should be included. In 1995, presum ably 16 countries will form the European Union—only Switzerland and Iceland will be missing to com plete the Erhard formula. I do not want to go into this grand design of European questions, even if I have indicated that I have always been inclined to be on the side of Ludwig Erhard. Let me com e back to the ques tion as to how a European M onetary Union can function. The cen ter piece of the EMU is the es tablishm ent of the European Central Bank, the only central bank w hich could issue the single *Sea: “ Protocol on the Convergence Criteria Referred to in Article 109j of the Treaty Establishing the European Com munity,” Office for Official Publication of the European Communities (1992). 3For instance, “ price stability” is given if a member state has an inflation rate not more than 1.5 percentage points higher than “ the three best-performing member states.” The question is, does it mean the average of these three states or that of the worst (or the best) of these three states? MAY/JUNE 1994 8 European currency. In many respects, this Bank will be modeled on the German Bundesbank, w hich m eans that it will have the priority target of keeping the value of the currency stable and, for that purpose, it will: • be independent from the national and supra national governments (not accepting orders nor being allowed to ask for any); • have a Central Bank Council consisting of the governors of the national central banks and six m em bers of an executive board who can not be dismissed during their term in office for decision-making; • be granted the necessary instrum ents for m onetary policy, that is, open m arket policy, interest rate policy, minimum reserve require ments, if necessary, and so on; and • the European Central Bank will not prim arily be responsible for banking supervision, but it is possible to tran sfer corresponding tasks to it. CENTRAL BANKING GOALS AND MONETARY AGGREGATES As I have already m entioned, the m onetary union is well constructed on paper and a Euro pean Central Bank could work. The European Central Bank is to be established as a federal in stitution, much like the Bundesbank in Germany, which has the federal elem ents of the United States as its “grandfather,” so to speak. Why should it not be possible for such an institution to fulfill its tasks for Europe? Many people are asking w hether such a new central bank com prising people from different countries with a somewhat different “culture" of m onetary policy could really work. I have to answ er that m one tary policy decisions are never made on the ba sis of a full conform ity of opinions among the decision makers. This is not the case in the Cen tral Bank Council of the Bundesbank with its 16 mem bers, nor is this the case in the Federal Open Market Committee (FOMC) of the Federal Reserve System, as we can all see from the record of policy actions. Both Councils are dem ocratic institutions in which decisions are made through majority, but certainly these insti tutions have a basis of a com m on attitude, and have a certain consensus about what to decide and how to work. Take the U.S. case. David Mullins, the form er vice chairm an, says that 4 percent inflation is FEDERAL RESERVE BANK OF ST. LOUIS unacceptably high. And Wayne Angell, a long standing m em ber of the Board of Governors, w rote in his letter of retirem ent to the President of the United States that “I am pleased... [that] the Board of Governors was enunciating a goal of zero inflation. But if the quest for price-level stability is replaced by an acceptance of an infla tion rate stabilized at say 2-1/2 percent per year, then we are accepting the fact that the domestic value of the dollar will be cut in half every generation, 28 years.” Thus, there is a consen sus, with some room for interpretation, betw een both (ex-)members. In the German case, the Bundesbank is som e times attacked because part of its interm ediate m onetary target—the growth of M3—is derived on the basis of a so-called unavoidable increase of about 2 percent in the price level. Presum a bly, Wayne Angell would also be a little bit dis appointed about this. W hich price level target will the European Central Bank form ulate, if it w ere to form ulate one at all? The 1.5 percentage points over the average of the best-perform ing countries cannot be a target, it is the entry criterion. Up to now, there seems to be no discussion about a price stability target for this future Central Bank. The core countries would presum ably agree on a figure like that of the Bundesbank, and—if that were done—one would be also in agreem ent with Professor Stanley Fischer, who recently wrote “that anyone should be com fortable with a 1-3 percent target” . I do not want to go into the details of the use of a declared price stability target. Canada and New Zealand are using one, and have indeed reached and kept to it up to now. It is worth noting that the last Governor of the Bank of Canada, whose bank was successful, had to re sign w hen a new governm ent came into power. Price stability targets in the sense of a stable consum er price level are a very ambitious exer cise. So far, New Zealand’s central bank keeps it at 0-2 percent inflation, even against opposition. Nobody would be surprised—especially not at the Federal Reserve Bank of St. Louis—if I were to suggest that the European Central Bank should start immediately with targeting on the basis of a money supply target. I m ust admit that this would be rath er complicated at the beginning, because nobody could know exactly how the Eu ropean money supply, the money demand fun c tion, and so on, will behave. But, in my opinion, 9 it is indispensable that this new central bank has an idea o f how strongly the money stock should grow and, especially, of how far the m onetary basis should be extended, the money w hich this new European Central Bank System will create itself. This question reminds me of the fact that after the deutsche m ark had been introduced in 1948, the military government prescribed that any additional increase of cu r rency in circulation, exceeding a total circulation of $10 billion, needed a very restricted proce dure of agreem ents by a three-fourths majority of the Central Bank council and had to be ap proved by the Lander Governments. This was a primitive way of control, but it made clear that any extension of money must be limit ed, especially if it is a new money, which does not inspire the same confidence at the begin ning than a well-proven old currency. The discussion of w hether or not to have money supply targets will be a point of con troversy in the EMI. The President of the EMI, Baron Lamfalussy (1994), has form ulated the different positions that will presum ably be held, for example, using the money supply as an in term ediate target, as is practiced in Germany and some other countries. Or, having no in ter mediate target but a concrete price stability tar get, such as New Zealand. Or neither of them, which he seems to p refer—reading betw een the lines—by saying “can one not use m onetary ag gregates at least as an inform ation variable?” . This seems to be a new form ula which was described, for instance, by Benjamin Friedman (1994). I find this new label sounds good but I have to add that such eclecticism , as a m atter of fact, is not new. I think that advanced central banks follow the development of the money sup ply aggregates everyw here and at all times, b e cause they value the inform ation that m onetary aggregates provide and their im portance to the econom ic developments, but they also take other factors into account. CONCLUSIONS W hen the EMI has completed its w ork of preparing the future European Central Bank, when the examination of the entry criteria has been made and the heads of state of the Euro pean Union have decided that the m onetary un ion can start, then the adventure will begin. At the beginning, only the exchange rates betw een the different national currencies of the m em ber countries have to be fixed without any margin and without any possibility of changing them in future. In a later stage, the single curren cy can replace these national currencies: This would be the fulfillment of the European M onetary Union. It may seem to you that the way to EMU is a relatively long one, that many preconditions which have been form ulated give the impression that the fathers of this program are themselves not sure about the success o f the whole plan. And you could add that Europe needs too much time to find solutions com parable with the dy namics in the United States. But even here in the United States, it took quite a long time b e fore you got a Federal Reserve System, even though one currency existed. Now, in Europe, we have central banks. In quite a num ber of our countries we have currencies which are relatively stable and we have a situation in w hich the exchange rates are no longer fluctuat ing very strongly, although they could. In other words: The integration of the econom ies in Eu rope and, especially, in the European Union can and will continue to progress toward m ore com plete econom ic integration. For this purpose, it is not so im portant—either for our countries or the rest of the world—w hether the European M onetary Union starts in 1997, 1999, or even a little later. W hat is decisive is that we replace the different European currencies with a single currency that is as stable as the best-perform ing national currencies. REFEREN CES Board of Governors of the Federal Reserve System. Federal Reserve Bulletin. Board of Governors of the Federal Reserve System, 1994, p. 235. Brash, Donald. Speech delivered to the Wellington Chamber of Commerce, in Bank for International Settlements BIS Review, No. 45 (1994). Erhard, Ludwig. “ Was Wird aus Europa?” (What Will Be come of Europe?), Handelsblatt (December 23-24, 1960), reprinted in Ludwig Erhard, Deutsche Wirtschaftspolitik. Diisseldorf, Wien, Frankfurt, 1962, p. 530. Fischer, Stanley. “ Costs and Benefits of Disinflation,” in A Framework for Monetary Stability: a paper and proceedings of an international conference organized by De Nederlandsche Bank and the Center for Economic Research at Amsterdam, the Netherlands, October 1993. J. Onno de Beaufort Wijnholds, Sylvester C.W. Eijffinger, and Lex H. Hoogduin, eds. Kluwer Academic Publishers, 1994, pp. 31-36. Friedman, Benjamin M. “ Intermediate Targets Versus Informa tion Variables as Operating Guides for Monetary Policy,” in A Framework for Monetary Stability, J. Onno de Beaufort Wijnholds, Sylvester C.W. Eijffinger, and Lex. H. Hoogduin, eds. Kluwer Academic Publishers, 1994, p. 109. MAY/JUNE 1994 10 Lamfalussy, M. Alexandre. Address at the Hessian Bankers’ Association, in the Bank for International Settlements BIS Review, No. 41 (1994). Mullins, David W., Jr. “ A Policy Maker’s Perspective,” in A Framework for Monetary Stability, J. Onno de Beaufort Wijn holds, Sylvester C.W. Eijffinger, and Lex H. Hoogduin, eds. Kluwer Academic Publishers, 1994, p. 6. Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER Office for Official Publication of the European Communities. Protocol on the Convergence Criteria Referred to in Article 109j of the Treaty Establishing the European Community. Luxembourg, 1992. Wallich, Henry C. Triebkrafte des deutschen Wiederaufstiegs. Frankfurt: F. Knapp, 1955. 11 Clemens J.M. Kool and John A. Tatom Clemens J.M. Kool is associate professor of international and monetary economics at the University of Limburg, the Nether lands. John A. Tatom is an assistant vice president at the Federal Resen/e Bank of St. Louis. Jonathan Ahlbrecht provided research assistance. The authors are grateful to Martin M.G. Fase, Eduard Hoechreiter and Dieter Proske for helpful comments. The P-Star M odel in Five Small Economies / h e QUANTITY THEORY a n d its equation of exchange provide a proven and useful fram e w ork to empirically analyze the relevance of money in the econom y. During the past decade, however, doubts about this approach arose b e cause of the perception that the links betw een money and prices and money and output had loosened or vanished.1 Recently, Hallman, Porter and Small (1991)—h en ceforth HPS—have drawn new attention to the quantity theory by explicitly linking the determ inants of long-run equilibrium prices to the short-run dynamics of actual infla tion in the so-called P-star approach. In this fram ew ork, deviations of the actual price level from equilibrium push cu rren t prices and infla tion in the direction of equilibrium. The em pirical results obtained so far for a wide set of countries, are supportive of the Pstar approach, although it seems to w ork b etter for larger than sm aller countries.2 One neglected aspect w hich may explain this apparent dichoto my is the im portance of the prevailing exchange rate regime for the determ inants of prices and inflation. T he original P-star approach assumes that the equilibrium price level is a function of the domestic money supply. Under a system of fixed exchange rates, however, the domestic price level in a small country is determ ined abroad and the domestic money stock becom es endogenous and demand-determined. This article develops a generalized P-star model that accounts fo r this international effect by including cross-country price gaps. It is test ed using annual data from 1960 to 1992 for five small European cou ntries—Austria, Belgium, Denmark, the Netherlands and Switzerland. During the Bretton W oods period, these coun tries pegged their exchange rates to the United States dollar. Since then, fou r of the five pegged their curren cies to the German m ark, with varying degrees of success. Only Switzerland has had a floating exchange rate regim e con tinuously since the breakdow n of Bretton Woods. W e investigate the extent to which prices in these countries have been affected by developments in Germany, as well as domesti- 1For an overview of past discussions on this issue, see Bat ten and Stone (1983), Dwyer and Hater (1988) and Dewald (1988). 2See, for example, Hoeller and Poret (1991). They also indi cate that there generally are superior models for forecast ing inflation movements, even for countries where the P-star model is not rejected. MAY/JUNE 1994 12 cally. To assess the tests used here, we also in vestigate the effect of U.S. price developments on the price level in these five countries and in Germany. The results below indicate that the five small countries’ equilibrium price levels are d eter mined in Germ any under the fixed exchange rate regim e and that this effect has been pro portional to the tightness of the exchange rate peg. In contrast, foreign-based equilibrium prices are found to be insignificant for the United States and Germany, the tw o countries that generally floated over the period exam ined here. In the next section, the theoretical fram ew ork is developed, focusing first on the P-star model for a closed econom y. Then, a generalized vari ant of the m onetary approach to the balance of payments is used to show that under fixed ex change rates, domestic price developments in a small country are determ ined abroad, and that domestic money becom es endogenous. Combin ing these ingredients, it is shown that both their own price gap and a price gap based on equi librium prices determ ined abroad can affect a cou ntry’s inflation, depending on the exchange rate regim e. The subsequent sections discuss the data, present the em pirical results, and summ arize the paper. TH E P-STA R MODEL IN CLOSED AND OPEN ECONOMIES The Closed E conom y M odel The simple Quantity Th eory’s equation of exchange is w here P* denotes the equilibrium price level to w hich actual prices converge in the long run, Y* is potential real output, and V* is the equi librium velocity of money. Following the Quanti ty Theory, they assume that V* and Y* are determ ined independently, and, m ore im portant ly, that both are independent of the m oney stock. Thus, the equilibrium price level moves proportionally with the stock of money. HPS fu rth er hypothesize that the equilibrium price gap, (In P -ln P *), has a theoretical value of zero so that P adjusts to equal P*. The com bination of equations 1 and 2 implies that the change in the actual price level should be negatively relat ed to the existing gap betw een P and P*. This relationship is form ally indicated by the hypothesis that a, is negative in the second term of the inflation equation: (3) AlnPt = a {) + + I! (ln P -ln P *)t x Pj MnP,_j + e,. The inflation lags AlnPtJ are added to the equa tion to account for short-run dynamics and e, is the random erro r term . If actual inflation, AInP, is nonstationary, then it does not have a fixed theoretical mean, possi bly leading to problem s in the estimation of equation 3. To accom m odate this possibility, equation 3 can be rew ritten w ithout loss of generality as (4) Ax, = a 0 + a, U n P -ln P *),_, N - 1 (1) P = M (V/Y), w here -ir now denotes inflation. w here P denotes the price level, M is the do m estic stock of money, Y is real output and V is the velocity of money. For convenience, time subscripts are omitted. Equation 1 simply pins down (nonobservable) actual velocity for given observations on P, M and Y. HPS (1991), how ever, hypothesize the follow ing long-run equilibrium relationship based on the identity in equation l : 3 (2) P* = M(V*/Y*), 3Humphrey (1989) gives a review of the precursors of this approach and shows that a variant can be traced back to the work of David Hume. FEDERAL RESERVE BANK OF ST. LOUIS If inflation is not stationary (and its first difference is stationary), then 6n in equation 4 has a theoretical value of zero, and lagged infla tion can be omitted. Thus, equation 4 would contain only stationary variables (since tti i can be omitted). Since this is not the case in equa tion 3 unless inflation is stationary, equation 4 is generally a m ore useful equation to estimate. Although it would be possible to include other transitory influences on prices, such as pricecontrol proxies or energy price shocks, we 13 abstract from these factors and focus on the interaction betw een changes in inflation and deviations from long-run equilibrium. HPS (1991) originally applied a version of equation 4 to quarterly U.S. data. They use M2 as the money stock and assume that the c o r responding equilibrium velocity is a constant.4 HPS conclude that the model is supported by the data.3 Hoeller and Poret (1991) extend the P-star approach to 20 m em ber countries of the Organization for Econom ic Co-operation and Development (OECD). They use the HodrickPrescott filter to extract equilibrium time series for output and velocity from the data.6 The evi dence provided by Hoeller and Poret is mixed. The P-star approach leads to satisfactory esti mated equations for most, but not all, countries. In particular, the evidence fo r small countries tends to reject the P-star model, while the evi dence for larger countries tends to conform to the P-star model. So far, research on the link betw een exchange rate regim es and m acroeconom ic adjustm ent of prices and output is quite limited. Recently, Bayoumi and Eichengreen (1992) use impulseresponse functions to analyze the differences betw een the Bretton W oods and post-Bretton Woods period in this respect fo r the G7 coun tries (United States, Canada, United Kingdom, Germany, France, Italy and Japan). They find evidence that, under the floating rate regime, countries' aggregate demand curves becom e steeper so that various shocks give rise to less output variability and greater price variability than under a fixed rate regim e. Tatom (1992) analyzes Austrian price behavior; the P-star model for Austria is rejected but, due to the fixed exchange rate regim e, a long-run relation 4Tatom (1990) points out that M2 velocity has exhibited semipermanent trends over time, so that the assumption of a constant equilibrium velocity may be flawed. 5HPS also split up the total price gap in separate output and velocity gaps, but find no additional explanatory power from this less-restricted variant. 6Hoeller and Poret also conduct tests using simple linear trends for the equilibrium levels of output and velocity. The Hodrick-Prescott filter allows time series with stochastic trends to be detrended. See Hoeller and Poret (1991) for a discussion. King and Rebelo (1989) contains a more tech nical analysis. An application and an appendix with the ap propriate formulas can be found in Mills and Wood (1993). ship betw een German and Austrian inflation is not rejected .7 Here, we intend to investigate in more detail how foreign price developments affect domestic prices under a system of fixed exchange rates and the implications of this linkage for the P-star model. The Generalized M onetary Approach to the Balance o f Payments The starting point of our analysis is a fixed exchange rate regime in which one large country (such as Germany) is the anchor of the system and sets its m onetary policy to achieve its own domestic objectives, independent of the ob jec tives of the smaller countries within the system. The large country is assumed to be sufficiently large so that it is unaffected by policy actions and outcom es in the small countries.8 Each small country, in contrast, takes the anchor cou ntry’s m onetary policy as given and is committed to a fixed exchange rate objective. Equations 5 and 6 represent money demand and money supply, respectively, in the large foreign cou ntry:9 (5) M[j = PfYf (6) M{ = Mf, w here the inverse of velocity is assumed to be a function of real output (KO and a vector of nomi nal interest rates (fV).10 W hen both output and the real interest rate are at their long-run equilibrium values determ ined elsew here in the economy, money m arket equilibrium determ ines 9A superscript f will be used to denote the large foreign country (Germany or the United States), while a super script d will be used for the small domestic country (Aus tria, Belgium, Denmark, the Netherlands and Switzerland). 10The relationship of equilibrium velocity and equilibrium nominal interest rates is generally ignored in formulations of the P-star model. This practice is consistent with the as sumptions that movements in the equilibrium real rate are not empirically significant and that movements in the ex pected rate of inflation have little effect on velocity; more over, even if this latter effect is not small, it is captured in the growth of the money stock or, given the dynamics in cluded in the P-star model, in the lagged inflation terms. 7The Bundesbank (1992) develops and tests a P-star model for Germany based on its M3 measure. 8ln the P-star model, this means that the large country’s potential output, equilibrium velocity and long-term inflation objectives are independent of foreign developments. MAY/JUNE 1994 14 the equilibrium price level in the large foreign econom y (Germany). The exchange rate constraint in a fixed ex change rate system then determ ines the equilibrium domestic price level for a small country as (7) = E P r*/ER *, w here E is the fixed nominal exchange rate, equal to the num ber of equilibrium domestic curren cy units per unit of foreign currency, and ER * is the corresponding equilibrium real exchange ra te .11 W ith the domestic price level conditioned by equation 7, the domestic money stock must adjust to bring about equilibrium in the domestic money market. P-Star in Open Econom ies Under Fixed Exchange Rates The above analysis has two m ajor implications for the short-run price dynamics in small coun tries under fixed exchange rates. First, a price gap determ ined abroad through the exchange rate constraint should be expected to influence domestic inflation. This gap can be defined as Second, the domestic price gap should lose its influence if the exchange rate is pegged; domes tic money becom es endogenous. Suppose, for example, cu rren t domestic prices are consistent with the foreign-determ ined P-star m easure, that is, the foreign-determ ined gap is zero, while simultaneously the domestic gap is posi tive, because actual prices exceed the equilibri um m easure of prices indicated by the domestic money stock. In this case, the domestic gap is expected to close by adjustm ent of the money stock, not by an adjustm ent of domestic prices and inflation. The extent to w hich this holds will be a function of capital mobility. The litera ture on sterilization and capital offset suggests that small countries m ay have some freedom to manipulate the domestic money supply in the interm ediate run to determ ine m onetary condi tions at home to the extent that capital mobility is limited.14 Both of these hypotheses are tested below. In particular, the model in equation 4 is sup plem ented with the foreign price gap so that the appropriate equilibrium gap m easure is a weighted average of the domestic gap in equa tion 4 and the foreign-determ ined gap in equa tion 8 or (9) (1 —w) (p d- p *) + w {pd- p d*)l (8) GAPf = (p d- p d*) = [pd - ( pf * + e -e r * )], w here low er-case symbols denote logarithm ic levels and a d superscript has been added to the logarithm of the domestic price level to distinguish it from a foreign m easure. W hen domestic prices exceed the foreign-determ ined equilibrium price level, downward pressure on cu rren t domestic inflation and prices results.12 The am ount of pressure this gap exerts on cu r ren t domestic inflation and the speed of adjust m ent tow ard equilibrium depend on the extent of arbitrage in goods and capital m arkets, and the degree to w hich the econom ies are integrated.13 11ln the traditional pure monetary approach to the balance of payments, the real exchange rate is assumed to be a constant and may be deleted from the analysis. ^Alternatively, the gap can be closed by a discrete decision to correspondingly devalue the currency. Afterwards, the peg could be resumed. 13Although the degree of integration may have increased over time and may also be a function of the exchange rate regime, rising with a credible fixed rate regime, the effects of these changes are ignored below. FEDERAL RESERVE BANK OF ST. LOUIS w here w is the weight attached to a fixed ex change rate regime. For a closed econom y or a floating exchange rate regime, w equals zero and the appropriate equilibrium price level and gap m easures are the conventional, domestically determ ined ones used in equation 4. If th ere is a credible fixed exchange rate regime w ith the domestic cu rren cy pegged to the foreign coun try, /, then w equals one and the equilibrium price level is that determ ined abroad and indi cated as p d* in equation 8. In this case, the ap propriate P-star and its related gap m easure are determ ined abroad. Since w may change over the sample period, but is unobservable, the ,4See Roubini (1988) for an overview of the literature, and Kool (1994) for a recent empirical analysis. Also, see Stockman and Ohanian (1993). 15 coefficients on the gaps are theoretically the m ean levels reflecting the sample experience and they sum to the mean level of W hile it would be desirable to characterize the prevailing exchange rate regim e over time for each country and to incorporate this in for mation in the analysis, this is not feasible. It may seem straightforw ard to distinguish b e tw een fixed and floating exchange rate regimes, but departures from these idealized extrem es are econom ically and qualitatively im portant in practice and are hard to quantify. Moreover, changes in the degree of international capital mobility and econom ic integration over time may change the speed of response to existing gaps. Finally, the limited num ber of observa tions available below severely constrains the use of extensive sets of dummy variables. For these reasons, we include both the domestic price gap (defined in equation 4) and the foreigndeterm ined price gap (defined in equation 8) in the final specification. TH E DATA AND TH EIR TIME SERIES P R O P ET IE S Annual data for seven countries—Austria, Belgium, Denm ark, Germany, the Netherlands, Switzerland and the United States—for the period 1960-92 are used to test the model. Con sistent nominal and real GDP data have been obtained from the Organization fo r Economic Co-operation and Development (OECD, 1993) for the six foreign cou ntries.15 U.S. Departm ent of Com m erce data are used for the U.S. nominal and real GDP m easures. These series have been used to compute the implicit GDP deflator. Average U.S. dollar exchange rates have been taken from the International M onetary Fund's Internation al Financial Statistics (IFS) database (line rf). Similarly, two money stock definitions from the IFS have been used: narrow money (line 34), w hich is called M l here, and the sum of narrow money and quasi-monev (line 35), w hich is com parable to M2 and will be denoted here as broad money, MB. The main advantage of using these series is their harm onization across countries. For Belgium, the m onetary aggregates series stop in 1990. Exchange Rate M ovem ents Figure 1 presents the nominal exchange rate (defined as the domestic cu rren cy price of Ger man marks) time paths for all countries relative to Germany, with the 1960 exchange rate in dexed to 100. The Bretton Woods fixed ex change rate regim e is clearly visible until the late '60s. Note, that although the form al end of Bretton Woods is often set in 1973 and some times in 1971, exchange rates start moving already in the period 1967-70. A fter the b reak down of Bretton Woods, exchange rates move least for Austria and the Netherlands. These countries have most persistently sought fixed exchange rates with Germany in the '70s and '80s. Much m ore exchange rate variability has been present, on the oth er hand, for the United States, w here the exchange rate has floated and, to a lesser extent, fo r D enm ark and Belgium. Despite a floating rate, the Swiss exchange rate has exhibited less variability than that in Den m ark and Belgium. The latter tw o countries have had mixed exchange rate regimes. W hile they have been on fixed exchange rates, at least nominally, they periodically devalued to escape the exchange rate constraint on domestic m one tary policy. Some degree of exchange rate stabilization appears to have set in the middleand late-’80s for Belgium and Denm ark, however, due to the effective functioning of the European M onetary System during that period.16 15Series in this publication are for 1960 and from 1963 to 1991. We are grateful to Amber DeBayser at the OECD for providing the consistent 1961 and 1962 data, which are not listed in the publication cited. Data for 1992 were comput ed from comparable OECD data. In Belgium, Denmark and Germany, 1992 data were not included due to lack of com parability. 16ln August of 1993, both the Belgian franc and Danish krone were forced to accept wider fluctuation margins in the European exchange rate mechanism and experienced a considerable depreciation; subsequently, their exchange rates moved back into the narrow bands that existed earli er, although the wider margins officially still are in place. This experience is outside the sample used here. MAY/JUNE 1994 16 Figure 1 Exchange Rates Relative to Germany Index (1960 = 100) The decision by the four countries, Austria, Belgium, Denm ark and the Netherlands, to peg their curren cies to the m ark is motivated by a desire to "im port” German inflation, one of the lowest rates in the world from I9 6 0 until re cently. W hile the Swiss chose to float, their m onetary policy has also achieved a similarly low inflation rate. The decision of the four small, open econom ies to peg to the m ark is also presum ably influenced by the fact that they are closely tied to Germany through trade. For example, in 1985-89, trade with Germany (both exports and imports) was 20 percent of Den m ark’s total trade, 21.4 percent of Belgium’s, 26.5 percent of the N etherlands, 27.4 percent of Sw itzerland’s and 39.2 percent of Austria’s. For trade within the six-country block, the shares w ere 28.8 p ercen t for Germany, 30.6 percent for Denm ark, 38.7 percent for Sw itzer land, 41.3 percent fo r Belgium, 44.8 percent for the Netherlands and 51 percent for Austria. Real exchange rates, defined as the nominal exchange rate multiplied by the ratio of German FEDERAL RESERVE BANK OF ST. LOUIS prices relative to each cou ntry’s price level, are displayed in Figure 2. For the United States, Switzerland and Belgium, sizeable perm anent real exchange rate changes relative to Germany appear to have taken place. Nominal and real exchange rate patterns are quite similar for these th ree countries. Real exchange rate move ments have been smaller in magnitude for Aus tria, Denm ark and the Netherlands. W hile the Danish krona has continuously depreciated in nominal term s over time, the real exchange rate has fluctuated around the same level fo r the en tire sample. Unit R o o t Tests One important issue for the co rrect specifica tion of the price equation to be estimated, is the (non)stationarity of the variables involved. Tables 1 and 2 report the results of standard augmented Dickey-Fuller (ADF) tests for both log levels and grow th rates of prices, output, narrow and broad money, the corresponding velocities of narrow and broad money, and the 17 Figure 2 Real Exchange Rates Relative to Germany Index (1960 = 100) nominal and real exchange ra tes.17 In the tables, we report the t-statistic on the one-period lagged level for the p referred specification; this specification is given below the f-statistic. Sig nificance at the 5 percent level is indicated (*) and implies rejection of nonstationaritv. W ith few exceptions, Table 1 indicates that the nonstationarity of the logarithm of the levels of the variables cannot be rejected. Consequently, com putation of the equilibrium values of V* and Y* by m eans of a regression with a determ inistic trend generally is incorrect. The most im portant implication of these results is that a procedure capable of handling stochastic trends is required to model the equilibrium levels of V* and Y*. The Hodrick-Prescott filter is used to find the equilibrium output and velocity paths. 17The specification used in each case is a regression of the first difference of the series on a constant, a trend, the one-period lagged level, and up to three lags of the firstdifferenced variable. Insignificant lags of variables are re moved step by step starting at the longest lag, for specifi cations that include or do not include a time trend. If the The grow th rates of output, narrow money, and narrow and broad money velocity, and the nominal and real exchange rates, all appear to be stationary according to the unit root tests reported in Table 2. For inflation and broad money grow th, a unit root generally cannot be rejected .18 Note that th ere are a considerable num ber of borderline cases. The (marginal) non stationarity of inflation suggests that equation 4 is appropriate for the ensuing gap analysis. TESTS O F TH E P-STA R MODEL In this section, we analyze the impact of different price gaps on short-run inflation dy namics. First, we focus on country-by-country estim ation using each country’s domestically de term ined price gap only. Then we proceed to trend is statistically significant, this version is reported in the table; otherwise, the estimate without the trend is reported. 18Unreported results show that a unit root for the change in inflation and broad money growth can be rejected. MAY/JUNE 1994 18 Table 1 ADF Unit Root Results (log levels) P Y M1 V1 MB VB E ER Austria Country -2 .1 9 (T,1) -2 .6 7 (C,0) -1 .8 2 (C,0) - 2 .5 4 (T,0) -2 .0 8 (C,1) -2 .0 6 (C,0) -1 .7 8 (C,0) -1 .4 5 (C,1) Belgium -1 .1 7 (C,1) -3 .2 6 * (C,0) -1 .4 5 (C,1) -3 .0 0 (T,0) -1 .2 1 (C,1) -2 .9 5 (C,0) -2 .6 7 (T,1) -3 .3 6 (T,1) Denmark -1 .9 8 (C,1) -2 .7 8 <C,0) -2 .9 3 (T,3) -2 .3 6 (C,3) -2 .2 3 (T,1) -0 .8 9 (T,0) -2 .0 7 (T,2) -5 .3 1 * (T,3) Germany -2 .1 9 (T,3) -1 .9 3 (C,0) -2 .6 1 (T,3) 1.73 (C,2) -4 .1 1 * (C,0) -2 .2 4 (C,0) NA NA Netherlands -1 .6 4 (C,1) -4 .6 2 * (C,0) -2 .0 3 (C,0) -1 .7 2 (C,0) -1 .9 5 (C,1) -2 .1 6 (T,0) -1 .0 9 (C,0) - 1 .7 7 (C,0) Switzerland -2 .2 2 (T,1) -1 .6 7 <C,1) -3 .0 7 * (C,2) 0.96 (C,2) -1 .7 9 (C,1) -3 .8 2 * (T,1) -1 .4 9 (C,3) -2 .3 2 (T,0) United States -1 .3 3 (C,1) -2 .6 5 (T,0) -2 .2 8 (T,1) -2 .2 5 (C,1) -1 .3 1 (C,1) —3.01 (T,1) -2 .8 6 (T,1) -1 .9 3 (C,1) Note: The entries show the relevant test statistic; the information in parentheses indicates the use of a constant only, C, or a constant and trend, T, followed by the number of lagged depen dent variables included. For the longest sample period used, the 5 percent significance level critical values are -3 .5 6 and -2.96, with and without the inclusion of a trend, respectively. Table 2 ADF Unit Root Results (growth rates) Country P Y M1 V1 MB VB E ER Austria -2 .1 9 (C,0) -4 .6 0 * (T ,0) -4 .7 2 * (C,0) -6 .3 3 * (C,0) -1 .5 4 (C,0) -4 .5 2 * (C,0) -4 .2 1 * (C,0) -3 .2 5 * (C,0) Belgium -2 .2 1 (C,0) —5.06* (T,0) -3 .0 9 * <C,0) -5 .2 8 * (C,0) -2 .4 3 (C,0) -5 .5 3 * (C,1) -3 .8 2 * (C,0) -3 .8 4 * (C,0) Denmark -1 .0 6 <C,0) -5 .6 6 * (T,0) -5 .3 4 * (C,0) -5 .6 8 * (T,0) -2 .4 7 (C,0) -3 .5 8 * (C,0) -4 .0 8 * (C,0) -4 .5 9 * (C,0) Germany -2 .1 5 <C,0) -3 .8 2 * (C,0) -4 .4 2 * (C,0) -5 .5 8 * (T,1) -3 .9 0 * (T,0) -4 .0 7 * (C,0) NA NA Netherlands -3 .2 6 (T,0) -3 .8 7 * (T,0) -3 .8 1 * (C,0) -5 .0 4 * (C,0) -2 .7 1 (T,0) -5 .6 8 * (T,0) -4 .7 0 * (C,0) -4 .0 1 * (C,0) Switzerland -2 .7 5 (C,0) -3 .0 1 * (C,0) -6 .0 7 * (T,1) -6 .1 6 * (C,1) -2 .9 6 (C,0) -4 .5 5 * (C,1) -5 .4 8 * (C,0) -6 .0 5 * (C,0) -1 .6 6 (C,0) -3 .9 6 * (C,0) -3 .8 1 * (T,1) -3 .8 9 * (T,1) -2 .8 3 (C,0) -4 .1 7 * (C,0) -3 .8 6 * (C,0) -3 .6 9 * (C,0) United States Note: The entries show the relevant test statistic; the information in parentheses indicates the use of a constant only, C, or a constant and trend, T, followed by the number of lagged depen dent variables included For the longest sample period used, the 5 percent significance level critical values are - 3 .5 7 and -2 .9 6 , with and without the inclusion of a trend, respectively RESERVE BANK OF ST. LOUIS FEDERAL 19 (11) GAP2 = (p - p * 2 > = (InVB-InVB*) - ( I n Y - l n Y *), Table 3 ADF Unit Root Results: Own Price Gaps Based on M1 and MB________ Country__________________ GAP1 _______________GAP2 Austria -3 .1 4 * (C,0) -3 .3 9 * (C,1) Belgium -4 .8 3 * (C,1) -3 .6 8 * (C,1) Denmark -2 .7 5 (C,0) -3 .9 1 * (C,3) Germany -3 .1 7 * (C,3) -3 .2 5 * (C,1) Netherlands -3 .4 4 * (C,1) -2 .9 9 * (C,0) Switzerland -6 .4 7 * (C,1) -3 .9 9 * (C,1) United States -4 .3 9 * (C,1) -4 .2 8 * (C,1) Note: The entries show the relevant test statistic; the in formation in parentheses indicates the use of a constant only, C, or a constant and trend, T, followed by the num ber of lagged dependent variables included. The critical value for 5 percent significance level is - 2.96, for the longest sample period used. the m easurem ent and inclusion of foreignbased price gaps. The Dom estic Price M odel The Hodrick-Prescott filter is used to deter mine equilibrium time paths fo r output QnY), narrow money velocity (InVl) and broad money velocity (InVB).19 Subsequently, two domestic price gaps are computed for each country: (10) GAP' = (p - p = *') U n V l- ln V l* ) - (InY -InY *), 19Basically the trend is derived by minimizing an objective function that consists of the sum of squared deviations of actual observations from the (unobservable) trend and a multiple, A, times the sum of squared changes of this trend. A smoothing factor A of 100 is used here, following Kydland and Prescott’s (1989) suggestion that this value is appropriate for annual data. Hodrick and Prescott (1981) show, using quarterly data, that a choice of A of up to four times or one-fourth as large has no practical effect on the results of applying the filter. The limiting case, A approach ing infinity, is a linear trend; this case was also examined (for Y, V and the real exchange rate below). Differences arising from the use of linear trend filters are noted below because the results are sensitive to this choice. w here p * 1 equals (InMl + ln V l* - I n Y * ) , and p *2 equals (lnMB + In V B *-In Y *), the respective meas ures of the equilibrium price level based on domestic M l and broad money. Table 3 shows ADF test results for these price gaps. The layout follows that of Tables 1 and 2. Overall, the price gaps appear to be stationary, with the ex ception of the M l price gap for Denmark. This Danish price gap is stationary only at the 10 percent level. Stationarity of the price gap is a necessary condition for the fu rth er analysis. Short-run inflation dynamics are theoretically assumed to be influenced by the price gap b e cause of the existence of an underlying equilibrating adjustm ent process. If actual prices do not converge to the computed equilibrium prices, as is the case with nonstationary gaps, the fundamental P-star hypothesis that, in the long run, prices converge to these equilibrium m easures is rejected; no theoretical foundation exists for including such a m easure of the gap in the inflation specification. This is the case for the Danish M l-based P-star model.20 Table 4 presents estim ates of the domestic P-star model based on each country’s own M l (GAP1) and broad money (GAP2). In the general specification, one lag of the dependent variable and the lagged inflation level are included along w ith the gap and a constant. The reported results are for estim ates in which insignificant lagged inflation variables have been dropped from the general specification. The results are supportive of the P-star ap proach. Save for Denm ark, the price gap with respect to broad money is significant with the correct negative sign. The price gap calculated with narrow money is significant only for the United States, Switzerland and Denmark, and 20When linear trends are used to construct equilibrium measures for the broad money-based P-star variables, only the U.S. and Swiss gap measures are stationary; the domestic gap measures for Germany and for the four other countries based on these measures are not stationary. The domestic version of the P-star model based on these equilibrium estimates is rejected because these measures cannot be equilibrium levels. Thus, the failure to reject the P-star model in the text is conditional upon the method of estimating equilibrium output and velocity. The Swiss and U.S. models using linear trends do not fit the data as well as the estimates reported below. MAY/JUNE 1994 20 marginally so for Germany. Since D enm ark’s M l-based domestic price gap is nonstationary, this result in D enm ark may well be spurious. Thus, D enm ark’s evidence rejects the domestic P-star model, a result obtained by Hoeller and Poret (1991) as well. In the other countries, the broad money-based P-star model generally fits the data som ew hat b etter (judged by the adjusted-R2) and fails to reject the model (judged by the statistical significance of the negative coefficient on the price gap).21 For five of the seven countries there is a high coh eren ce betw een the narrow and broad money price gap; the exceptions are Austria and the Netherlands.22 Apparently, deviations from equilibrium levels fo r narrow and broad money velocity or, m ore precisely, deviations from equilibrium nominal GDP, usually are closely related. Based on the superior results for broad money in Table 4 and the close coherence of the broad and narrow gaps, only the broad money-based domestic gaps are used in the discussion of the open econom y model. To com pute this price gap, how ever, one needs a m easure of the equilibrium value of the real exchange rate (er*).23 Two alternative m easures o f the equilibrium real exchange rate are p re sented here. The first m easure assumes that the equilibrium real exchange rate is a constant, w hich is equivalent to assuming that purchasing pow er parity (PPP) holds in the long run. The second m easure is based on using the HodrickPrescott filter to find the equilibrium com po nent of the real exchange rate. This m easure is less restrictive and m ore consistent with the data, but the oth er assumption, the PPP-based m easure, has strong theoretical appeal.24 PPP-B a sed Measures Suppose that each real exchange rate is stationary and converges to a constant longrun equilibrium level.25 Then, the gap with this constant, equilibrium real exchange rate removed is (12') GAPf1 = [pd - {pr* + e)]. Calculating An Appropriate Foreign-Based Price Gap To exam ine the German influence on each of the five small European countries, the foreign (German) gap is defined as (12) GAPf = [pd - (p>* + e - er*)]. 2 Hoeller and Poret (1991) also report that the domestic 1 P-star model is rejected for Austria and the Netherlands when the Hodrick-Prescott filter is used, but not when linear trends are used to find equilibrium output and veloc ity. Their study uses semiannual data and, for the seven countries examined here, uses sample periods beginning in 1962 for the United States, 1963 for the Netherlands, 1964 for Denmark, 1969 for Austria and Germany, 1971 for Belgium and 1973 for Switzerland. The last data point is in 1989 in each country. Their results also differ in choosing the monetary aggregates used for each country based on each country’s target. These are generally broad meas ures, but not necessarily the broad measure used here. 22The correlation coefficient for the U.S. broad-money and narrow-money-based price gaps is 0.74 and that for the German broad-and narrow-based price gaps is 0.71. The correlations of the gaps based on narrow and broad domestic monetary aggregates are 0.79 in Belgium and Switzerland, and 0.83 in Denmark. In the Netherlands, this correlation is only 0.25 and in Austria it is only 0.09; neither of these is statistically significant at a 5 percent level. 23For an analysis of the U.S. impact on the six European countries, see the appendix to this article. 24A third measure was also investigated. It assumes that the measured real exchange rate is always the equilibrium value, so that all changes in the real exchange rate are Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER This gap is consistent with the pure theory of the m onetary approach to the balance of pay m ents.26 As could be expected, however, this gap is generally nonstationary. Relevant ADF test results are shown in the first column of Table 5. This gap m easure is generally not the p referred specification because the statistical permanent. With this measure, the foreign-based price gap is Germany’s domestic price gap, which is stationary (see Table 3). The inflation model using this real exchange rate assumption is consistently dominated by the fit of the model using the Hodrick-Prescott filtered real exchange rates. 25Although the analysis above and the evidence in Figure 2 suggest this assumption is incorrect, it provides a con venient and insightful benchmark. Tatom (1992) uses this assumption for Austria; the constant level of the exchange rate is also removed from the foreign gap measures, so that Austrian prices are hypothesized to equal a German P-star in equilibrium. This model is rejected, however, although Austrian inflation is found to be tied to such an equilibrium German inflation measure. These results could arise from ignoring the effects of real exchange rate move ments on the level of prices, but the same results—the ab sence of a tie of the level of prices to an equilibrium level, but a strong tie of inflation to an equilibrium inflation rate—occur for a German money (M3)-based P-star mea sure and German prices. 26For recent discussions of the evidence against PPP, see Coughlin and Koedijk (1990), Dueker (1993) and Huizinga (1987). 21 Table 4 Short-Run Inflation Equations Including Own Price Gaps Country C An_i Belgium -7 .7 8 (1.24) — -0 .2 5 (2.01) -14 .4 1 (1.41) — -0 .2 7 (2.59) — 1.23 (1.84) — Germany -0 .0 6 (0.26) — — -0 .0 6 (0.22) Denmark — — 1.55 (2.78) GAP1 -0 .3 2 (2.41) -0 .2 8 (2.23) 1.40 (2.21) 1.23 (2.05) 1.36 (2.38) Austria "-1 0.33 (1.89) -0 .3 9 (2.92) — — -1 1 .3 6 (2.15) — -1 0 .2 8 (1.86) 1.26 (3.02) Netherlands Switzerland United States — -0 .3 2 (3.22) 0.82 (1.29) 0.80 (1.40) — -0 .1 9 (1.71) -0 .1 9 (1.86) -1 1 .1 8 (1.32) -0 .4 4 (3.05) -0 .5 0 (4.09) -1 2 .8 3 (2.71) — 1.87 (2.64) 2.12 (3.53) — 0.07 (0.40) 0.01 (0.04) — — — — — — — — -2 1 .1 2 (4.13) — R2 SEE LM (4) CHOW (77) Last Period 0.137 1.242 0.71 0.76 92 0.221 1.180 0.64 2.29 92 0.148 1.610 0.89 0.75 91 0.355 1.400 1.53 2.52 91 — 0.111 1.381 0.53 0.72 91 -6 .4 8 (1.19) 0.014 1.454 0.27 1.01 91 0.260 1.069 0.46 1.36 91 0.511 0.860 2.57 2.36 91 0.080 1.775 0.43 1.98 92 0.242 1.612 0.61 5.43* 92 0.333 1.653 3.13* 1.07 92 0.523 1.400 0.89 0.78 92 0.349 1.027 0.40 1.46 92 0.408 0.978 0.68 1.58 92 GAP2_ i — -2 4 .1 7 (2.17) — -3 5 .2 9 (3.36) — -29.01 (4.82) — -4 2 .3 4 (2.84) — - 20.28 (4.63) — -2 1 .0 0 (4.66) Notes: LM (4) is a Breusch-Godfrey test on serial correlation of the residuals using four lags of the residual; it has a * 2(4) distribution. CHOW (77) is a test on parameter stability with the break point in 1977; it follows an F-distribution. evidence rejects the hypothesis that domestic prices have a long-run relationship to this m ea sure of the foreign-determ ined, equilibrium price level. The irrelevance of this m easure, ex cept for Denm ark, is consistent with the evi dence for real exchange rates in Table 1, which indicates that the real exchange rate is nonstationary in all cases except for Denmark. Equilibrium Real Exchange Rates f r o m the Hodrick-Prescott Filter The second alternative explicitly tries to find a statistical estim ate for the time path of the equilibrium real exchange rate. To this end, we w rite the real exchange rate as (13) er = e + pf - p d = er * + u, w here u is a stationary, unobservable erro r term . In this case, the actual real exchange rate is equal to its long-run value (er*) plus a transi tory deviation. Neither e r * nor u are observable. It is possible, how ever, to obtain an estim ate of the equilibrium com ponent of the real exchange rate, e r * , again using the Hodrick-Prescott filter. This equilibrium com ponent, er * , is substituted into equation 8 to obtain: (12") GAPf* = [pd - (pf* + e - er*)]. T he second column of Table 5 shows this gap to be stationary at the 5 percent level for Aus MAY/JUNE 1994 22 tria, Belgium, Denm ark and the Netherlands. For Switzerland, a unit root can be rejected only at the 10 percent level.27 A Comparison o f Alternative Foreign Gap Measures To com pare the alternative m easures of the foreign-determ ined domestic price gap, labeled GAP?', and GAPf2, respectively, equation 4 is reestim ated with each of these gaps replacing the domestic gap. Table 6 contains the coeffi cient on the gap, the absolute value of its f-statistic in parentheses, and the adjusted fi-squared of the equation in square b rackets.28 GAPf1 is only relevant for Denm ark, w here its stationarity and that of the real exchange rate are supported by the data; nevertheless, this gap has been included as a benchm ark for the other countries as well. All coefficients in Table 6 are of the co rrect sign. Judged both by sig nificance and the amount of explanatory power, GAPf2 outperform s the other m easure, except in Denm ark, w here GAPf' is better. Table 7 similarly contains results for regres sions with both the domestic price gap and the Germ an-based price gap included. The dynamic specifications are the same as those in Tables 4 and 6.29 In Austria, the com parison of explana tory pow er favors GAPf1 slightly, but GAPr' is nonstationary and, judged by the most relevant com parison of perform ance shown in Table 6, GAP^ again has m ore explanatory pow er. Over all, the evidence suggests that GAP^ is the p referred m easure fo r both em pirical and theo retical reasons, except in Denm ark, w here GAP1' is preferred . These gaps are shown in Figure 3 along with the domestic price gaps based on the broad m oney aggregate. 27When linear trends are used to find equilibrium real ex change rates, output and velocity, the foreign-based gaps are not stationary, indicating that this approach to deriving the foreign-based P-star measure is inappropriate. 28The dynamic specifications used for the results reported in Table 6 follow those used in Table 4, although different specifications could have been used in two cases without changing the qualitative results. For Denmark, the lag of inflation is statistically significant at a 10 percent level (f= -1.83) when GAP1 is used, but it is omitted in Table 6 to 2 facilitate comparison to the GAP'1 case and to the Table 4 results. When this term is included using GAP'1 and G AP1 , the comparable adjusted-R2 are 0.266 and 0.240, 2 respectively. For Switzerland, adding the first lagged de pendent variable is statistically significant with either for eign gap; in this case, the adjusted-R2 is 0.386 and 0.616 for GAP1 and GAP'2, respectively. 1 FEDERAL RESERVE BANK OF ST. LOUIS Table 5 ADF Unit Root Results: Price Gaps Relative to Germany (only MB) Country GAP’1 GAP*2 Austria -1 .5 2 (C,1) -2 .4 4 (T ,0) —3.72* (C,1) - 1 .4 4 (C,0) -0 .7 5 (C,0) -3 .0 1 * (C,1) -3 .3 6 * (C,1) -3 .6 0 * (C,1) -3 .0 2 * (C,1) -2 .7 2 (C,0) Belgium Denmark Netherlands Switzerland Note: The entries show the relevant test statistic; the infor mation in parentheses indicates the use of a constant only, C, or a constant and trend, T, followed by the num ber of lagged dependent variables included. For the lon gest sample period used, the 5 percent significance level critical values are -3 .5 7 and -2 .9 6 , with and without the inclusion of a trend, respectively. The Impact o f Foreign-Based Price Gaps Table 8 restates the results of estimating equa tion 4 using the m ost appropriate foreign price gap for each country. For each of the five, small, European countries, Table 8 contains an equation with only the German-based gap in cluded (from Table 6) and an equation with both the domestic gap and the German-based gap (from Table 7). Comparing the results of Tables 4 and 8, a num ber of findings em erge: 29For the Swiss equation, using G APf1, a lagged dependent variable is significant at a 10 perent level (f = 1.85), but it is omitted in the table to facilitate comparison with the equa tion containing the second foreign gap measure in which the lagged dependent variable is not statistically signifi cant. When this lagged dependent variable is included with each foreign gap measure, the relevant adjusted-R2 is 0.624 for the first foreign gap measure and 0.680 for the second foreign gap measure, so the comparison of the two remains unaffected. 23 Table 6 Table 7 Comparison of the Impact of Two German-Based Price Gaps_______ Comparison of the Impact of Two German-Based Price Gaps, including Own Price Gap GAP*1 Austria Belgium Denmark Netherlands Switzerland GAP « -1 1 .0 7 (3.86) [0.406] -1 9 .6 4 (3-91) [0.411] -2 .1 5 (0.58) [0.097] GAP2 GAP'1 GAP2 GAP*2 Austria - 24.02 (3.52) [0.373] -1 4 .1 5 -9 .8 1 (1.41) (3.32) [0 .427] -1 3 .1 8 -1 7 .3 9 (1.30) (3.31) [0.425] Belgium -1 6 .7 3 (3.00) [0.216] -1 8 .8 9 (2.68) [0.176] - 35.67 -2 .6 2 (3.37) (0.83) [0 .348] -3 0 .1 9 -2 0 .8 0 (3.39) (3.54) [0.548] Denmark -1 0 .7 3 (2.21) [0.168] - 27.79 (3.23) [0.288] -1 1 .0 6 -2 0 .0 8 (2.38) (3.75) [0 .328] -1 0 .4 0 -2 2 .5 3 (2.16) (3.29) [0.271] Netherlands -4 .5 2 (2.32) [0.293] -2 7 .2 9 (5.35) [0.583] -3 5 .7 9 -7 .8 3 (2.39) (1.68) [0 .288] -2 4 .7 2 -2 0 .0 7 (1.49) (2.03) [0.318] Switzerland -1 8 .9 6 -3 .5 8 (4.64) (2.40) [0. 592] -11 .8 1 -1 9 .1 9 (2.66) (3.47) [0.658] Note: In each cell, the top entry is the coefficient for the gap, the middle entry is the absolute value of the f-statistic and the R2 is the lowest entry given in brackets. • The German-based gap provides greater ex planatory pow er than the cou ntry’s own domestic price gap w hen each is considered alone. • Table 8 indicates that adding the Germanbased gap to a specification already containing the domestic gap (Table 4), on the other hand, always leads to a statistically significant im provem ent in the inflation model. • In the case of Denm ark, the addition of the foreign gap means the d ifference betw een rejecting the P-star model and not doing so. The closed econom y model rejects the P-star model in Denm ark, but the open econom y model does not. • Adding the domestic gap to the specification already containing the German-based gap 30The appendix shows that U.S.-based price gaps have had little impact on European inflation developments over the sample, regardless of the measure used. This provides ad ditional support for our hypothesis that the exchange rate regime determines which price gaps are relevant, that is, to what “ equilibrium” measure of prices, foreign or domes tic, will actual domestic prices converge. Under floating rates, it should be domestic price gaps that matter, while under fixed rates, the foreign influence will increase. It is leads to a significant im provem ent in the cases of Belgium, Denm ark and Switzerland. • Overall, the results are supportive of the hypothesis that the domestic price gap is of little im portance under a regim e of fixed ex change rates, but that, instead, cu rren t infla tion developments at home are determ ined by m onetary conditions abroad.30 A com parison across countries reinforces these conclusions. Austria, for instance, has been most closely linked to Germany over most of the sample, followed by the N ether lands. Consequently, no significant additional inform ation is provided by their own domes tic price gap, once account has been taken of Germany’s impact. For Belgium and Denmark, on the oth er hand, both the domestic price gap and the German-based price gap are imtrue that for a few years early in the period studied here, all these countries were pegged to the dollar, but the P-star influence of U.S. prices due to this experience is not statistically significant. Presumably, the period of the dollar peg in this sample is too brief for the dollar-based gap to be significant. MAY/JUNE 1994 24 Figure 3 Domestic and Foreign-Based Price Gaps Austria Belgium Denmark* *The average value o f the log o f the real m ark exchange rate for Denmark was added to the foreign gap. FEDERAL RESERVE BANK OF ST. LOUIS 25 Figure 3 (continued) Netherlands Switzerland portant. This may reflect the difficulties these two countries have experienced during the '70s and ’80s in keeping their cu rren cies’ values and their inflation rates in line with Germ any’s. By infrequent devaluations, they have allowed their own m onetary policy—and related domes tic price gaps—to affect dom estic inflation. Through their continued efforts to converge to German inflation levels over time, however, German price gaps have m attered as well. The m ost interesting set of results is for Switzerland. In contrast with the oth er four small European countries under consideration, Switzerland has followed a floating exchange rate policy since the breakdow n of Bretton Woods. As a result, the impact of German-based price gaps should be insignificant, according to the hypothesis. Our estim ates, how ever, suggest that the German-based price gap has dominated the domestic Swiss gap in the sense that the form er has m ore explanatory pow er, considered alone, than the latter. Although far from conclusive, th ere are some possible reasons fo r this apparent anomaly. First, m onetary policies in Switzerland and Germany have been quite sim ilar during much of the sample period. Both countries faced similar infla tionary pressures towards the end of Bretton MAY/JUNE 1994 26 Table 8 Short-Run Inflation Equations Including a German-Based Price Gap GAP*2. , R2 SEE LM (4) CHOW (77) Last period — -1 9 .6 4 (3.91) 0.411 1.026 1.50 1.33 92 - 0 .3 7 (3.33) -1 3 .1 8 (1.30) -1 7 .3 9 (3.31) 0.425 1.014 1.15 1.83 92 1.46 (2.59) - 0 .3 0 (2.84) — -2 4 .0 2 (3.52) 0.373 1.381 0.77 3.05* 91 1.45 (3.01) - 0 .2 9 (3.32) -3 0 .1 9 (3.39) - 20.80 (3.54) 0.548 1.172 1.61 1.33 91 Country C "-1 Austria 1.72 (3.24) -0 .3 9 (3.52) 1.62 (3.06) Belgium Denmark GAP2, , — — -1 6 .7 3 (3.00) 0.216 1.296 1.73 0.34 91 - 26.48 (3.76) Netherlands -2 2 .0 6 (3.01) — -1 1 .0 6 (2.38) -2 0 .0 8 (3.75) 0.328 1.200 1.96 0.44 91 -0 .2 3 (2.30) — -2 7 .7 9 (3.23) 0.288 1.562 0.78 1.94 92 0.94 (1.71) -0 .2 2 (2.23) -2 4 .7 2 (1.49) -2 0 .0 7 (2.03) 0.318 1.529 0.34 2.78 92 2.42 (4.27) -0 .5 5 (4.83) — - 27.29 (5.35) 0.583 1.306 1.49 0.83 92 2.38 (4.63) Switzerland 1.00 (1.78) -0 .5 5 (5.31) -11.81 (2.66) -1 9 .1 9 (3.47) 0.658 1.184 1.39 0.52 92 Notes: For Denmark, the foreign gap used is GAP*1 instead of GAP'2. LM (4) is a Breusch-Godfrey test on serial correla tion of the residuals using four lags of the residual; it has a x2(4) distribution. CHOW (77) is a test on parameter stability with the break point in 1977; it follows an F-distribution. Woods, and implem ented similar m onetary ta r geting policies in the mid-'70s to reduce inflation. Second, the Swiss fran c and the German mark have been attractive—and closely substitutable— investm ent cu rren cies in international portfolios. T he Swiss results could also be interpreted as stemming from close coordination of m onetary policies under floating exchange rates.31 Similar tests w ere conducted for the United States and German domestic P-star models to ex amine the pow er of the tests of the significance of foreign gaps reported here. In particular, fo r eign gaps constructed like the gap m easure GAPf2, using the five European countries, w ere constructed and added to the domestic P-star model for the United States and Germany. Also, 31This is not equivalent to fixing nominal exchange rates, as evidenced by the appreciation of the Swiss franc during the time period considered (see Figures 1 and 2). FEDERAL RESERVE BANK OF ST. LOUIS a German-based gap was added to the domestic P-star model for the United States. In no case was one of the foreign gap term s statistically significant in the dom estic model fo r the United States or Germany. This strengthens the evi dence that in floating countries the appropriate P-star model is a domestic one, while the domestic P-star is determ ined by the anchor country in a fixed rate regime. SUMMARY AND CONCLUSION The systematic link betw een domestic money and the general level of prices is central to the P-star model, w hich emphasizes this long-run 27 relationship as a determ inant of short-run move m ents in the level of prices and inflation. M one tary authorities in countries with fixed exchange rate regim es do not determ ine th eir own long-run level of prices, however. Instead, their long-run equilibrium price level is im ported from the countries w hose curren cy is the basis of the peg. To varying degrees five, small open European countries have pegged their cu rren cy to the German m ark. Econom ic theory suggests that, to the degree they did so, these countries’ longrun equilibrium price levels and their inflation rates should be dominated by the German price developments, w hich, in turn, are presum ably controlled by the Bundesbank. An open econom y model of inflation in countries with fixed ex change rates must take into account the external basis of the equilibrium price level. evidence shows that the long-run equilibrium price level tow ard w hich domestic prices adjust is determ ined by foreign m onetary policy. This article develops such a P-star model for domestic prices from 1960 to the 1990s in five European countries: Austria, Belgium, Denm ark, the Netherlands and Switzerland. These econo mies b ord er Germany and have, to varying degrees, fixed their domestic exchange rates based on a peg to the German m ark. The evi dence presented h ere shows that the open econ omy, fixed exchange rate P-star model is not rejected for the countries considered. The infla tion model’s fit improves for all five countries when allowance is made for the statistically sig nificant foreign (German) influence on equilibri um domestic price levels during fixed exchange rate periods. Dewald, William G. “ Monetarism is Dead; Long Live the Quantity Theory,” this Review (July/August 1988), pp. 3-18. Perhaps the best example is Denm ark, w here the domestic P-star model is rejected. In the open econom y model, how ever, the broad money-based domestic gap and the German P-star-based gap are both highly significant and with the co rre ct sign, showing the im portance of accounting for the foreign influence. Two other countries in w hich the domestic gap is significant in tests of the open econom y model are Belgium, the oth er interm ediate case, and Switzerland. In Austria and the Netherlands, w here currencies have been m ost tightly pegged to the m ark, the German-based P-star model outperform s the respective domestic models and, w hen included with the domestic gap, the domestic gap is not statistically significant. Overall, the results confirm the long-run link betw een m onetary aggregates and domestic prices for both closed or large, flexible exchange rate countries, as well as for fixed exchange rate countries. In the latter case, how ever, the REFERENCES Batten, Dallas S., and Courtenay C. Stone. “Are Monetarists an Endangered Species?,” this Review (May 1983), pp. 5-16. Bayoumi, Tamim, and Barry Eichengreen. “ Macroeconomic Adjustment Under Bretton Woods and the Post-BrettonWoods Float: An Impulse-Response Analysis,” NBER Work ing Paper No. 4169 (1992). Coughlin, Cletus C., and Kees G. Koedijk. “ What Do We Know About the Long-Run Real Exchange Rate?," this Review (January/February 1990), pp. 36-48. Deutsche Bundesbank. “ The Correlation Between Monetary Growth and Price Movements in the Federal Republic of Germany,” Monthly Report of the Deutsche Bundesbank (January 1992), pp. 20-8. Dueker, Michael J. “ Hypothesis Testing with Near-Unit Roots: The Case of Long-Run Purchasing-Power Parity,” this Review (July/August 1993), pp. 37-48. Dwyer, Gerald P., Jr., and R.W. Hafer. “ Is Money Irrelevant?,” this Review (May/June 1988), pp. 3-17. Hallman, Jeffrey J., Richard D. Porter, and David H. Small. “ Is the Price Level Tied to the M2 Monetary Aggregate in the Long Run?” The American Economic Review (Septem ber 1991), pp. 841-58. Hoeller, Peter, and Pierre Poret. "Is P-Star a Good Indicator of Inflationary Pressure in OECD Countries?,” OECD Econom ic Studies No. 17 (autumn 1991), pp. 7-29. Hodrick, Robert J., and Edward C. Prescott. “ Post-War U.S. Business Cycles: An Empirical Investigation,” Northwestern University Discussion Paper No. 451 (May 1981). Huizinga, John. “An Empirical Investigation of the Long-Run Behavior of Real Exchange Rates,” Carnegie Rochester Conference Series on Public Policy (autumn 1987), pp. 149-214. Humphrey, Thomas M. “ Precursors of the P-Star Model,” Federal Reserve Bank of Richmond Economic Review (July/August 1989), pp. 3-9. King, Robert G., and Sergio T. Rebelo. “ Low Frequency Filtering and Real Business Cycles,” University of Rochester, Center for Economic Research Working Paper No. 205 (October 1989). Kool, Clemens J.M. “ The Case for Targeting Domestic Money Growth Under Fixed Exchange Rates: Lessons from the Netherlands, Belgium, and Austria: 1973-1992,” METEOR research memorandum 94-011 (February 1994), University of Limburg, Maastricht. Kydland, Finn E., and Edward C. Prescott, “A Fortran Subrou tine for Efficiently Computing HP-Filtered Time Series,” Federal Reserve Bank of Minneapolis research memoran dum (April 1989). Mills, Terence C., and Geoffrey E. Wood. “ Does the Ex change Rate Regime Affect the Economy?,” this Review (July/August 1993), pp. 3-20. Organization for Economic Co-operation and Development, National Accounts, Main Aggregates, vol. 1, 1960-91. OECD, 1993. MAY/JUNE 1994 28 Roubini, Nouriel. "Offset and Sterilization Under Fixed Ex change Rates with an Optimizing Central Bank” NBER Working Paper No. 2777 (1988). Stockman, Alan C., and Lee E. Ohanian. “ Short-Run In dependence of Monetary Policy Under Pegged Exchange Rates and Effects of Money on Exchange Rates and In terest Rates,” University of Rochester, Center for Economic Research Working Paper No. 361 (September 1993). Tatom, John A. “ The P-Star Model and Austrian Prices,” Empirica, vol. 19 (1992), pp. 3-17 ________“ The P-Star Approach to the Link Between Money and Prices,” Federal Reserve Bank of St. Louis Working Paper 90-008 (1990). Appendix Are European Prices Influenced By U.S. Monetary Policy? The analysis in the text focuses on the con nections betw een Germany and a num ber of small European countries with strong ties to Germany. A similar analysis can, of course, be applied to the United States in relation to the six European countries. Our maintained hypothesis suggests that under the floating ex change rate regime for the U.S. dollar over most of the period, European countries should have been insulated from inflationary or defla tionary pressures arising from the United States. For the '60s and early ’70s, on the other hand, U.S.-determined price gaps should have influenced Europe, because the United States was the anchor country in the fixed exchange rate system of Bretton Woods. Figure 1 shows, however, that the relevant Bretton Woods period in our sample has been too short to perform a meaningful test of the significance of U.S.-determined gaps. Our regressions start in 1962 or 1963, depending on the lags included, and nominal exchange rates start moving in 1967-68, thereby reducing the potential impact of U.S. m onetary conditions abroad. Thus, for our sample, we do not expect coefficients on U.S.-determined gaps to be sig nificantly different from zero. First, we present test statistics on the stationarity of the relevant gap variables in Table A l. Gap definitions are similar to those in the main text, with Germany replaced by the United States. The results are very close to those in Table 5, w here unit root statistics for German-based gaps are displayed. GAPfl is nonstationary, while GAP& is stationary. Tables A2 and A3 are com parable to Tables 6 and 7, respectively, and have the same layout. Estimated gap coefficients are generally small in magnitude and insignifi FEDERAL RESERVE BANK OF ST. LOUIS Table A1 ADF Unit Root Results: Price Gaps Relative to U.S. (only MB)________ Country GAP’ 1 G AP'2 Austria -2 .7 0 (T.1) -3 .3 5 * (C,1) Belgium -2 .6 7 (C,1) -3 .4 6 * (C,1) Denmark -1 .9 4 (C,1) -3 .3 2 * (0,1) Germany -1 .8 3 <C,1) -3 .4 3 * (C,1) Netherlands -1 .5 6 (C,1) -3 .0 8 * (C,1) Switzerland - 0 .7 7 <C,0) -3 .2 0 * (C,1) Note: For the longest sample period used, the 5 percent significance level critical values are - 3 .5 6 and -2 .9 6 , with and without the inclusion of a trend, respectively. cant. Only fo r Belgium are small significant coefficients found fo r GAPf1 and GAP& in Table A3. Overall, the evidence rejects a link from U.S. m onetary conditions to inflationary pres sures in Europe. This supports our hypothesis. To address the issue of the potential impact of the United States on other countries during the Bretton W oods period adequately, a longer sam ple going back to the early ’50s or late ’40s would be required. This is left for future research. 29 Table A2 Table A3 Comparison of the Impact of Two U.S.-Based Price Gaps Comparison of the Impact of Two U.S.-Based Price Gaps, Including Own Price Gap_________________ 0.21 (0.14) [0.09] 0.12 (0.05) [0.09] -1 .0 4 (0.91) [-0 .0 1 ] 0.42 (0.18) [-0 .0 3 ] - 0 .2 7 (0.27) [0.16] - 1 .0 4 (0.56) [0.17] -0 .7 4 (0.57) [0.03] - 1 .3 9 (1.41) [0.28] Belgium Denmark Germany1 Netherlands Switzerland1 Country GAP2 GAP'1 GAP2 GAP1 2 Austria -2 5 .6 7 (2.23) - 0 .5 7 (0.61) [0.20] -2 7 .8 6 (2.39) -2 .1 2 (1.04) [0.22] Belgium -5 1 .6 3 (4.34) -3 .5 3 (2.37) [0.45] -5 3 .8 3 (4.57) -5 .7 1 (2.65) [0.47] Denmark (1.02) CD - 0 .6 3 (0.30) [0.09] I o GAP'2 -0 .1 3 (0.13) [0.09] — i. GAP'1 I cn Country Austria (0.68) [0.00] -7 .7 2 1.52 (1.32) (0.62) [-0 .0 1 ] Germany -2 9 .9 7 (4.95) -3 0 .6 0 (5.23) 0.59 (0.20) [0.02] -0 .8 0 (1.08) [0.51] -2 .4 4 (1.79) [0.55] Netherlands -5 0 .4 0 (3.31) -5 4 .1 3 (3.20) -3 .3 5 (1.25) [0.27] -1 .9 6 (1.68) [0.29] -4 .1 4 (1.40) [0.27] Switzerland -19 .5 1 (4.34) -0 .6 7 (0.85) [0.52] -2 1 .8 0 (4.38) 1.64 (0.67) [0.51] ’ The specification includes a statistically significant lagged dependent variable which was not significant and not included in that used in Table 4. MAY/JUNE 1994 31 Charles W. Calom iris Charles W Calomiris is associate professor of finance, . University of Illinois at Urbana-Champaign; faculty research fellow, National Bureau of Economic Research; and visiting scholar at the Federal Reserve Bank of St. Louis. Greg Chaudoin, Thekla Halouva and Christopher A. Williams provided research assistance. The data analysis for this article was conducted in part at the University of Illinois and the Federal Reserve Bank of Chicago. Is the Discount Window Necessary? A Penn Central Perspective XN RECENT YEARS, ECONOMISTS have come to question the desirability of granting banks the privilege of borrow ing from the Federal Reserve’s discount window. The discount win dow’s d etractors cite several disadvantages. First, the Fed’s control over high-powered money can be ham pered. If bank borrow ing b e havior is hard to predict, open m arket opera tions cannot perfectly peg high-powered money, w hich some econom ists believe the Fed should do. Second, there are m icroeconom ic concerns about potential abuse of the discount window (Schwartz, 1992). Critics argue that the discount window has been misused as a tran sfer schem e to bail out (or postpone the failure of) troubled or insolvent financial institutions that should be closed quickly to prevent desperate acts of fraud or excess risk taking by bank m anagem ent. In response to growing criticism of Fed lending to prop up failing banks, Congress mandated limits on discount lending to distressed banks, which w ent into effect on D ecem ber 19, 1993. Some econom ists (Goodfriend and King, 1988; Bordo, 1990; Kaufman, 1991, 1992; and Schwartz, 1992) have argued that th ere is no gain from al lowing the Fed to lend through the discount window. These critics argue that open m arket operations can accom plish all legitim ate policy goals without resort to Federal Reserve lending to banks. Clearly, if the only policy goal is to peg the supply of high-powered money, open m arket operations are a sufficient tool. Similar ly, the Fed could peg interest rates on traded securities by purchasing or selling them . Any argum ent for a possible role fo r the discount window must dem onstrate that pegging the ag gregate level of reserves in the economy, or controlling the riskless interest rate on traded securities, is insufficient to accomplish a legiti m ate policy objective that can be accomplished through Fed discounting. In this article, I exam ine theoretical assump tions that may justify the existence of the dis count window. I argue that th ere is little cu rren t MAY/JUNE 1994 32 role for the discount window to protect against bank panics. The main role of the discount win dow is in defusing disruptive liquidity crises that occu r in particular n on ban k financial m ar kets. I discuss evidence from the Penn Central crisis of 1970, which seems consistent with that view, and conclude by considering w h eth er this evidence is relevant for today’s relatively sophisti cated financial environm ent. Backup protection for financial m arkets through the discount window could be achieved at little cost if access to the discount window w ere confined to periods of financial disruption. During norm al times, open m arket operations and interbank lending would be sufficient for determ ining the aggregate amount of reserves in the banking system and their allocation among banks. A first step tow ard envisioning a role for any financial institution or policy instrum ent, includ ing the discount window, must be the relaxation of the assum ptions of zero physical costs of transacting and/or sym m etric inform ation. The discount window’s benefit, if any, must be relat ed to its role in helping to econom ize on costs in capital m arkets, w hich them selves are a fun c tion of physical or inform ational "im perfections.” I divide the discussion of potential justifications for the discount window into two parts— assistance to financial interm ediaries and assistance to particular financial m arkets. THE DISCOUNT WINDOW AND BANKING PANICS The Federal Reserve System was created in 1913 with three prim ary objectives: to eliminate the "pyram iding” of reserves in New York City and replace it with a polycentric system of 12 reserve banks; to create a m ore seasonally elas tic supply of bank credit; and to reduce the propensity for banking panics. The discount 1lf money-demand disturbances were the cause of banking panics, as envisioned in Diamond and Dybvig (1983), then open market operations, as normally defined, would be a sufficient tool for policy if the central bank were permitted to purchase bank loans. Since bank loans are not “ spe cial” in that framework (that is, there is no delegated con trol and monitoring function performed by the banker and, hence, no potential for adverse selection or moral hazard), it is natural to think of standard open market operations as including purchases of bank debt in the context of that model. If, however, banking panics are produced by confu sion over the incidence of shocks to the value of bank as sets, as argued in Calomiris and Gorton (1991), and if RESERVE BANK OF ST. LOUIS FEDERAL window was the prim ary m echanism for achiev ing these goals. The 12 regional Federal Reserve Banks offered an alternative to the private in ter bank deposit m arket as depositories of bank reserves. T he architects of the Fed expected to eliminate reserve "pyram iding,” w hich chan neled interbank deposits to New York, w here they often w ere used to finance securities m ar ket transactions (White, 1983). Interbank lend ing was viewed by some as a problem because it encouraged dependency of the nation’s banks on New York bankers and placed funds into the hands of securities m arket speculators. The discount window also promised to reduce the seasonal volatility of interest rates and in crease the seasonal elasticity of bank lending by providing an elastic supply of reserves, allowing bank balance sheets to expand seasonally without increasing the loan-to-asset ratio. Prior to the creation of the Fed, bank expansion o f loans in peak seasons led to costly increases in portfolio risk (a higher loan-to-asset ratio), or costly seasonal im portation of specie. This implied an upward sloping loan supply function with large interest rate variation over the seasonal cycle (Miron, 1986; Calomiris and Hubbard, 1989; and Calomiris and Gorton, 1991). Finally, the availability of the discount window was also expected to reduce the risk of bank panics in two ways. First, by increasing the availability of reserves, the discount window limited seasonal increases in portfolio risk and reductions in bank liquidity during high-lending months, thus reducing the risk of panics. Sec ond, the discount window would provide a source of liquidity to banks if an unpredictable withdrawal of deposits in the form of curren cy created a shortage of reserves that threatened the liquidity of the banking system (as in Dia mond and Dybvig, 1983).1 But the discount win dow offered limited protection to banks from a panic induced by adverse econom ic news. Because access to the discount window was banks have special information about their portfolios, then a government policy of purchasing bank loans during a cri sis at pre-panic prices would have the same costs and benefits as allowing banks access to the discount window. 33 limited by strict collateral requirem ents, bank borrow ing was limited to the amount of eligible collateral the bank possessed.2 Thus, Federal Reserve Banks could not use the discount win dow to shore up banks if their depositors lost confidence in the quality of the bank’s illiquid loan portfolio. The collateral required for discount window lending was subsequently broadened in the 1930s. The history of the pre-Fed era suggests that the early limitations on discount window lend ing w ere im portant. Gorton (1989), Calomiris and Gorton (1991), and Calomiris and Schweikart (1991) have argued that sudden w ithdraw als from the banking system occu rred w hen depositors received news about the state of the econom y that was bad enough to make them think that some banks w ere insolvent. Because depositors w ere uninform ed about the incidence of this disturbance across individual banks (be cause of depositors’ limited inform ation about bank portfolios) all banks’ depositors had an incentive to withdraw funds from their banks until they could b etter ascertain the risks of in dividual banks. Thus, relatively small aggregate insolvency risk could have large costs through disinterm ediation from banks. Costs associated with banking panics can moti vate a m ore aggressive policy than one req u ir ing riskless collateral for all central bank lending. The central bank could provide loans to the banking system on illiquid collateral to offset the tem porary withdrawal of depositors’ funds. The rationale for this intervention lies in in for mational externalities caused by panics. Banking panics create negative externalities among banks and their custom ers. Banks whose assets have not declined in value, and their b orrow ers and depositors, su ffer because of the confusion over w hether they are among the banks holding lowvalue assets. The banks lose business, the b o r row ers lose access to credit, and the depositors lose interest and pay transaction costs of tran s ferring funds out of the banking system. Banks and their borrow ers benefit by keeping the banking system from shrinking. If bank credits and deposits play special roles in financing and clearing transactions, then con tractions in bank activity will be costly. The dis count window can be thought of as a way to coordinate a mutually beneficial decision among depositors not to w ithdraw their deposits during panics. Removing the private incentive fo r depo sitors to w ithdraw th eir funds makes all deposi tors better off. W hile private deposits fall, public “deposits” made through the discount window (the indirect assets of the public) rise to com pensate. Open m arket operations would not be an adequate substitute policy. Open m arket operations would simply insulate the money supply from the reduction in the money mul tiplier as bank deposits and bank credit fell; they would not reduce withdrawals from banks. Thus, one could argue for central bank adop tion of the following rule for use of a “backup” discount window: Under norm al circum stances (when there is no general systemic banking pan ic reducing private deposits in banks), the cen tral bank provides no loans to banks. During a systemic crisis, the central bank agrees to pro vide loans to banks up to the amount of deposi tor withdrawals (at an interest rate that fairly com pensates the governm ent fo r the default risk of the average bank). Such crisis loans must be short-term and paid in full after the crisis passes (which, if history is any guide, should be no longer than two months). The governm ent might increase the interest rate it charges on loans to banks over time to encourage them to assist in resolving the inform ation asymmetry m ore quickly (for example, by sharing inform a tion about themselves and one another). The central bank might even charge a fee to banks ex post as a function of actual losses, to fu rth er encourage good banks to bring the crisis to a 2These limitations were eliminated in the 1930s. For a dis cussion of changing collateral requirements on Fed lend ing, see Friedman and Schwartz (1963, pp. 190-5). Note that lending from the Fed, even on riskless collateral, can provide special assistance to banks (up to the amount of their riskless collateral) because the Fed enjoys a special right to “jump the queue” of debt seniority. By taking the best assets of the bank as collateral, the Fed effectively subordinates existing debt claims. Private creditors would not be able to do so and, thus, would not be able to lend to the bank on the riskless collateral. MAY/JUNE 1994 34 speedy conclusion.3 As deposits retu rn to the banks, they would use them to repay the governm ent loans. Banks that fail to attract depositors (relative to other banks) as the crisis draw s to a close would be denied continuing ac cess to credit and would be allowed to fail. In principle, banks might be able to prevent panics by pooling resources privately without any intervention by the central bank. If the banking system w ere able to allocate funds to insure against banking panics by agreeing to treat deposits as a collective liability of all banks during a systemic crisis (as some groups of banks did historically), then, so long as the pub lic was confident of the aggregate solvency of the banking system, th ere would be no threat of systemic bank runs and no need for a governm ent-run discount window to reduce the costs of banking panics.4 Kaufman (1991) argues that interbank m arkets did not operate e ffec tively historically, but that this is no longer the case. He claims that the existence of the m odern federal funds m arket obviates the need fo r the discount window during crises because open m arket operations, com bined with in ter bank tran sfers, can funnel cash to w hichever solvent banks experience large withdrawals. If banks as a group are willing to pool their governm ent security holdings during a crisis, then Fed purchases of securities com bined with interbank tran sfers to banks that lack sufficient governm ent securities can keep the banking sys tem afloat, and possibly prevent runs (if in ter bank insurance is credible ex ante). Despite the existence of a delivery m echanism (the fed funds market), lending among banks during a crisis may not occu r due to asym m et ric inform ation. If banks are unable to regulate and credibly m onitor each o th er’s portfolios and behavior, they will be reluctant to insure one another during a banking panic. Even though the interbank m arket operates quite well during norm al tim es among most banks, it cannot 3There must be an implied “ subsidy” relative to the terms by which private lenders would be willing to lend to the bank, or else government lending cannot prevent runs. The actuarialy fair government lending will be lower than the rates banks would pay in the private market, since govern ment intervention reduces default risk. 4Calomiris (1990, 1992c) argues that a nationwide branch banking system would not have experienced aggregate in solvency risk even during the worst episodes of bank failure and bank panic. FEDERAL RESERVE BANK OF ST. LOUIS necessarily be relied upon to protect the bank ing system from panics. The interbank loan m arket can operate effec tively so long as banks have adequate inform a tion about and control over each oth er’s actions. Lending banks m ust be confident that b orrow ers are not abusing the interbank m arket to subsidize excessive risks o r provide a bailout to insider depositors of a failed bank. Although this "incentive com patibility” requirem ent may be difficult to satisfy, th ere are many examples that show it is possible to do so. Gorton (1985, 1989), Calomiris (1989a), Calomiris and Kahn (1990, 1991), and Calomiris and Schw eikart (1991) argue that inform ation asym m etry about bank borrow ers and the consequent risk of panics prompted cooperative behavior among banks historically. Coordination among banks in response to panics characterized many coun tries’ banking systems (notably England's during the Baring Crisis of 1890, and Canada's repeatedly during the 19th and 20th centuries). But in the United States, laws limiting bank branching and consolidation effectively limited interbank cooperation. As the num ber of U.S. banks and their geographical isolation from one another in creased, the feasibility of national cooperation was undermined. A bank’s cost of monitoring and enforcing cooperative behavior rises with fragm entation, while the benefit to any bank from m onitoring and enforcing declines with the num ber of m em bers in the coalition (the benefit is shared by all). Thus, the need for discount window assistance to banks is magnified by unit banking laws that make private interbank cooperation, lending and mutual insurance infeasible. Absent such regulations, the potential for costly banking pan ics would be substantially reduced, and the ex pected benefits of discount window protection of the banking system would be sm aller.5 In closing, four points are w orth noting. First, I have not assumed that the governm ent has 5See the related discussion of other countries’ experiences in Bordo (1990) and Calomiris (1992a). 35 superior inform ation regarding individual bank solvency—an alternative justification for govern m ent lending to banks even in noncrisis states. W hile such an argum ent can be made (based on the governm ent’s access to inform ation by vir tue of its supervisory role), the recen t history of bank failures and losses, and of regulatory agen cies’ inabilities to anticipate, observe or prevent widespread abuse seems to argue against such a presumption. Kane (1988) argues that regulators face distorted incentives to collect and report inform ation about banks. These incentive prob lems may outweigh regulators’ special channels of inform ation due to supervisory authority. Second, discount lending can be motivated by physical transaction costs that limit interbank lending. Such physical costs m ean that open m arket operations will have uneven effects on the supply of reserves available to different banks if the m arket fo r reserves is segmented. Although this may have been a legitimate moti vation fo r the discount window historically, as Kaufman (1991) argues, cu rren t interbank reserve tran sfers are accom plished at little cost. Third, I have not addressed the possible role o f the discount window in bailing out a banking system that is insolvent as a whole. Even in a concentrated, mutually insuring banking system, interbank insurance and lending could never deal with enorm ous adverse asset shocks (that is, those larger than aggregate bank capital). Partial governm ent deposit insurance (with large deductibles) for mutually insuring groups of 6lt is beyond the scope of this article to examine all of the relative advantages of government deposit insurance or discount lending for stabilizing a fragmented (uncoordinated) banking system. Perhaps the most obvious potential advan tage of discount window lending is that government inter vention can be state-contingent. If a bank fails when there is no systemic panic, the bank’s depositors will not be bailed out by government insurance. This reduces the moral-hazard costs of the government’s “ safety net.” This argument for the relative desirability of the discount window as a means to insure against panics presumes that the central bank will not cave in to the political pressures of special interests to bail out banks in noncrisis times. Recent accusations by the House Banking Committee of inap propriate lending by the Federal Reserve to insolvent banks cast some doubt on the ability of current institutions to make and enforce appropriate distinctions regarding when banks should have access to the discount window (see Business Week, July 15, 1991, pp. 122-3). Schwartz (1992) argues that the history of the discount window is replete with such examples. Congress has mandated, and the Fed has implemented, specific new guidelines that limit Fed lending to distressed banks (The American Banker; August 12, 1993, pp. 1-2). banks can protect against this unlikely event b etter than w holesale bailouts through discount window "lending” (Calomiris, 1992b). Fourth, the need fo r the discount window to protect the cu rren t U.S. banking system from financial panic has been substantially curtailed by deposit insurance.6 Under the cu rren t deposit insurance system, discount window intervention would be largely redundant as protection against systemic risk. Insured depositors have little incen tive to run their banks during a financial crisis. Although deposit accounts in excess of $100,000 under cu rren t law are not protected (de jure) by governm ent deposit insurance, larger deposits may be covered if a general run on the banking system ensued. The FDIC Im provem ent Act (FDICIA) of 1991 establishes a form ula for determining w h eth er a system ic threat w arrants the coverage of larger-denom ination deposits.7 Fed lending does retain a potentially im portant role in providing implicit protection for the in ter bank clearing system, w hich is discussed below .8 NONBANK LENDING AND THE RO LE O F THE DISCOUNT WINDOW In an econom y in w hich physical costs of in terbank tran sfers are small, and interbank coor dination and mutual insurance, or governm ent deposit insurance, protects the banking system from the risk of panic, there is no additional need for the discount window to facilitate the in 1995, the FDIC, the Secretary of the Treasury (in consul tation with the President), and a supernumerary majority of the boards of the FDIC and the Federal Reserve, must agree that not doing so “ would have serious adverse ef fects on economic conditions or financial stability.” If unin sured deposits are covered through this provision, the insurance fund must be reimbursed through emergency special assessments. Because the nation’s largest banks would end up paying a disproportionate cost of such a bailout, they would be expected to lobby against the exten sion of insurance to uninsured deposits, unless the criteria for assistance were truly met. 8The protection afforded to bank clearing houses is consi dered in more detail in the conclusion to this article. 7Under 12 U.S.C. § 1823 (c) (4) (G) of FDICIA, for insurance to be extended to uninsured liabilities of a bank, beginning MAY/JUNE 1994 36 operation of the banking system. But even in such an environm ent, problem s that arise out side the banking system may motivate central bank lending through the discount window. In particular, securities m arkets may be vulnerable to externalities arising from asym m etric in for mation. I will argue that these problem s may be addressed effectively by channeling funds through banks that borrow from the window, rath er than through direct lending from the cen tral bank to nonbank firm s.9 The example that I will focus on is the com m ercial paper m arket "ru n ” that followed Penn Central’s 1970 bankruptcy. As many research ers have stressed, the bank ing system is particularly vulnerable to confu sion about the incidence of disturbances for tw o reasons. First, its assets (that is, bank loans) typically are not traded in centralized m arkets. Thus, it is difficult for an uninform ed bank depositor to keep abreast of the effect of a given news item on the value of his bank’s as sets. Second, the fact that banks finance through large quantities of demandable debt allows n er vous depositors to withdraw from the bank rath er than wait to see w h eth er their bank will survive or fail. Although these two attributes that make banking panics possible—nontraded assets and demandable debt—seem to set the banking sys tem apart from oth er m arkets, the banking sys tem is just an extrem e case of a m uch m ore general phenom enon. The condition necessary to generate a costly panic in a d e b t m a r k e t is that the tim e horizon for rolling over debt is less than the time it takes to m ake accurate reappraisals of firm -specific risk during episodes of general bad news. Lenders’ lack of inform a tion about the attributes of specific firm s may result in the pooling of borrow ers w ith com mon observable characteristics. In such circum stances, firm s will face tem porarily high “lemons prem ia” in debt and equity m arkets, w hich will increase the cost of finance and reduce investm ent, even for firm s whose true “fundam entals” are u naffected by the bad news. 9Mishkin (1991a) also argues that asymmetric information is relevant outside the banking sector. He uses data on interest rate spreads between risky and riskless debt instruments to support this view. He finds evidence of an increase in these spreads (which he interprets as reflecting an increased inability to sort borrowers according to risk) coinciding with or prior to the Penn Central crisis of 1970 and the stock market crash of 1987. FEDERAL RESERVE BANK OF ST. LOUIS Firms with short-term debts (which must be rolled over regularly) can be particularly vulner able to systemic risk and the possibility of a run. A liquidity crisis that would prompt a general calling in of debt by creditors could lead firm s with outstanding short-term debt to experience high costs of debt rollover or asset sale not experienced by oth er firms. Furtherm ore, if interm ediaries for particular m arkets (for example, com m ercial paper deal ers) su ffer losses from one firm ’s issues, they may be less able to deal in the paper of other firm s. This, too, can force firm s to pay higher costs for funds tem porarily in the affected m ar ket, or switch to new, higher-cost sources of funds. Firms that face liquidity problem s in nonbank debt m arkets may have difficulty borrow ing from bankers, too, particularly if they lack exist ing bank-lending relationships. To the extent that banks have special inform ation about b o r row ers’ attributes, due to their past involvement with firm s and their ongoing m onitoring of firm com-pliance with lending covenants, banks may be able to assist firm s w hen their costs of funds rise in other credit m arkets. For firm s that moved away from reliance on bank credit, how ever, th ere may be no strong banking relation ship to fall back upon. Assistance from banks for these firm s would be forthcom ing only at higher interest rates, which would com pensate banks for the transaction and inform ation costs of drafting em ergency lending arrangem ents. In particular, if the bank expects only a tem porary relationship with the firm in need (for the dura tion of the “em ergency”), the bank will have to charge higher in terest rates to recoup its fixed costs over a sh orter lending period. Given the high cost of substituting bank credit for oth er credit on short notice, a credit m arket run may force some solvent firm s into financial distress, or simply reduce their ability to invest or to lend to other firm s.10 If the social costs of such disruptions to short-term debt m arkets are large, Fed intervention to defuse such crises may be w arranted. Specifically, the Fed could 10Calomiris, Himmelberg and Wachtel (forthcoming) find that nonfinancial commercial paper issuers of the 1980s tended to be net lenders to other firms through accounts receivable. 37 supply banks with funds at low cost through the discount window for the express purpose of refinancing maturing short-term debts of firm s suffering from disruption in the short-term debt m arket. In a competitive banking system, this subsidy would be passed on to borrow ers and would mitigate high short-run costs of switching to banks for credit. New financial m arkets may be particularly vulnerable to negative externalities among firms or tem porary disruptions to m arket dealers. The lack of data on the risks and liquidity of new products, and relatively thin trading, in creases the likelihood of systemic risk in new m arkets. In the following section, I consider w hether the com m ercial paper m arket experienced such a financial crisis in mid-19 70, and w h eth er that crisis w arranted discount window intervention. The com m ercial paper m arket o f m id-1970 is an especially interesting case to exam ine for six reasons. First, most com m ercial paper m atured quickly—with an average m aturity of under 30 days (Stigum, 1983, p. 632). This meant that a sudden disinclination by investors to hold com m ercial paper would entail substantial problems for firm s trying to roll over their com m ercial paper debt. Second, com m ercial paper was a new and growing method of finance during the 1960s.11 Institutional arrangem ents for rating and sup porting com m ercial paper issues w ere virtually nonexistent; thus, inform ation im perfections w ere potentially im portant. Third, com m ercial paper finance originated as a substitute for bank credit. Many firm s that had moved to this m arket in the 1960s may have curtailed or term inated their relationships with com m ercial banks (making the disruption in the supply of paper m ore costly). Fourth, during the early years of rapid grow th in this m arket, there was a m ajor shock to the com m ercial paper m arket, namely the failure of Penn Central in 1970, which w as associated with substantial contraction of outstanding paper (that is, a "ru n ”). Fifth, com m ercial paper issuers include many of the econom y’s largest firm s, and other firm s often depend upon them for credit (Calomiris, Himmelberg and W achtel, forthcoming). In creases in the cost of funds for this class of b orrow ers thus may have significant secondorder effects on the cost of credit fo r other firms. Finally, the Fed intervened during this crisis largely by encouraging banks to com e to the discount window to finance the payoff of com m ercial paper. Evidence from the Penn Central com m ercial paper crisis of 1970 allows a detailed case study of “inform ation externali ties,” the potential for a run in m arkets for traded short-term debt, and an evaluation of Fed intervention in response to such a crisis. Penn Central's Failure and the Liquidity Crisis o f Mid-1970 The facts surrounding the com m ercial paper run following the Penn Central failure are com monly known (see Schadrack and Breim yer, 1970; Maisel, 1973; Tim len, 1977; Brim m er, 1989; and Mishkin, 1991a), but some important details are w orth reviewing. Along with many other firm s, Penn Central's financial condition deteriorated during the recession of 1 9 6 9 -7 0 . Penn Central was a m ajor issuer of com m ercial paper, w ith m ore than $84 million outstanding, m uch of w hich cam e due in June, July and August of 1970. As Penn Central’s cash flow declined, its debt holders and their agents ap pealed to the federal governm ent for financial assistance, w hich the Nixon Administration sup ported. The Administration proposed a $200 million loan guarantee to a syndicate of some 70 banks, w hich w ere to provide a two-year loan in that amount. The loan guarantee would be autho rized through a loose interpretation of the "T h e re had been an earlier incarnation of the commercial paper market that thrived from the 1870s and declined in importance during the 1920s. Calomiris (1992a) argues that this operated effectively as an interbank loan-sale market, moving high-quality borrowers from high credit-cost areas to low credit-cost areas. Consistent with that argument, James (1994) views the demise of this market as the result of the bank merger wave of the 1920s, which provided an alternative means to channel credit through the financial system. MAY/JUNE 1994 38 Defense Production Act. Although th ere was in creasing congressional opposition to this plan, as late as Friday, Ju n e 19, the Wall Street J o u r nal reported that "the opposition doesn't yet ap pear strong enough to halt the $200 million loan guarantee.” That article also reported the possi ble existence of a secret m em orandum from the Federal Reserve Bank o f New York, recom mending "that the loan be granted, based on an investigation that bank is believed to have con ducted into the credit-w orthiness of Penn Cen tral.” Contrary to the Wall Street Jo u rn a l report, no such m em orandum existed, and that same Friday the Penn Central plan was rejected by Congress. T he Nixon Administration then asked the Federal Reserve Board (through the New York Fed) to make a loan to Penn Central to help it m eet immediate obligations. T he New York Fed recom m ended against the loan, and it was denied. This news forced Penn Central’s bankruptcy on Sunday, Ju n e 21. The surprising news of the unwillingness of Congress and the Fed to prop up Penn Central created widespread concern over the weekend that the Penn Central failure would have rep er cussions elsew here in the econom y, particularly for other firm s that had large outstanding com mercial paper issues. It is not easy to explain this con cern w ithout invoking an "inform ation externality” of some form . That is, one needs to explain why the bad news about Penn Central would raise doubts about oth er firms. The bad new s about Penn Central on Ju n e 19 had two parts. First, prior to that date, the Wall Street Jo u rn a l reported that the New York Fed had made a favorable audit of Penn C entral’s underlying financial strength. A fter Friday, quite the opposite was known. The reaction of the m arket, as reported in the press, was that if Penn Central's financial state could so rapidly and unexpectedly have turned sour in the previ ous year, w hat other "blue chip” com m ercial paper issuers might be in the same position? This con cern was fueled by the fact that the incom e reductions during the recession of 1 9 6 9 -7 0 , which potentially affected many firms, w ere not know n at the firm level with any p re cision at the time. Those concerns about other firm s began to be voiced even b efore the revela tion of the New York Fed’s audit. For example, a lead article in the Jo u rn a l on June 12 queried: “How many other U.S. corporations are so short 12See Schadrack and Breimyer (1970, p. 283). FEDERAL RESERVE BANK OF ST. LOUIS of cash that they may soon find themselves similarly unable to pay their bills?” Until the m arketplace could assess the extent to which Penn Central’s financial position was the result of idiosyncratic shocks and mismanagement, as opposed to a signal of a com m on problem likely to be faced by many firm s, Penn C entral’s failure would cast doubt on the financial posi tion of other firm s. The second elem ent of general bad news revolved around the fate of Penn Central and its creditors. It becam e clear that, w hatever its underlying condition, the governm ent would not guarantee Penn Central’s debt and that, th ere fore, Penn Central’s creditors faced the possibili ty of substantial losses. The incidence of losses on the firm 's com m ercial paper was unknow n, but it was rum ored that ow nership was quite concentrated. For example, on June 15 the J o u r nal reported that Morgan Guaranty owned or acted as “agent” for nearly $84 million in Penn Central’s com m ercial paper. According to Federal Reserve data on holdings of com m ercial paper, in early 1970 nonfinancial corporations owned about 74 percent of outstanding paper.12 The Ju n e 12 Jo u rn a l article cited above also asked: "If even one m ajor corporation should becom e insolvent, would its failure bring down other cash-short com panies because the failing company couldn’t pay its bills? Could that, in turn, intensify the present severe strain on the cash resources of banks and corporations into a liquidity crisis, draining the flow of money and credit and plunging the nation into a depres sion?” W hile this “domino” scenario o f economywide depression may seem a bit farfetched, it would have been less farfetch ed to imagine that one or two m ajor com m ercial paper issuers (who may have been creditors of Penn Central) might also find it hard to repay their debts. Thus, lack of inform ation about the effects of the recession on oth er firm s (which Penn Cen tral’s failure indicated might be large), and about the identities of Penn Central’s creditors and their creditors in turn, could have produced legitimate, rational con cern about rolling over the com m ercial paper of oth er firm s at p re existing term s. The com m ercial paper m arket was especially vulnerable to these sorts of doubts because it was a fast-growing new financial m ar ket, as shown in Figure 1. From 1956 to 1966, 39 Figure 1 Commercial Paper Outstanding Billions of dollars Source: Schadrack and Breimyer (1970), Chart I. the am ount of nonbank com m ercial paper is sued rose at a 16 percent compounded annual rate. From 1966 to 1970, it rose 29 percent per year. The num ber of com panies issuing paper rose from 335 in 1965 to 575 in April 1970. In the later period, grow th was especially con cen trated in dealer-placed paper (which includes all nonfinancial com m ercial paper), which grew from 1966 to 1970 at an annual rate of 57 per cent. Rising interest rates and regulatory restric tions on banks (especially Regulation Q ceilings) are widely cited as the cause of this boom in the com m ercial paper market. The m arket pricing and rating of paper issues on a large scale was in its infancy (Stigum, 1983, p. 635; Standard and Poor's, 1979, p. 1), and the recession of 1 9 6 9 -7 0 was the first dow nturn to test the burgeoning com m ercial paper m arket. Fu rtherm ore, com m ercial bank lending or standby com m itm ents for com m ercial paper issues did not exist at this time; thus, com m ercial paper holders faced g reater risk than they do today.13 It would not be farfetched to argue that learning was occurring “in real tim e” and that the first time a recession oc curred, and a com m ercial paper issuer failed, the m arket might have found it difficult to as sess the ram ifications for others with any great confidence. Indeed, it may have been necessary for the m arket to reevaluate its methods for pricing paper generally in light of this surpris ing event. Professor Roger M urray o f Columbia University argued that com m ercial paper mar- 13The nature of these arrangements for supporting commer cial paper issues is discussed below, as well as in Calomiris (1989b). MAY/JUNE 1994 40 ket pricing had been too optimistic in the 1960s. His (post-crisis) study of Penn Central’s financial position in the 1960s concluded that there was m uch to be learned from the Penn Central col lapse about the need for g reater caution in valu ing com m ercial paper: "A carefu l financial analyst might well have recom m ended...against the purchase of Penn Central com m ercial paper a year or m ore b efore the events of May and Ju n e 1970.”14 M urray accounted for the poor ex ante evaluation of risk by the fact that so "m any new faces appeared in that m arket for large sums at the time and Penn Central was hardly noticed as an unusual case.” Schadrack and Breim yer (1970) provide a simi lar perspective. They claim that before the Penn Central failure, "the confusion of corporate size with liquidity tended to mask some deteriora tion during [the late 1960s] of the quality of com m ercial paper outstanding...the fact that a num ber of firm s in the m arket by 1970 had very high debt-to-equity ratios and/or income flows of dubious quality (some conglom erate, franchising and equipm ent leasing companies, fo r example) suggests such a deterioration in the quality of outstanding paper.”15 They also argue that, in addition to the con cern about other com m ercial paper b orrow ers brought on by the failure of Penn Central, the bank’s failure raised con cern about some of the m ajor b ro k er age houses, w hich acted as dealers and pur chasers in the m arket. Commercial paper dealers maintain open positions in the paper they sell eith er as part of an underw riting arrangem ent, or through a com m itm ent to m aintain a secon dary m arket in the paper (Stigum, 1983, p. 640). The th reat of a liquidity crisis for firm s and their dealers led to a collapse of demand for the debt instrum ents o f others. These fears fueled the flight to cash. Schadrack and Breim yer (1970) also argue that the crisis led to refined methods of pricing com m ercial paper, which is consistent w ith M urray’s view that th ere was room for im provem ent. In particular, after the Penn Central crisis they found a w ider disper sion of rates for dealer-placed paper, w hich 14See Murray (1971). Whitford (1993) applied Altman’s (1968) “ z-score” model to Penn Central’s accounts as of Decem ber 1969, and found a remarkably low z-score of 0.135. Alt man had found that no healthy firms had z-scores of below 1.81 and no bankrupt firms had a score above 2.99. 15See Schadrack and Breimyer (1970, p. 289). Digitized for FEDERAL RESERVE BANK OF ST. LOUIS FRASER they interpreted as the result o f “greater inves tor selectivity.” Also, they noted a persistent shift tow ard bank CDs and Treasury bills. As Mishkin (1991a) and Schadrack and Brei m yer (1970) point out, the spread betw een com m ercial paper and Treasury bills widened during and after the crisis. This widening seems to reflect a persistent revision in the evaluation of com m ercial paper risk. Schadrack and Breim yer (1970) report that in November 1970 the dealer paper rate averaged 103 basis points above the Treasury bill rate, com pared to previous spreads of roughly half that amount. A similar pattern is visible in Table 1, w hich reports the federal funds rate, three-m onth T reasury bill yields, the discount rate, and the four-to-six-month prime com m ercial paper rate before, during and after the crisis. The "flight to quality,” visible in the declining yields of T reasury bills and rising short-term spreads, is also visible in long-term yields and spreads, shown in Table 2. From Ju n e 20 to Ju n e 27, Treasury bond yields fell as corporate bond yields rose. The spread betw een the Treasury bonds and the Aaa corporates reached a peak on July 11. Interestingly, the spread b e tw een Aaa- and Baa-rated bonds was essentially constant during the crisis, but rose afterw ards. This is consistent with the view that during the crisis, increased riskiness was attributed to all securities, but that, after the crisis, investors w ere b etter able to sort firm s into risk categories. Concerns about the financial condition of com m ercial paper issuers and dealers proved unw arranted ex post (since no other com m er cial paper issuers defaulted), but seem to have been im portant ex ante, as evidenced by move ments in the stock m arket and com m ercial paper m arket. Firms, especially those with large outstanding debt, saw large stock price declines in the first three days of the crisis. During that time, the Dow Jones Industrial Average lost 28 points (a fall of roughly 4 percent). Chrysler, General M otors and IBM all saw large losses as rum ors circulated that they faced risks of being unable to m eet their debts (W all Street Jou rn al, 41 Table 1 Selected Yields and Interest Rates Date 1970 January February March April May June 1 2 3 4 July 1 2 3 4 August September October November December 3-month Treasury bill yield Federal funds rate 7.89% 6.88 6.16 6.59 7.00 6.82 6.76 6.71 6.51 6.46 6.62 6.46 6.34 6.25 5.80 5.84 4.99 4.83 9.04% 8.41 7.45 8.43 7.64 7.84 7.98 7.80 7.21 7.23 7.34 7.59 7.16 6.34 6.05 6.11 5.16 4.82 Discount rate 6.00% 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 6.00 5.85 5.52 4-6 month prime commercial paper 8.55% 8.50 8.03 8.00 8.13 8.13 8.15 8.25 8.25 8.38 8.35 8.25 8.35 7.70 7.20 6.63 5.75 5.75 NOTES: Data are all end-of-month, except for June and July, which are reported end-of-week. Trea sury bill and commercial paper yields are quoted on June 6, 13, 20 and 27 and July 4, 11, 18 and 25. Federal funds rates are for June 3, 10, 17 and 24, and July 1, 8, 15 and 22. SOURCES: Board of Governors of the Federal Reserve System (1976), Table 12.5B, Table 12.6B, and Table 12.7B; Federal Reserve Bank of St. Louis. Ju n e 23-25, 1970, "A breast of the M arket”). Busi n ess W eek quoted one stock m arket analyst as saying that "investors think that any company... w ith...debt is going bankrupt” (June 27, p. 42). Perhaps the best indicator of the extent of these fears is the contraction in the volume of com m ercial paper outstanding from late Ju n e to mid-July. Total outstanding nonbank com m ercial paper fell from $32 billion on Ju n e 24 to $29 billion on July 15, with $2.3 billion of that decline in the first week of the crisis (see Figure 2). Interestingly, com m ercial paper rates showed little change during the crisis, although the spread betw een paper rates and other money m arket rates did widen. T he reason for this was the speedy reaction o f the Federal Reserve to the failure of Penn Central. Luckily, it occurred over a weekend, which gave the Fed time to prepare for the opening of financial m arkets on Monday. The Fed pursued four courses of action. The Fed's Discount W indow Policy During the Crisis First, the Fed contacted m em ber banks and notified them that “as they made loans to enable their custom ers to pay off m aturing com m ercial paper and thus needed m ore reserves, the Fed eral Reserve discount window would be availa b le."16 The meaning of "available” is of paramount im portance. The Federal Reserve let m em ber banks know that if they borrow ed at the dis count window for purposes of making loans to com m ercial paper issuers, they would be able to do so w ithout incurring any costs other than the discount rate. The Fed was inform ed by banks when their discount borrow ing resulted from financing com m ercial paper rollovers, and the total amount o f such discount borrow ing to taled some $500 million in the weeks im mediate ly following Penn Central (Melton, 1985, p. 158). Beyond the amount lent through the discount window, access to the window for com m ercial 16See Treiber (1970, p. 16). MAY/JUNE 1994 42 Table 2 Long-Term Yields and Spreads Date1 Long-term government bonds2 1970 January February March April May June 6 13 20 27 July 4 11 18 25 August September October November December Aaa corporate bonds3 6.84% 6.25 6.33 6.70 7.21 7.00 7.09 7.05 6.89 6.73 6.56 6.61 6.54 6.73 6.52 6.65 5.97 6.05 7.91% 7.83 7.92 7.83 8.21 8.30 8.42 8.55 8.60 8.60 8.55 8.49 8.40 8.13 8.06 8.07 8.02 7.51 Baa corporate bonds3 Spread between Aaa and government bonds Spread between Baa and Aaa corporate bonds 1.07 1.58 1.59 1.13 1.00 1.30 1.33 1.50 1.71 1.87 1.99 1.88 1.86 1.40 1.54 1.42 2.05 1.46 0.90 0.90 0.74 0.91 0.89 0.83 0.76 0.71 0.76 0.81 0.89 0.90 0.98 1.34 1.26 1.27 1.35 1.51 8.81% 8.73 8.66 8.74 9.10 9.13 9.18 9.26 9.36 9.41 9.44 9.39 9.38 9.47 9.32 9.34 9.37 9.02 1All data are end-of-month, unless otherwise indicated. 2Maturity varies. 3Rated by Moody’s. SOURCE: Board of Governors of the Federal Reserve System (1976), Table 12.12B. paper rollovers gave assurance to the financial markets that the liquidity essential to their operation would be preserved. If panicky investors refused to renew their holdings of commercial paper, preferring Treasury bills, bank deposits— anything!—instead, their extreme preference for safety would not be allowed to contribute to widespread insolvency. Once every one understood that, there was little reason for panic (Melton, 1985, p. 158). Fed encouragem ent to use the discount w in dow to finance the payoff of com m ercial paper was associated with reduced costs of borrow ing from the Fed, even though the discount rate re mained unchanged. Normally, the costs of b o r rowing from the discount window include the discount rate and a nonpecuniary “hassle” cost. That is, the Fed does not want to encourage abuse of the privilege of borrow ing from the discount window and banks that may be seen as abusing the privilege run the risk of exam i nation and regulatory sanctions. This penalty explains the positive d ifference betw een the fed Digitized for FEDERAL RESERVE BANK OF ST. LOUIS FRASER funds rate and the discount rate. If th ere w ere no penalty, banks would be indifferent betw een borrow ing from oth er banks and the Fed’s dis count window. In this case, the two rates would be identical. In the presence of a nonpecuniary cost of borrow ing from the Fed, as long as b o r rowings are positive, the fed funds rate will be higher than the discount rate since, on the m ar gin, banks will be indifferent betw een paying the fed funds rate in the interbank m arket and borrow ing from the Fed (which entails a dis count rate cost and a hassle cost). Figure 3 provides a simple illustration of the simultaneous determ ination of the federal funds rate and borrow ed reserves, w hich is helpful in analyzing the effect of discount window lending during the Penn Central crisis. Reserve demand is shown as a negative function of the federal funds rate. The position of the demand sched ule varies with loan demand, reserve requ ire ments, and the demand for excess reserves. The Fed determ ines the am ount of nonborrow ed 43 Figure 2 Commercial Paper and Business Loans June-August 1970 Billions of dollars Billions of dollars June July * Including business loans sold to affiliates. Source: Schadrack and Breimyer (1970), Chart V. reserves through its open m arket operations. Borrow ed reserve costs are given by an upward sloping schedule, w hich sums a constant pecuni ary cost (the discount rate) with an increasing nonpecuniary hassle cost. The m ore reserves that are borrow ed, the more the Fed is liable to penalize borrow ing. Figure 3 illustrates equilibri um in the reserve m arket for Ju n e 17 and July 15, 1970, using actual data on the discount rate (which rem ained at 6 percent throughout the crisis), nonborrow ed reserves, borrow ed reserves and the federal funds rate. Assuming equilibri um in the reserve m arket, we can identify shifts betw een these two days in reserve demand (as bank loans rose to com pensate for the con trac tion in com m ercial paper) and in reserve sup ply. The reserve supply function shifted in slightly (nonborrow ed reserves fell due to in creased cu rren cy demand, w hich was only partly offset by open m arket operations) and rotated dow nward as the Fed reduced its non pecuniary penalty fo r borrow ing. August The downward rotation of the borrow ed reserve supply function illustrates how the Fed eral Reserve lowered the nonpecuniary cost of borrow ing from the discount window during the crisis. O ther evidence on the composition of bank lending, bank borrow ings from the Fed, and the different rates charged to different types of bank custom ers suggests that the reductions in nonpecuniary costs w ere linked (as the quotation above suggests it was) to in direct subsidies for com m ercial paper rollovers. That is, it seems that loans to m em ber banks for this purpose w ere granted a special "subsi dy” by the Fed (in the form of lower, or possi bly zero, nonpecuniary costs). Consistent with this account, the composition of m em ber bank borrow ings changed during the crisis. As of Ju n e 24, large com m ercial banks (primarily m oney-center banks) accounted for only 75 percent of borrow ing from the Fed. The trebling of m em ber bank borrow ing from MAY/JUNE 1994 44 Figure 3 Shifts in the Reserve Market June 17-July 15 * Discount rate Ju n e 24 to July 15 was due to an increase in m oney-center borrow ing, as one would expect if it was earm arked for com m ercial paper payoff. As shown in Table 3, total borrow ed reserves rose by $1,196 billion, while borrow ed reserves of large com m ercial banks rose $1,224 billion. These same banks w ere the only ones that saw a large grow th in loans to businesses and finance com panies during the crisis. Loans in creased by $2.3 billion from Ju n e 24 to July 15, almost an exact offset of the amount by which com m ercial paper was reduced during this peri od. This rise of 2.6 percent in total loans for this group of banks was highly unusual. The average rate of increase for the preceding four years during this period of the y ear had been 0.03 percent, and the highest rate of grow th in the preceding four years had been 0.25 percent in 1968. Finally, there is w eak evidence that large b o r row ers from m oney-center banks as of August 1970 (which would have included form er com m ercial paper issuers) received loans on rela FEDERAL RESERVE BANK OF ST. LOUIS tively favorable term s. Available data on average loan interest rates fo r the first two weeks of May and August 1970 by size of borrow er and region show that large, short-term b orrow ers in N ortheastern financial cen ters experienced the smallest increase in lending rates over this peri od (although differences are small). As Table 4 shows, the largest classes of borrow ers in New York City actually saw slight reductions in aver age loan interest rates. Other Fed Reactions to the Crisis The discount window announcem ent ta r geting assistance to com m ercial paper issuers was only the first of the Fed's fou r policy responses to the crisis. On Tuesday, Ju n e 23, the Fed suspended regulation Q ceilings on large-denomination bank CDs. This allowed a flood of m oney into the com m ercial banks, so that maturing com m ercial paper could be directly recycled through CDs, which financed bank loans to form er issuers. As shown in Table 3, from June 24 to July 15, large negotia 45 Table 3 Banking System Changes During the Penn Central Crisis Date1 Federal funds rate minus discount rate 1970 January February March April May June 3 10 17 24 July 1 8 15 22 August September October November December 3.04 2.41 1.45 2.43 1.64 1.84 1.98 1.80 1.21 1.23 1.34 1.59 1.16 0.34 0.05 0.11 -0 .6 9 -0 .7 0 Loans to business Large negotiable Borrowed reserves and finance companies CDs at large Total of large commercial by large commercial borrowed banks commercial banks banks2 reserves $1,071 873 1,594 926 979 1,335 834 459 840 923 1,598 2,036 1,216 1,044 852 418 1,144 252 $ 807 522 1,334 680 675 1,063 624 273 613 671 1,402 1,837 1,044 941 788 341 1,098 224 $83,423 83,549 83,903 84,122 83,265 83,545 83,811 85,785 85,331 87,212 87,161 87,590 87,472 86,067 88,426 86,514 86,385 89,130 U.S. government securities held by Federal Reserve Banks $10,444 10,839 11,795 13,022 12,984 12,964 12,956 12,741 12,949 14,118 15,199 15,980 16,911 20,157 22,227 23,546 25,201 26,075 $55,568 55,749 55,621 56,085 57,115 57,698 57,552 57,823 57,005 57,714 57,671 58,839 58,138 59,618 60,055 59,283 61,209 60,632 1 data are end-of-month, unless otherwise shown. Dollar amounts are in millions. All 2These are the sum of commercial and industrial loans by large commercial banks, and loans to personal and sales finance companies, etc. SOURCES: Table 1 and Board of Governors of the Federal Reserve System (1976), Table 4.1B, Table 10.1D. ble CDs at large com m ercial banks increased from $13 billion to $16 billion, and the growth continued, with CDs of large banks in excess of $26 billion by the year's end .17 The third policy intervention by the Fed was open m arket operations. From Ju n e 17 to July 15, total U.S. governm ent securities held by the Fed increased from $57.8 billion to $58.8 billion. As noted above, how ever, open m arket opera tions w ere not sufficient to m aintain the stock of nonborrow ed reserves, given the increased demand for cu rren cy by the public. Thus, b o r rowed reserves w ere relied upon as the primary vehicle fo r expanding reserves during the crisis. loans, directly or indirectly, to "w orthy” b o r row ers who w ere otherw ise unable to secure credit. The Fed never made such loans because its other policies proved sufficient to contain the run on com m ercial paper, but it is clear that the Fed was willing to provide direct lend ing if banks had been unwilling to m ake ap propriate loans for com m ercial paper rollovers. In his statem ent to Congress on July 23, the Chairman of the Board of Governors, Arthur Burns, made this com m itm ent clear. He viewed the discount window as the key to preventing a liquidity crisis, and saw direct lending by the Fed to firm s in need, if necessary, as an ap propriate fail-safe m easure: The Fed was also prepared to use “standby procedures” so that, if necessary, it could make Credit demands on the banking system at large can be accommodated by open market operations, 17An unintended cost of Regulation Q was that it removed an “ automatic stabilizer” from the financial system by making it less attractive for investors to hold bank debt at times of crisis in other markets. MAY/JUNE 1994 46 Table 4 Average Loan Rates on Short-Term Loans_________________ New York City Loan amount All sizes 1,000- 9,000 10,000- 99,000 100,000-499,000 500,000-999,000 1 million and over $ May August 8.24% 9.05 8.91 8.53 8.31 8.13 8.24% 9.07 8.95 8.59 8.23 8.12 Other Northeastern financial centers May 8.86% 9.23 9.34 9.01 8.72 8.45 August 8.89% 9.41 9.42 9.01 8.68 8.49 SOURCE: Board of Governors of the Federal Reserve System (1976), Table 12.9A. while the needs of individual banks can be met through the discount window...We have found, also, that minor adaptations of conventional mone tary tools can provide solutions to special financial problems...it was made clear that the discount window would be made available to assist banks in meeting the needs of businesses unable to roll over maturing commercial paper, and member bank borrowings for this purpose subsequently have risen...These conventional tools are but tressed with standby procedures to permit the Federal Reserve to make funds available to credit worthy borrow ers facing unusual liquidity needs through 'conduit loans’—that is, loans to a mem b er bank to provide funds needed for lending to a qualified borrower...Furtherm ore, the Federal Reserve could—under unusual and exigent cir cumstances—utilize the limited power granted by the Federal Reserve Act to make direct loans to business firms on the security of Government obli gations or other eligible paper, provided the bor row er is creditworthy but unable to secure credit from other sources.18 Here, Burns explicitly allows for Fed loans backed by com m ercial paper or other eligible collateral. In dealing with the Penn Central crisis, the Fed did not simply focus on controlling the money supply or an interest rate, which it could have done easily through open m arket operations. Rather the Fed coaxed deposits into banks by relaxing Regulation Q ceilings, and 18See Burns (1970, pp. 624-5). FEDERAL RESERVE BANK OF ST. LOUIS used the discount window to encourage banks to make loans to custom ers experiencing distress —especially com m ercial paper issuers. T he logic of the Fed’s com bined approach was that m one tary aggregates, bank credit and assistance to the com m ercial paper m arket could b e targeted independently by using th ree instrum ents. Relaxation of Regulation Q, rath er than expan sionary open m arket operations, allowed bank credit grow th w ithout (narrow) m oney grow th. The discount window was directed tow ard the special difficulties in the com m ercial paper m ar ket. The Fed left open the possibility of lending directly to firm s in need if they w ere turned down by bankers. Evaluating Discount W indow Policy During the Crisis It is not self-evident that the Fed’s policy response was correct. Schw artz (1992) has a r gued that the Penn Central crisis was not a "real” financial crisis and that discount lending served no useful purpose. Of course, the ab sence of a financial collapse in m id-1970 may have been attributable to Fed intervention itself, a possibility Schw artz does not take into ac count. But even if Schw artz is too quick to dismiss the potential seriousness of the Penn Central crisis—particularly given the evidence on yield-spread m ovements and contraction of the volume of com m ercial paper—that does not prove that the discount window was a n eces sary instrum ent for dealing with the crisis. If the failure of Penn Central increased doubts about the solvency of all firm s in the economy, then a tem porary expansion of open m arket operations or a Regulation Q relaxation—to in crease the supply of credit available to all b o r row ers through relatively inform ed financial interm ediaries—would have been a desirable response to an economy-wide need for liquidity, and there would have been no need to use the discount window. On the other hand, if the crisis involved a special reappraisal of the creditw orthiness of com m ercial paper issuers and com m ercial paper dealers in particular, and a reassessm ent of the desirability of lending through the com m ercial paper m arket, then increasing the supply of loanable funds from banks may not have been as effective as targeting tem porary assistance (a short-run subsidy fo r bank loans to com m ercial 47 paper issuers) using the discount window as a means to smooth issuers’ costs of rollover.19 In this case, open m arket operations or Regulation Q relaxation would have been a blunt instru m ent for dealing with a run on com m ercial paper p e r se, while discount window subsidies for the payoff of com m ercial paper would have provided targeted assistance without a ffect ing m onetary aggregates or interest rates on all traded assets. If some com bination of an economy-wide reassessm ent of firm s and a com m ercial paper run characterized the crisis, then policy could have com bined an aggregate in crease in open m arket operations or Regulation Q relaxation with targeted assistance to com m ercial paper issuers. Thus, to assess the desirability of the use of the discount window during the crisis, one must exam ine the incidence of the crisis across firm s. Was it purely an economy-wide phenom e non or did it pose a special threat to com m er cial paper issuers? An Event Study o f the Penn Central Crisis To investigate the extent to w hich the Penn Central crisis posed a special threat to com m er cial paper issuers, I exam ine data on firm s’ abnorm al stock returns during the crisis. Did firm s with outstanding com m ercial paper suffer abnorm al negative retu rns relative to other firm s during the onset of the crisis, and w ere those negative retu rns reversed by Fed in ter vention? To answ er this question, I com bine CRSP data on daily stock retu rns with Compustat data on annual incom e and balance sheet variables for nonfinancial corporations to m ea 19The moral hazard costs of government pass-throughs were minimal, since the banks, not the government, bore the default risk on the loans. This statement requires some qualification. If the pool of borrowers faced large aggregate default risk, then bank failures might have resulted from the loans, in which case the government would have borne some of the losses. Moreover, if some banks had been on the brink of failure, they might have been willing to make subsidized loans to the riskiest firms, thus concentrating overall default risk and making the government’s indirect default risk greater. The central assumptions underlying my claim that the government’s share of the risk was small are that banks were not on the verge of failure at the time, and that the average quality of the commercial paper borrowers pool was high. The relaxation of Regulation Q ceilings on CDs was also helpful in limiting the government’s risk, since it limited the amount of borrowing from the Fed. CDs also provided a natural vehicle for financing fixed-term commercial paper, and did so without affecting the money supply. sure cross-sectional differences in abnorm al retu rns over various dates, and to link them to firm financial characteristics m easured at the beginning of 1970. I employ standard m easures of abnorm al returns, using residuals from fore casts of m arket retu rns based on estim ates of firm s’ betas (from a 100-day pre-sample period) and the aggregate contem poraneous movements in the m arket. Specifically, consider a standard model of firm s’ stock returns, which decomposes retu rns into system atic and idiosyncratic factors: (1) RIt = a + b I R, + e I,,' t t w here R m easures returns, z indexes firm s, t denotes the date, and a and b are param eters to be estimated. The erro r term e m easures ab n o r mal retu rn s—the firm -specific, idiosyncratic daily retu rn at each date—or, in other words, the part of the stock retu rn that is not forecastable using the simultaneous aggregate return for the m arket and the firm s' estimated correlations with the m arket (b ). Each firm ’s b is estim ated using observations on daily stock retu rns fo r 100 trading days prior to the event (in this case, June 12). Cumulative abnorm al retu rns over any “win dow” are the accum ulation of abnorm al retu rns for each of the dates included in the window. Cumulative retu rns generated from the above forecasting equation are “standardized” such that they can be interpreted to have been drawn from a unit normal distribution.20 This adjustm ent results in a cross-section of stan dardized cumulative abnorm al retu rns (SCARs) for each firm in the sample over the event w in d o w . 20For details, see Wall and Peterson (1990). MAY/JUNE 1994 48 The event windows are defined as Ju n e 12Ju n e 22 and Ju n e 23-July 9. Early concerns about com m ercial paper issuers reported in the Wall Street Jo u rn a l date from Ju n e 12. Ju n e 22 is the date after w hich Fed intervention should have improved the position of com m ercial paper issuers. By the second w eek of July, the con traction in outstanding com m ercial paper began to be reversed. T he goal of the event study is to examine w hether (likely) com m ercial paper issuers suffered abnorm al negative stock retu rns dur ing the Penn Central crisis, and w hether Fed in tervention reversed those costs to com m ercial paper issuers, after controlling for other m eas ures of cross-sectional d ifferences among firms. To control for other influences that would not have been specific to the com m ercial paper m arket, I add a variety of balance sheet and in com e statem ent variables taken from the Jan u ary financial reports of these nonfinancial firms. All firm balance sheet and incom e data are m easured as of the beginning of 1970.21 The control variables included are: the ratio of debt to assets; the ratio of short-term debt to assets; the size of the firm (m arket value of capital); the ratio of net incom e to m arket value of capi tal; the ratio of inventories to sales; and the squares of each of these variables. These varia bles are included to control for the possibility that the share prices of firm s w ith high ex posure to m acroeconom ic shocks (firms with high leverage, or with large financing needs relative to sales) may have responded m ore strongly to econom ic news, irrespective of w hether or not they w ere com m ercial paper issuers. For example, if Penn Central’s failure increased the cost of debt for all firms, then leverage ratios or inventory-to-sales ratios would identify cross-sectional d ifferences in SCARs. Isolating the effect on SCARs of reliance on the com m ercial paper m arket is not straightfor ward, since data on outstanding com m ercial paper issues of firm s are not available fo r this period. The regular reporting of com m ercial paper ratings was largely a consequence of the Penn Central crisis. Standard and Poor's began publishing some com m ercial paper ratings in The B on d O utlook in July 1970, but these rat 21This was dictated by the superior data available on the annual Compustat tape. Quarterly Compustat data for this period are often incomplete. FEDERAL RESERVE BANK OF ST. LOUIS ings w ere for only a handful of issuers, most of w hich w ere financial firm s. M oody’s Industrial M anual and other similar publications, which today provide some data on com m ercial paper issues by firm s, did not provide such data in 1970. Outstanding com m ercial paper cannot be inferred by looking at firm s’ reported balance sheets. Commercial paper can appear in firm balance sheets either as long-term or short-term debt. Although it is usually included in short term debt, even in that case it cannot be sepa rated from oth er short-term debt (loans from banks, finance companies, and so on). The Board of Governors of the Federal Reserve Sys tem did not collect firm-level data on issuers, only on aggregate am ounts of outstanding is sues, based on dealers’ reports. Despite searches of various publications by the rating agencies, I have been unable to uncover any com prehen sive listing of firms w hich issued com m ercial paper in 1970. Given the lack of data identifying issuers, I use bond ratings to sort firm s according to w hether they w ere likely to have issued com m ercial paper in 1970. In the 1970s, com m ercial paper issuance was usually restricted to the firm s with the highest bond ratings (Standard and Poor’s, 1979, p. 47). Having a AA or AAA rating in 1970 is likely to be the best proxy for the likelihood of being a com m ercial paper is suer. Eight of the 11 nonfinancial firm s whose ratings w ere published in Standard and Poor’s B on d O utlook in 1970 and 1971 w ere rated AA or AAA (the rem ainder w ere A-rated). Also, data from later years indicate a close relationship b e tw een high bond ratings and com m ercial paper access. Standard and Poor's first com prehensive listing of rated com m ercial paper issuers, The C om m ercial P aper Ratings Guide, was published in 1978. Of the 90 nonfinancial firm s that had AA or AAA bond ratings in 1970, 64 w ere issu ing com m ercial paper in 1978. Of the 146 non financial firm s listed in Compustat with AA or AAA bond ratings in 1978, 93 w ere com m ercial paper issuers. In 1978, 94 of the 207 A-rated nonfinancial firm s in Compustat w ere com m er cial paper issuers, and only 43 firm s with bond ratings below A issued com m ercial paper (all of these w ere firm s w ith BBB or BB ratings). Using the AA rating as our cutoff, th erefore, seems 49 advisable. Based on available data, it seems reasonable to assume that a m ajority of AA or AAA nonfinancial firm s w ere com m ercial paper issuers in 1970, and that a m uch sm aller p er centage of rem aining firm s w ere issuers.22 The total num ber of nonfinancial firm s in our sam ple (that is, those without missing observations, and covered by both CRSP and Compustat in 1970) is 1,482. Of these, 90 had bond ratings of AA or AAA. If com m ercial paper issuers experienced a special problem during the crisis, and if Fed in tervention reversed the strain on issuers, the coefficient on the high-rating indicator variable should be negative during the onset of the crisis and positive after Fed intervention. The use of AA or AAA bond ratings as an indicator of a com m ercial paper issuer provides a “conserva tive” m easure of the problem s in the com m er cial paper m arket, for three reasons. First, m easurem ent erro r (the existence of some A-rated com m ercial paper issuers, and of non issuers with AA or AAA ratings) biases the coefficients on the high-rating indicator variable toward zero. Second, the excluded A-rated com m ercial paper issuers likely would have ex perienced the largest adverse effects of the crisis, since their debt was riskier to begin with. Third, the flight to quality during a financial crisis should produce a relative im provem ent in the value of high-rated firms, w hich would im ply positive effects on AA and AAA firm s, after controlling for other firm characteristics, during the onset of the crisis. Table 5 reports regression results for SCARs for two windows around the Penn Central crisis—Ju n e 12 to Ju n e 22, and Ju n e 23 to July 9.23 It is im portant to emphasize th ree points b efore reviewing Table 5. First, coefficients on the control variables in this regression m ust be interpreted cautiously. For example, while rela tively high leverage ratios may have created 22lt is less clear whether the data on A-rated firms in 1978 is representative of A-rated firms in 1970. From 1970 to 1978, market analysts argue that the growth in commercial paper issuers brought more firms with lower ratings (A or BBB) into the market; thus, it might not be appropriate to as sume that 1970 saw the same high proportion of A-rated firms issuing paper as in 1978 (45 percent). For purposes of constructing an indicator variable, given the uncertainty about the number of issuers with A ratings in 1970, it is best to exclude A-rated firms from the group of likely is suers because A-rated firms are a small fraction of total firms with ratings below AA, but a large fraction of AA or AAA firms. problem s for firm s during the crisis, high debt ratios may them selves have been associated with firm attributes (like creditw orthiness) that helped firm s w eather the crisis b etter (and led to relatively higher stock values). Thus, it is not possible to in fer "stru ctu ral” relationships from these cross-sectional findings. The main point of including the control variables is to separate the effect of com m ercial paper issuance per se from factors unrelated to com m ercial paper issuance. Second, the abnorm al retu rns m easures are purged of cross-sectional differences in firm s’ betas that might be correlated with the various regressors. For example, higher debt ratios might be associated with low er retu rns crosssectionally because leverage increases a firm's beta. But, by construction, the abnorm al retu rns used in Table 5 are uncorrelated with the firm ’s beta. Third, squared term s w ere added fo r all regressors, but they do not affect the direction of the results. In no case does a squared term m ore than offset the linear effect of the same variable w hen both coefficients are evaluated at the mean of the regressor (given in Table 6). The direction of association betw een SCAR and any regressor is that of the linear effect. The results reported in Table 5 indicate that the ratio o f debt to assets and the ratio of in com e to net w orth (both m easured at the begin ning of the year) may have been associated with m ore negative retu rns cross-sectionally during the onset of the crisis. Firm size per se had no effect on retu rns in the presence of squared term s for debt ratios. For the period after June 22, the total debt ratio and the profit ratio are associated with a positive effect on returns, indicating a reversal of the stock price move m ents during the period prior to Fed in ter vention. T he inventory-to-sales ratio and the short-term debt-to-assets ratio are both nega tively associated with abnorm al retu rns after Ju n e 22. 23The results reported below are not sensitive to whether June 22—which arguably could have been included in the second window— is included or excluded from either win dow. The results of the first period are driven by pre-June 22 returns, and the results of the second window are driven by post-June 22 returns. MAY/JUNE 1994 50 Table 5 Event Study Regression Results for Standardized Abnormal Returns (standard errors in parentheses) 6/12/70 - 6/22/70 6/23/70 - 7/9/70 (1) (2) (3) (4) Intercept -0 .8 1 (1.19) -0 .4 2 (1.20) -1 .5 7 (1.30) -2 .3 9 (1.31) Debt/Assets -0 .8 1 (0.55) -0 .6 8 (0.55) 0.58 (0.60) 0.31 (0.60) 0.61 (0.74) 0.50 (0.74) -0 .5 0 (0.81) -0 .2 6 (0.81) STD/Assets -0 .7 1 (0.99) -0 .8 0 (0.99) -2 .1 4 (1.08) -1 .9 6 (1.08) (STD/A)-sq. 2.39 (2.30) 2.51 (2.30) 4.31 (2.52) 4.06 (2.50) Size(MVE) 0.09 (0.21) 0.01 (0.21) 0.12 (0.23) 0.29 (0.23) (MVE)-sq. 0.00 (0.01) 0.01 (0.01) 0.00 (0.01) -0 .0 1 (0.01) -0 .8 6 (0.33) -0.82 (0.33) 1.67 (0.36) 1.60 (0.36) (NI/MVE)-sq. 0.82 (0.43) 0.80 (0.43) -0 .6 4 (0.47) -0 .5 9 (0.47) INV/SALES -0 .3 9 (0.40) -0 .5 1 (0.41) - 1 .8 5 (0.44) -1 .5 9 (0.44) (INV/SALES)-sq -0 .0 4 (0.41) 0.04 (0.41) 0.76 (0.45) 0.59 (0.45) — -0 .3 0 (0.15) — 0.64 (016) (D/A)-sq. NI/MVE AA or AAA Adj. R-squared 0.040 0.042 0.080 0.089 A fter controlling for observed balance sheet and incom e characteristics, firm s with AA or AAA bond ratings experienced significant, nega tive, abnorm al retu rns during the onset of the crisis and significant, positive retu rns after Fed intervention. The addition of this indicator vari able increases the adjusted R-squared in both regressions. The evidence provides support for the notion that, in addition to the economy-wide liquidity crisis during the Penn Central crisis, com m ercial paper issuers faced a special prob lem. This, in turn, lends support to the argu ment that discount window subsidization of lending may have been useful in targeting FEDERAL RESERVE BANK OF ST. LOUIS assistance to the com m ercial paper m arket. Thus, the Fed may have been co rrect to divide policy into tw o com ponents: Regulation Q relax ation to provide liquidity to all firm s through banks, and discount window lending to target subsidized assistance to com m ercial paper is suers to offset the special disorder in that m ar ket. That is not to say Fed policy achieved the right mix. For example, negative retu rns for firm s with high inventory-to-sales ratios or high short-term debt after Ju n e 22 may indicate that credit supply was too tight overall. Changes in the Commercial Paper Market After the Crisis The com m ercial paper m arket changed as a result of the Penn Central crisis. In addition to increased diligence in evaluating credit risk, two oth er changes have reduced the possibility of a similar problem in the future. First, in August of 1970, the Fed passed a regulation to restrict the grow th of bank com m ercial paper. Bank paper would be treated, for reserve req u ire ment purposes, the same way as demand or time deposits, depending on the m aturity of the paper. This eliminated the advantages of offbalance sheet financing through bank com m er cial paper and led to the contraction of bank paper. This had little effect on banks or on the grow th of the com m ercial paper m arket, which has been robust (Post, 1992). It simply propelled banks tow ard relying on negotiable CDs (virtual ly identical to com m ercial paper) as an alterna tive source of funds. Of g reater im portance w ere institutional changes in the way com m ercial paper is m ar keted. First, rating agencies made finer distinc tions in their ratings of com m ercial paper issues (Stigum, 1983, p. 637). An im portant elem ent in the rating becam e evidence of com m ercial bank backup arrangem ents behind com m ercial paper programs. Commercial bank support for com m ercial paper program s was a private innova tion. After, and largely as a result of Penn Central, com m ercial paper issuers increasingly sought “hurricane insurance” in the form of backup loan com m itm ents (Stigum, 1983, pp. 633-4; Standard and Poor’s, 1979, p. 47). Most of these loan com m itm ents (roughly 85 percent in 1989) are not credit guarantees to com m ercial paper holders, but rath er promises for as sistance during a general liquidity crisis if the borrow er rem ains creditw orthy (Calomiris, 1989b). W ithin a few years of the Penn Central 51 Table 6 Means, Standard Deviations, and Correlations Among Regressors_______________________________________ Mean Standard deviations C orrelations (p-values in parentheses) STD/A MVE NI/MVE I/S -0 .0 5 (0.05) AA + 0.11 (0.000) D/A 0.28 0.21 0.52 (0.000) -0 .1 1 (0.000) 0.28 (0.000) STD/A 0.07 0.09 — -0 .2 8 (0.000) 0.06 (0.03) 0.26 (0.000) -0 .0 8 (0.003) MVE 11.2 1.6 — — -0 .0 3 (0.22) -0 .1 3 (0.000) 0.32 (0.000) NI/MVE 0.17 0.13 — — — -0 .1 3 (0.000) 0.07 (0.004) I/S 0.17 0.13 — — — -0 .1 9 (0.000) AA + 0.06 0.23 — — — crisis, backup lines w ere almost always 100 p er cent of outstanding issues, except for large, toprated, highly liquid issuers like GMAC or large com m ercial banks. These loan com m itm ents w ere issued by banks for the same reason bank assistance had been relied on during the Penn Central crisis: Banks have access to the discount window and believe that they can rely on the Fed (which m aintains no official policy in this regard) to tem porarily suspend norm al non pecuniary discount window penalties to grant lending subsidies during an em ergency. Institu tionalizing Fed discount window protection through explicit bank loan com m itm ents, one could argue, reduces the tim e to process credit rollover during a crisis. Furtherm ore, the exis tence of clear com m itm ents to lend during a crisis may itself reduce the threat of a general liquidity squeeze and, thus, m ake crises less likely. Currently, the use of backup lines of bank credit, "backed” by access to the discount win dow, has virtually eliminated risk of another Penn Central crisis in the com m ercial paper m arket. But this does not imply an end to the role played by the discount window. The pro tection offered through backup lines of credit depends on banks’ potential access to funds through the discount window. — EVALUATING O TH ER PO SSIBLE FED INTERVENTIONS Thus far, I have argued that both economywide policy (open m arket operations and Regu lation Q relaxation) and targeted discount lend ing may have been desirable interventions during the Penn Central crisis. But the Fed was willing to go beyond these interventions, if necessary, as Chairman Burns’ com m ents cited above indicate. W as the Fed right to have provided fo r the possibility of direct lending to firms, or should it have been willing to rely only on the discount window and open m arket operations? Was the Fed right to have allowed Penn Central to fail in the first place? The Fed’s decision not to prevent the failure of Penn Central is easy to defend. The success o f the capitalist system requires that firm s face "hard ” budget constraints. As refo rm ers in Eastern Europe and the Soviet Union have been saying for years, protecting large corporations from bankruptcy through assistance from the state imposes large costs on m ore successful growing enterprises. More fundamentally, allow ing corporations to fail is what encourages them to succeed. It is w orth emphasizing that the public policy rationale for insulating financial m arkets from tem porary inform ation externali MAY/JUNE 1994 52 ties during panics does not in any way justify bailing out discernably insolvent institutions. W ith regard to the other question—w hether direct Fed lending to corporations is ever justifi able—it is also hard to justify intervention. As Mishkin (1991b) notes, it is b etter to d ecentral ize the decision over who receives how much, and place it in the hands of relatively inform ed bankers who have incentives to avoid making bad loans. If banks had been unwilling to finance the payoff of the com m ercial paper of certain firm s, even on highly subsidized term s, that would have indicated the likely insolvency of those individual issuers.24 Discount window protection should not be used to save individual firm s w hich are viewed as insolvent by their creditors. Of course, creditors are not always right, but part of the rationale for corporate re organization under bankruptcy law (increasingly popular since the 1978 changes in the bankrupt cy code) is to minimize unnecessary costs of li quidating defaulting firm s who turn out to be solvent. Given the availability of the reorganiza tion option, it may be best for the governm ent to allow private m arkets to decide w h eth er in dividual corporate borrow ers are viable. COULD A SIMILAR CRISIS HAPPEN TODAY? Although I have argued that the possibility of another Penn Central crisis today in the com 24Of course, the Fed could have done even more to en courage banks to make pass-throughs than it did during Penn Central by making its subsidy larger. The subsidy that the government can grant is potentially very large. By lowering the discount rate to zero and discriminating in imposing nonpecuniary penalties across banks (for exam ple, charging a zero hassle cost to banks borrowing for tar geted pass-throughs and a prohibitive rate on other borrowing), the subsidy can be increased to the level of the equilibrium fed funds rate without affecting monetary control. 25Gorton and Pennacchi (1992) argue that there is no evi dence for “ contagion” among commercial paper issuers or finance companies. They examined the failures of several issuers and finance companies and found that a failure did not lead investors in securities markets to lower the price of other issuers’ or finance companies’ securities, ceteris paribus. It is premature, however, to interpret this as evi dence that issuers or finance companies are now immune to panics, or more broadly, that financial technology has improved so much that intermediaries are not potentially vulnerable to panics. Gorton and Pennacchi’s sample of events is small, and the events they examine may simply have been transparently idiosyncratic (unlike, for example, the Penn Central crisis). It is possible that events unlike those in their sample could produce panics. Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER mercial paper m arket is rem ote, in other new and growing financial m arkets the potential for a crisis similar to Penn Central may loom larg e r.25 For example, w ithin the banking system a large overdraft default in the Clearing House In terbank Payments System (CHIPS) might lead to a general run of uninsured liabilities o f CHIPS m em bers, due to problem s of unraveling which banks stood to lose from the default. Subsidized lending to CHIPS m em bers might be w arranted to prevent a panic.20 The Fed is cognizant of its potential role in assisting banks in the event of a systemic crisis in the payments system, and it regulates the payments system accordingly. Like many other central banks, the Fed limits over drafts of bank accounts with the central bank and requires private bank clearing systems to limit overdrafts among their m em bers. Such limits include collateral requirem ents, quantity limits on overdrafts, and pre-established losssharing arrangem ents. These regulations are m eant to ensure that the potential protection af forded by the Fed is not abused. It is also conceivable that discount window in tervention could be used to target assistance to m arkets fo r financial derivatives. In the swap m arket, for example, if a m ajor swap provider becam e insolvent, its counterparties, and third parties who have contracted with those coun terparties, could experience unpredictable changes in their m arket risk exposures and, consequently, in their default risks. Because of the interrelatedness of the various positions and 260 f course, the discount window is not the only way to deal with such a problem. Alternatively, deposit insurance could be extended to the CHIPS clearinghouse as a whole. For example, the government could offer insurance to CHIPS with a large deductible, with the liability for the deductible shared by all clearing members. 53 uncertainty as to w hich swap contracts will su r vive the crisis, it might be difficult for cou n ter parties to gauge their true exposure to m arket risk. This could produce a flight to cash by all parties. Fu rtherm ore, a reversal of m arket opin ion about the reliability of swaps as hedging devices could suddenly affect the m arket’s per ception o f firm s with large swap positions. In this case, tem porary disruptions to the supply of credit to certain classes of firm s could con ceivably result. These problem s could motivate discount window subsidies as in the Penn Cen tral crisis. T he lesson in this dismal scenario is not that swaps are a bad idea. They offer real long-run systemic risk reduction as a low-cost vehicle to hedge interest rate risk. But reaping the advan tages of this and other financial innovations requires a period of learning about how to m easure and control the risks created by new financial instrum ents. The existence of the dis count window provides a safety valve to protect the financial system from growing pains like the ones it suffered in 1970. Recent financial inno vations in derivatives and asset securitization may have increased the need fo r the discount window as an instrum ent of public policy. Its role is not just to protect the banking system from systemic runs on com m ercial banks (in deed, it may have little im portance here in the presence of deposit insurance); rath er, its role is to effect occasional, contingent and focused credit subsidies to particular m arkets through banks during mom ents of tem porary disruption, like that of the Penn Central crisis. Another example of a potential application of the discount window is a run on a futures clearing house. Individual clearing m em bers stand betw een all contracting parties and the clearing house provides mutual insurance among all m em bers against default. To limit the risk of default by clearing m em bers, clearing houses impose reserve requirem ents in the form of cash or Treasury bills on open positions and frequently m onitor those positions. Still, it is conceivable that a very large, sudden price drop (say, in the stock market) might bankrupt a clearing m em ber with a large open position and conceivably threaten the clearing house. This could cause a run on the futures m arket as holders of contracts, w ary of the credibility of the clearing house’s solvency, try to sell their contracts. This could amplify the losses to the clearing house and legitimize the initial con cerns that prompted the run, leading to fu rth er cashing-in of positions. If the clearing house w ere to fail, many hedges would disappear with its demise, increasing the risk of many financial claims and causing confusion about the inci dence of the increased risk in ways that might provoke a general liquidity crisis. The Fed could reduce the chance of a run on a futures clearing house and the negative ex ter nalities attendant to such a run by agreeing tem porarily to lend through the discount win dow without penalty to banks making loans to clearing house m em bers, and could even lower the discount rate if necessary to encourage such subsidies. Indeed, this seems a reasonable characterization of the Fed’s response to con cern s about futures m arkets posed by the stock m arket collapse of O ctober 1987. T h ere is a m ore difficult policy question, however, that so far has not been addressed. If banks are unwilling to lend to a clearing house—even on highly subsidized term s—should the Fed let the clearing house fail? On one hand, ad hoc direct lending by the Fed runs the risk of encouraging lax self-regulation within the clearing house. On the oth er hand, the financial disruption from a clearing house failure might generate substantial negative externalities in the financial system. It might be desirable for the Fed to decide w hether it would stand behind the liabilities of failed fu tures clearing houses. If so, the Fed should con sider w h eth er existing private risk-management devices (like margin rules) are adequate. If not, it might recom m end changes to the Commodity Futures Trading Commission, w hich regulates these exchanges. As the volume of derivative transactions expands, so does the need to de velop policies fo r dealing with possible systemic risks related to these m arkets. Identifying a potential benefit from a "backup” discount window does not justify the cu rren t form of the discount window. T h ere may be no benefit from Fed lending to banks during n or mal times, and as Schw artz and others have ar gued, such lending may be costly. T h ere also rem ains the risk that governm ent agencies will abuse even a “reform ed ” discount window by defining noncrises as crises to make loans to fa vored parties. The evidence presented in this paper, th erefore, does not prove that the dis count window has been a net benefit as a poli cy tool, only that it has the potential to provide benefits as well as costs. MAY/JUNE 1994 54 REFERENCES Altman, Edward I. “ Financial Ratios, Discriminant Analysis and the Prediction of Corporate Bankruptcy,” Journal of Finance (September 1968), pp. 589-609. Board of Governors of the Federal Reserve System. Banking and Monetary Statistics: 1941-1970. Board of Governors of the Federal Reserve System, 1976. Bordo, Michael D. “ The Lender of Last Resort: Alternative Views and Historical Experience,” Federal Reserve Bank of Richmond Economic Review (January/February 1990), pp. 18-29. Brimmer, Andrew F. “ Distinguished Lecture on Economics in Government: Central Banking and Systemic Risks in Capital Markets,” Journal of Economic Perspectives (spring 1989), pp. 3-16. Burns, Arthur F. “ Statement to Congress,” Federal Reserve Bulletin (August 1970), pp. 619-26. Business Week. “ Can Stocks Shake the Liquidity Jitters?” June 27, 1970, p. 42. Calomiris, Charles W. “ Regulation, Industrial Structure, and Instability in U.S. Banking: An Historical Perspective,” in Michael Klausner, and Lawrence J. White, eds., Structural Change in Banking. Business One Irwin, 1992a, pp. 19-116. ________ “ Getting the Incentives Right in the Current Deposit-lnsurance System: Successes from the Pre-FDIC Era,” in James R. Barth, and R. Dan Brumbaugh, Jr., eds., The Reform of Federal Deposit Insurance: Disciplining the Government and Protecting Taxpayers. Harper Business, 1992b, pp. 13-35. . “ Do ’Vulnerable’ Economies Need Deposit Insur ance?: Lessons from U.S. Agriculture in the 1920s,” in Phil lip L. Brock, ed., If Texas Were Chile: A Primer on Banking Reform. ICS Press, 1992c, pp. 237-314. ________"Is Deposit Insurance Necessary? A Historical Per spective,” Journal of Economic History (June 1990), pp. 283-95. ________“ Deposit Insurance: Lessons from the Record,” Federal Reserve Bank of Chicago Economic Perspectives (May/June 1989a), pp. 10-30. ________ “The Motivations for Loan Commitments Backing Commercial Paper,” Journal of Banking and Finance (May 1989b), pp. 271-7. , and Gary Gorton. “ The Origins of Banking Panics: Models, Facts, and Bank Regulation,” in R. Glenn Hub bard, ed., Financial Markets and Financial Crises. University of Chicago Press, 1991, pp. 109-74. _______ , Charles P. Himmelberg, and Paul Wachtel. “ Com mercial Paper and Corporate Finance: A Microeconomic Perspective,” Carnegie-Rochester Series (forthcoming, 1994). _______ , and R. Glenn Hubbard. “ Price Flexibility, Credit Availability, and Economic Fluctuations: Evidence from the United States, 1894-1909,” Quarterly Journal of Economics (August 1989), pp. 429-52. _______ , and Charles M. Kahn. “ The Role of Demandable Debt in Structuring Optimal Banking Arrangements,” The American Economic Review (June 1991), pp. 497-513. Cummins, Claudia. “ Fed Proposes Curbing Access to Ad vances by Troubled Banks,” The American Banker (August 12, 1993), p. 2. Diamond, Douglas W., and Philip H. Dybvig. “ Bank Runs, Deposit Insurance, and Liquidity,” Journal of Political Econ omy (June 1983), pp. 401-19. Friedman, Milton, and Anna J. Schwartz. A Monetary History of the United States: 1867-1960. Princeton University Press, 1963. Goodfriend, Marvin, and Robert G. King. “ Financial Deregu lation, Monetary Policy, and Central Banking,” in William S. Haraf, and Rose Marie Kushmeider, eds., Restructuring Banking and Financial Services in America. American En terprise Institute, 1988, pp. 216-53. Gorton, Gary. “ Self-Regulating Bank Coalitions,” University of Pennsylvania Working Paper (March 1989). ________“ Clearing Houses and the Origin of Central Bank ing in the U.S.” Journal of Economic History (June 1985), pp. 277-83. _______ , and George Pennacchi. “ Money Market Funds and Finance Companies: Are They the Banks of the Future?,” in Michael Klausner, and Lawrence J. White, eds., Structur al Change in Banking. Business One Irwin, 1992, pp. 173-214. James, John A. “ The Rise and Fall of the Commercial Paper Market, 1900-1930,” Paper presented at Anglo-American Finance: Financial Markets and Institutions in 20th-Century North America and the U.K., a conference held at New York University, New York, N.Y., December 10, 1993. Janssen, Richard F., and Charles N. Stabler. “ Empty Coffers: Cash Shortage Causes Worry that Big Firms Could Col lapse,” Wall Street Journal, June 12, 1970, p. 1. Kane, Edward J. “ How Incentive-Incompatible Deposit Insur ance Funds Fail,” Ohio State University Working Paper No. 88-35 (1988). Kaufman, George G. “ Lender of Last Resort, Too Large to Fail, and Deposit-lnsurance Reform,” in James R. Barth, and R. Dan Brumbaugh, Jr., eds., The Reform of Federal Deposit Insurance: Disciplining the Government and Protect ing the Taxpayer. Harper Business, 1992, pp. 246-58. ________“ Lender of Last Resort: A Contemporary Perspec tive,” Journal of Financial Services Research (October 1991), pp. 95-110. Maisel, Sherman J. Managing the Dollar. Norton & Co., 1973. McLean, David. “Abreast of the Market,” Wall Street Journal, June 25, 1970, p. 29. ________“Abreast of the Market,” Wall Street Journal, June 24, 1970, p. 31. _______ . “Abreast of the Market,” Wall Street Journal, June 23, 1970, p. 33. Melton, William C. Inside the Fed: Making Monetary Policy. Dow Jones-lrwin, 1985. Miron, Jeffrey A. “ Financial Panics, the Seasonality of the Nominal Interest Rate, and the Founding of the Fed,” The American Economic Review (March 1986), pp. 125-40. Mishkin, Frederic S. “Asymmetric Information and Financial Crises: A Historical Perspective,” in R. Glenn Hubbard, ed., Financial Markets and Financial Crises. University of Chicago Press, 1991a, pp. 69-108. _______ , a n d ________ “ The Efficiency of Cooperative Inter bank Relations: The Suffolk System,” University of Illinois Working Paper (1990). ________“Anatomy of a Financial Crisis,” NBER Working Paper No. 3934 (December 1991b). _______ , and Larry Schweikart. “ The Panic of 1857: Origins, Transmission, and Containment,” Journal of Economic His tory (December 1991), pp. 807-34. Murray, Roger F. “ Lessons for Financial Analysis: The Penn Central Debacle,” Journal of Commercial Bank Lending (December 1971), pp. 42-9. FEDERAL RESERVE BANK OF ST. LOUIS 55 Post, Mitchell A. “ The Evolution of the U.S. Commercial Paper Market Since 1980,” Federal Reserve Bulletin (December 1992), pp. 879-91. Schadrack, Frederick C., and Frederick S. Breimyer. “ Recent Developments in the Commercial Paper Market,” Federal Reserve Bank of New York Monthly Review (December 1970), pp. 280-91. Schwartz, Anna J. “ The Misuse of the Fed’s Discount W in dow,” this Review (September/October 1992), pp. 58-69. Standard and Poor’s. Standard and Poor’s Ratings Guide. McGraw-Hill, 1979. _______ . Commercial Paper Ratings Guide. Standard and Poor’s Corp., February 1978. ________The Bond Outlook. Standard and Poor’s Corp., various issues. Stigum, Marcia. The Money Market. Dow Jones-lrwin, 1983. Timlen, Thomas M. “ Commercial Paper — Penn Central and Others," in Edward J. Altman, and Arnold W. Sametz, eds., Financial Crises: Institutions and Markets in a Fragile En vironment. John Wiley and Sons, 1977. Treiber, William F. “ Problems of Financial Community Under Constant Scrutiny,” The American Banker (October 13, 1970), p. 16. Wall, Larry D., and David R. Peterson. “ The Effect of Con tinental Illinois’ Failure on the Financial Performance of Other Banks,” Journal of Monetary Economics (August 1990), pp. 77-99. Wall Street Journal. “ Patman Asks to See Pentagon’s Dossier on Penn Central Before Any Aid is Given,” June 19, 1970, p. 5. ___________ “ Three Outside Directors Quit Penn Central, Rail Unit; Bank Conflict May Affect Others,” June 15, 1970, p. 7. White, Eugene N. The Regulation and Reform of the American Banking System, 1900-1929. Princeton University Press, 1983. Whitford, David. “ Financial Ratios and the Prediction of Bankruptcy: The Case of Penn Central,” University of Illinois Working Paper (1993). Yang, Catherine, and Mike McNamee. “ Reform, Or A Crack down On Banking?,” Business Week (July 15, 1991), pp. 122-3. MAY/JUNE 1994 57 D a vid C. W h eelock and Paul W. W ilson David C. Wheelock is a senior economist at the Federal Reserve Bank of St. Louis. Paul W Wilson is associate . professor of economics at the University of Texas at Austin. Can Deposit Insurance Increase the Risk o f Bank Failure? Some Historical Evidence l i O M i LOSSES ASSOCIATED WITH declines in energy and agricultural prices, and the col lapse of com m ercial real estate m arkets were the proximate cause of the high num ber of bank and savings and loan (S&.L) failures of the past 12 years. Many researchers also blame governm ent policies, however, such as restric tions on branch banking and limitations on the services that banks and S&,Ls may offer. Such restrictions ham per diversification, thus leaving depository institutions particularly vulnerable to downturns in the regions w hich they serve. Deposit insurance has probably been the most criticized governm ent policy related to bank and S&L failures. Many econom ists believe that deposit insurance encourages banks and S&Ls to take excessive risks, thereby increasing their chance of failing.1 This article investigates empirically the con nection betw een deposit insurance and bank failure. Today, virtually all banks are insured by the Federal Deposit Insurance Corporation (FDIC) and, consequently, isolating the effects of insurance from oth er regulations and exogenous econom ic conditions that affect bank p erfo r m ance is difficult. We study the effects of deposit insurance by drawing on historical evi dence from a voluntary insurance regime that operated in Kansas betw een 1909 and 1929. Be cause m em bership in the Kansas deposit insu r ance system was optional, we are able to com pare insured and uninsured banks facing otherw ise similar regulations and econom ic con ditions in a way not possible with modern data. We estimate a model of bank failure to test for the impact of insurance on the likelihood of failure.2 We find that insured banks were less well capitalized and, in some years, less liquid 1Kane (1989) examines the problems of the S&L industry and the role of government policy. Mishkin (1992), Keeley (1990) and O’Driscoll (1988) discuss the relationship be tween deposit insurance and bank failures in the 1980s. ures of managerial inefficiency help distinguish failing from surviving banks. Grossman (1992) also investigates the effects of deposit insurance by comparing insured and uninsured S&Ls during the 1930s. 2Wheelock (1992b) also investigates how deposit insurance affected the probability of failure for Kansas banks in this era, but employs a different methodology and somewhat different data. Wheelock and Wilson (1993) use the same data set as the present study, but while considering the effects of insurance, focus largely on whether or not meas MAY/JUNE 1994 58 than uninsured banks, and that capitalization and liquidity w ere im portant determ inants of failure. The next section discusses how deposit insu r ance might increase the likelihood of bank failure. Next, we describe the Kansas deposit in surance system and the effects of a collapse of commodity prices in 1920 on com m ercial banks. The final sections develop the econom etric methodology used to model failure, specify the model, and present results and conclusions. D EPO SIT INSURANCE AND BANK FAILURE Federal deposit insurance was enacted in response to the bank failures of the Great Depression. Thousands of banks failed from 1930 to 1933, wiping out the funds of deposi tors, producing a collapse of the money supply, increasing the costs of interm ediation and in ter fering with the clearing of payments.3 Although large banks and many econom ists vigorously op posed deposit insurance, and President Franklin Roosevelt was reluctant to accept it, Congress deemed deposit insurance necessary to protect small, unsophisticated depositors from losses due to bank failures, and the payments system from a wholesale banking collapse like that of 1930-33.4 Until the 1980s, deposit insurance was g en er ally hailed for eliminating the possibility of widespread bank failures.5 M erton (1977) and Kareken and Wallace (1978) showed, however, that w hen insurance premiums are unrelated to the expected cost of failure to the insurance sys tem, banks have an incentive to take greater risks than they otherw ise would. Because depo sitors are protected in the event of bank failure (to the limit of insurance coverage), they have little or no incentive to m onitor their banks’ ac tivities or to demand risk premiums on deposit interest rates. Deposit insurance thus raises 3Studies of the causes and effects of bank failures during the Depression are too numerous to list. Friedman and Schwartz (1963), however, is the seminal investigation of the impact of bank failures on the money supply, and Bernanke (1983) is the most important investigation of non monetary effects of bank failures. “Golembe (1960) and Flood (1992) investigate the rationale for federal deposit insurance. 5For example, see Friedman and Schwartz (1963, pp. 434-42). 6For example, see Kane (1989). Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER the expected retu rn to banks from investing in risky loans and investments and encourages them to substitute debt, in the form of insured deposits, for equity. Consequently, unless regula tions inhibit risk-taking, the presence of deposit insurance could lead to m ore bank and S&L failures than there would otherw ise be. Many econom ists blame deposit insurance, coupled with inadequate regulation and supervi sion, political interference and a failure by regu lators to promptly close insolvent institutions, for the high num ber of S&L failures and bank ruptcy of the Federal Savings and Loan Insur ance Corporation during the 1980s.6 The bank ing industry’s problem s were, by comparison, less notorious. Banks faced higher capital re quirem ents and were m ore stringently super vised than S&Ls, w hich lessened banks’ incentive and ability to take excessive risks. But deregula tion of deposit interest rates, initiated by the Depository Institutions Deregulation and M one tary Control Act (DIDMCA) of 1980, the gradual removal of b arriers to branch banking, more liberal chartering polices and increased com peti tion from foreign banks and from nonbank financial institutions, all worked to lessen ch art er values and increase the incentive for banks (as well as S&Ls) to take greater risks.7 M ore over, in 1980, deposit insurance coverage was in creased from $40,000 per account to $100,000 for both banks and S&Ls, while the failure reso lution policy known as "too-big-to-fail” effectively extended insurance to all deposits at the largest banks, thereby enhancing their incentive to take risks.8 As is all too often the case, the bank and thrift debacle of the 1980s stemmed in part from the failure of policy m akers to heed les sons from the past. Flood (1992) argues that when deposit insurance legislation was being considered in 1933, policy makers understood the temptation that insurance gives bankers to take excessive risks. Accordingly, coverage was 7Keeley (1990) draws the connection between increased competition, deposit insurance and increased risk-taking. 8Too-big-to-fail was implemented to reduce that the failure of a very large bank could temic crisis, with depositor runs on many (1992) and Boyd and Gertler (1993) argue increased risk-taking by very large banks. the possibility produce a sys banks. Mishkin that this policy 59 limited to $2,000 per account and regulations were imposed to constrain risk-taking. Deposit interest rate ceilings prevented weak institutions from growing rapidly by bidding up interest rates, and regulators gave bankers added incen tive to act conservatively by limiting the issu ance of new bank charters. Many of the sources of increased com petition for banks and S&Ls that had em erged by 1980, such as money market mutual funds and the com m ercial paper market, were the product of technological changes that could not be foreseen in 1933.9 But deregulation of bank and S&L deposits and the expansion of deposit insurance coverage at a time w hen the industry was facing increased com petition contradicted the regulatory princi ple that underlay deposit insurance legislation in 1933. tions.1 Contemporaries believed that deposit in 1 surance had contributed to the high num ber of failures because it protected incom petent and dishonest bankers from m arket discipline.1 In 2 the following sections, we investigate empirically how deposit insurance might have contributed to the failure of banks operating under the deposit insurance system of Kansas during the 1920s. We study this case because just th ree of the eight state insurance systems had optional mem bership for state-chartered banks and, hence, perm it com parison of insured and unin sured banks facing otherw ise similar conditions. Of these, only the Kansas system lasted for many years with a large num ber of banks elect ing to join the system and a significant num ber remaining uninsured. The insights that policy m akers had in 1933 about deposit insurance cam e partly from prior state experiences with deposit insurance. Six states had experim ented with insurance in the pre-Civil War era, as did eight others betw een 1908 and 19 3 0 .1 None of the 20th-century sys 0 tems was able to fully protect depositors of failed banks from loss, and each closed before the onset of the Great Depression. The commodity-price collapse of 1920-21 triggered a wave of bank failures throughout the Midwest and the South, including seven of the eight states with deposit insurance. Although loan losses associat ed with the decline of state incom es was the proximate cause of bank failures, insured banks generally suffered higher failure rates than uninsured banks facing similar exogenous condi D EPO SIT INSURANCE IN KANSAS 9See Wheelock (1993). 10The 20th-century states and the years in which their insur ance systems operated are Oklahoma (1907-23), Texas (1909-25), Kansas (1909-29), Nebraska (1909-30), South Dakota (1909-31), North Dakota (1917-29), Washington (1917-29) and Mississippi (1914-30). Cooke (1909), Robb (1921), American Bankers Association (1933), Federal Deposit Insurance Corporation (1956) and Calomiris (1989) compare the features and performance of the systems. The Kansas deposit insurance system began operation in Ju n e 1909 and officially closed in 1929. Kansas was the second state to enact in surance legislation following the Panic of 1907, and was motivated partly by the adoption of an insurance system by Oklahoma in 1908.13 In contrast to the Oklahoma system, in w hich all state-chartered banks were required to carry in surance, the Kansas system was made optional for state banks because of complaints that insur ance forces conservatively managed banks to in sure depositors of banks that are m ore likely to fail.1 The state o f Kansas, like oth er states with 4 deposit insurance systems, did not guarantee in surance payments. In contrast to the experience of Reserve City Bankers (1933) and Robb (1921) for con temporary views about insurance. 13Robb (1921) describes previous attempts to enact deposit insurance legislation in Kansas and other states, and notes that Kansas banks located near the Oklahoma border were especially strong proponents of deposit insurance in Kan sas (pp. 107-12). 14The Comptroller of the Currency ruled in 1908 that national banks could not join state deposit insurance systems. 1'Thies and Gerlowski (1989) and Alston, Grove and Whee lock (1994) find that a state’s bank failure rate during the 1920s was higher if it had a system of deposit insurance, holding constant other possible causes of failure. Wheelock (1992a) reports similar evidence at the county level for Kansas. 12Commenting about the effects of the Kansas deposit insur ance system, Harger (1926, p. 278) wrote that insurance “ gave the banker with little experience and careless methods equality with the manager of a strong and conser vative institution. Serene in the confidence that they could not lose, depositors trusted in the guaranteed bank. With increased deposits, the bank extended its loans freely.” See also American Bankers Association (1933), Association MAY/JUNE 1994 60 with federal insurance in the 1980s, depositors, not taxpayers, suffered from any insurance fund deficiencies.1 5 Kansas banks were required to operate for at least one year and undergo an examination by state authorities before being admitted into the insurance system .1 Insured banks w ere also re 6 quired to maintain total capital of at least 10 percent of total deposits, and surplus and undis tributed profits of at least 10 percent of total capital.1 At first, deposit insurance was restrict 7 ed to noninterest bearing accounts, savings deposits of $100 or less, and time deposits of b e tw een six and 12 m onths maturity. Banks with insured deposits were not permitted to pay more than 3 percent interest on any deposit, w hether insured or n ot.18 Regulations were relaxed in 1911; insurance was extended to all deposits not otherw ise secured, including sav ings accounts in excess of $100, and the state banking com m issioner was given authority to adjust interest rate ceilings as he deemed appropriate. Insured banks were assessed annual premiums equal to 1/20 of 1 percent of their insured deposits less total bank capital. Although a bank could reduce its premium by increasing its capi tal, the saving was small. A bank with $100,000 of insured deposits and $10,000 o f capital was assessed an insurance premium of $45, w hereas a bank with $15,000 of capital had a premium of $42.50. Additional premium s could be as sessed to cover shortfalls in the insurance fund, but total annual premium s w ere capped at 1/4 of 1 percent of insured deposits less capital. Banks also w ere required to place cash or eligi ble bonds with the state banking com m issioner equal to 0.5 percent ($500 minimum) of their in sured deposits to guarantee insurance premium payment. Banks could withdraw from the insur ance system at any time, but rem ained liable for any premium s needed to reim burse depositors of banks which failed while the withdrawing 15Mississippi, however, ultimately issued bonds deficit of its insurance system. to retire the 16The requirement of one year of operation was waived if no other bank in the applicant’s town was an insurance sys tem member. 17The former requirement was eliminated in 1917. Warburton (1958, p. 21) argues that, if maintained and enforced, the requirement could have prevented much of the rapid growth of banks that ultimately resulted in large losses to the insurance system. Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER bank was insured, including the six m onths fol lowing notice of withdrawal. Deposit insurance proved popular in Kansas, and before 1920 the deposits of insured banks grew m ore rapidly than those of uninsured state and national banks. Figure 1 plots the participa tion rate of all Kansas banks and of those eligi ble for deposit insurance. Figure 2 illustrates the shares of all bank and eligible bank deposits held by insured banks.19 The percentage of the state’s bank deposits held by insured banks peaked in 1921 at 43.8 percent, and membership in the system peaked at 65.6 percent of eligible banks in 1923. In that year, 681 banks, holding $168 million of deposits, belonged to the insu r ance system, while 357 state banks, holding $64 million of deposits, did not. THE CHARACTERISTICS OF INSURED BANKS This section identifies some im portant differ ences betw een insured and uninsured banks that may explain why the failure rate of insured banks exceeded that of uninsured banks. If depositors believe that they will be protect ed from loss in the event of bank failure, they will be willing to accept a lower rate of retu rn on their deposits than they would in the ab sence of such protection. Because it lowers the cost of deposits, deposit insurance encourages banks to rely m ore heavily on deposits to finance their activities, as opposed to equity and nondeposit liabilities, than they otherw ise would. Economic theory suggests that banks also will choose to hold riskier assets w hen deposits are insured.20 Insured banks in Kansas had a higher failure rate than uninsured banks, which might have been caused by "m oral haz ard,” that is, by high-risk behavior encouraged by deposit insurance. Alternatively, because risky banks would stand to gain the most from insurance in term s of lower deposit costs, the 18For comparison, the annual average interest rates on prime four-six month commercial paper and on call loans in 1909 were 4.69 and 2.71 percent, respectively. 19AII banks include those with federal charters, trust compa nies and unincorporated banks. The source of these data is the FDIC (1956, p. 68). 20See Merton (1977) or Kareken and Wallace (1978). 61 Figure 1 Proportion of Banks Participating in the Deposit Insurance System Figure 2 Proportion of Deposits in Insured Banks MAY/JUNE 1994 62 failure rate of insured banks might have been higher simply because risky banks were more likely to join the voluntary insurance system, that is, because of “adverse selection." Of course, both effects might have been present and con tributed to the higher failure rate of insured banks. T he troubled history of the Kansas deposit in surance system raises the question of w hether depositors expected an insurance payoff in the event of bank failure. If they did not, then depo sitors would have had an incentive to m onitor their banks' activities and to demand the same term s from a m em ber of the insurance system as from an uninsured bank with equal likeli hood of failure. Indeed, if depositors thought that insured banks had, on average, a higher probability of failure and that an insurance payoff was unlikely, then they would have had an incentive to tran sfer deposits from insured banks to uninsured banks. No doubt some depo sitors did so, as the relative share of deposits in insured banks fell after 1921. Large num bers of depositors left their funds in insured banks, however, and because of the difficulty of assess ing the extent of protection from deposit insur ance at any point in time, might have expected at least partial reim bursem ent in the event of bank failure.2’ To investigate the relationship betw een deposit insurance and bank behavior, we com pare vari ous financial ratios of insured and uninsured banks in our sample in different years. Table 1 reports the m ean capital/assets, deposits/assets and cash reserves/deposits ratios of insured and uninsured banks in our sample in each year for which data are available.22 In general, insured banks maintained less capital relative to assets than uninsured banks and, hence, w ere more likely to fail as a result of loan losses or other declines in asset values. The hypothesis that the mean capital/assets ratios of insured and unin sured banks are equal can be rejected (at the .10 level or better) in each year. 2,Wheelock and Kumbhakar (1994) argue that before 1926, depositors had a reasonable expectation of an insurance payoff, and show that deposit insurance enabled members of the insurance system to hold lower capital ratios than uninsured banks until that year. 22The biennial reports of the state banking commissioner (Kansas, various years) provide balance sheet data for all state-chartered banks and trust companies on August 31 of each even-numbered year (except 1912 and 1916). Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER The greater reliance o f insured banks on deposits is indicated by the fact that, except for 1924, insured banks had higher deposits/ assets ratios than uninsured banks. Insured banks also held few er liquid assets (“reserves”), defined here as cash, cash items and the liabili ties of other banks, relative to deposits than uninsured banks in 1910, 1914 and 1924. Thus, for some of the period, insured banks w ere less liquid than their uninsured competitors. We find the reserves/deposits ratio to be particularly useful for distinguishing failing and nonfailing banks. The comparatively low capital/assets and reserves/deposits ratios of insured banks indicate that they w ere m ore risky than unin sured banks and, hence, the higher failure rate of insured banks is not surprising. We fu rth er examine the impact of deposit insurance on the probability of failure, and seek to identi fy oth er characteristics which distinguish fail ing from nonfailing banks in the following sections. THE BANKING COLLAPSE OF THE 1920s The num ber of banks and total bank deposits grew rapidly throughout the United States in the first two decades of the 20th century, espe cially during the inflationary boom of World War I. Kansas experienced a 30 percent increase in the num ber of banks betw een 1910 and 1920, w hen it had 1,096 state-chartered banks, 266 na tional banks and 18 unincorporated banks (Kan sas, 1920, and Bankers Encyclopedia Company, M arch 1921). After 1920, the num ber of banks in the United States fell sharply, especially in the Midwest and the South, w here waves of bank failures followed a collapse of commodity prices. Between Ju n e 1920 and January 1921, an index of wholesale commodity prices fell from 167 to 114; by January 1922, it had fallen to 91 (Board of Governors, 1937, p. 174). Sharply lower in comes left many farm ers who had borrow ed to 63 Table 1 Mean Financial Ratios of Insured and Uninsured Banks Year Type of bank Capital/ assets Deposits/ assets Reserves/ deposits Number of banks 1910 Insured Uninsured .188*** .238 .793*** .728 .320*** .424 41 186 1914 Insured Uninsured .205** .227 .755*** .717 .303** .341 124 128 1918 Insured Uninsured .124*** .142 .864*** .824 .328 .308 149 97 1920 Insured Uninsured .133** .148 .836** .806 .207 .200 158 84 1922 Insured Uninsured .162** .179 .784* .755 .205 .206 159 74 1924 Insured Uninsured .163* .179 .808 .798 .224*** .282 150 63 1926 Insured Uninsured .150** .172 .835** .810 .227 .233 135 63 Note: ***, ** and * indicate that the difference in the means for insured and uninsured banks is statistically different from zero at the .01, .05 and .10 levels (two-tailed tests). finance land acquisition and improvements b e fore 1920 unable to repay their loans. Loan loss es, in turn, caused the failure of many banks in commodity-producing regions, including 220 in Kansas betw een 1920 and 1929. The impact of agricultural distress on in dividual Kansas banks reflected the portfolio choices they had made prior to the collapse and as it unfolded. Between Septem ber 1920 and Septem ber 1926, 122 state-chartered Kansas banks failed. Of those, 94 had been m em bers of the insurance system (a 4.6 percent failure rate) and 28 had not (a 2.3 percent failure rate). By contrast, just six national banks failed (a 0.8 p ercent failure rate). Over the life of the insur ance system, depositors of just 27 failed banks recovered the entire amount of their insured deposits, and those of two oth er banks received 93 and 95 percent of their deposits, respectively (Warburton, 1958, pp. 27-9). No insurance pay m ents w ere made to depositors of 88 m em ber banks that failed (FDIC, 1956, p. 58). On aver age, holders of insured deposits received 53 p er cent of their funds from liquidation of bank assets and 18 percent from the deposit insur ance fund (7 percent o f which cam e from the reorganization of one bank, the American State Bank of Wichita). The rem aining 29 percent of insured deposits w ere never recovered. The sharp increase in bank failures beginning in 1920 quickly swamped the resources of the Kansas deposit insurance fund. W hen a m em ber of the Kansas insurance system failed, its depos itors were given interest-bearing certificates immediately upon closure, and received reim bursem ent only after the bank’s assets had been entirely liquidated. If the proceeds from liquida tion were insufficient to reim burse insured depositors, the insurance system was supposed to make up the difference. Depositors of the two banks that failed before 1920 w ere eventually fully reim bursed, but inadequate insurance funds m eant that depositors of most banks that failed after 1920 were not as fortunate. Because depositors were not reim bursed until after liquidation of a failed bank’s assets, the condition of the fund and the prospect that depositors of failed banks would eventually receive full reim bursem ent were difficult to de term ine at any point. The failure in Ju n e 1923 of the Am erican State Bank of Wichita, the state’s largest insured bank, presented the insur MAY/JUNE 1994 64 ance system with its greatest challenge. Eventu ally, the bank was reorganized with oth er in sured banks assuming $1.4 million of the loss and depositors accepting, on average, 40 percent of their deposits in the form o f stock in the new bank. The event marked a turning point in the history of the Kansas insurance system, however, as the num ber of banks and the deposits held in insured banks began to decline.23 Although a special insurance assessm ent was collected in 1922 and insurance premium s w ere set at their legal maximums beginning in 1924, losses from bank failures exceeded insurance system revenues from 1921 onward. In 1925, the state bank com m issioner stopped making pay m ents on all insurance claims, and in 1926 a state suprem e court decision effectively ended the system. The court decision resulted from the refusal of several banks that had withdrawn from the insurance system to pay additional in surance premiums. The court ruled that banks could withdraw w ithout additional liability by simply giving up the bonds they had pledged to guarantee premium payments. This led many banks to withdraw and, by 1927, insurance sys tem m em bership had fallen to less than 10 p er cent of eligible banks. Kansas appears to have suffered many of the problem s that have been associated with the bank and S&L debacle of the 1980s. In the 1980s, many depository institutions, especially insolvent S&Ls, bid up deposit interest rates and grew rapidly by issuing deposits through brok ers.24 In the 1920s, some banks appear to have evaded deposit interest rate ceilings in order to grow rapidly. In his report for 1922 (Kansas, 1992, p. 5), the state bank com m issioner also felt it desirable to limit deposit insurance to only the original holder of a deposit, and not to any as signee. Supervision was also reported to have been weak in Kansas, especially during the w orst failure years, and for a time state banking authorities perm itted weak and insolvent banks to rem ain open rather than closing them immediately upon recognition of trouble (War- 23See Wheelock and Kumbhakar (1994) and Warburton (1958) for additional detail about this failure. 24See Kane (1989). 25Kiefer (1988) provides a good introduction to the analysis of duration data; Kalbfleisch and Prentice (1980) and Lan caster (1990) provide more advanced treatments of the subject. RESERVE BANK OF ST. LOUIS FEDERAL burton, 1958, p. 19). W hether any such banks recovered is not known, but the lack of m ention in the biennial reports of the state banking com m issioner suggests that, like the attem pts at fo r bearance during the 1980s, the policy was probably not successful. MODELING TIME-TO-FAILURE W hile many Kansas banks failed during the 1920s, a m ajority of banks survived the decade. W hat characteristics distinguish the survivors from the failures? To identify im portant ch arac teristics o f failing banks, we employ an econo m etric technique that explicitly models timeto-failure. The analysis of duration data is rela tively new in economics. Engineers and bio medical scientists have analyzed time-to-failure for electrical and m echanical com ponents of m achinery and the survival times of subjects for many years, but econom ists have only recently begun to apply similar models, prim arily in the area of labor econom ics with a focus on the du ration of spells of unem ploym ent.25 Although models developed to analyze duration data are sometimes called time-to-failure models, the event of interest need not be characterized as a "failure”; all that is necessary is that the event be well-defined. Duration models differ from standard discrete choice models (such as probit or logit models) in that duration models use inform ation about how long banks survive in the estim ation of the in stantaneous probability of failure for a given set of observations on the independent variables. Param eter estim ates thus indicate w h eth er an increase in the value of an individual indepen dent variable will reduce or extend the expected time until failure occurs. By contrast, discrete choice models typically ignore inform ation about the timing of failures, and provide an estimate only of the probability o f failure within a given interval of time. Discrete choice models treat all banks that fail during an interval the same, as they do all surviving banks. Thus, for example, a bank that fails on the first day of a two-year in- 65 terval is treated the same as a bank that fails on the last day, and a bank that survives the interval but fails one day after that period ends is treated the same as a bank that sur vives an additional 10 years. Duration models explicitly incorporate such inform ation, and thus yield m ore efficient param eter estim ates.26 A detailed description of the duration model used in this article is presented in the appendix. In the present application, we observe the ch arter date for each bank in our sample. For some banks, we observe a failure date, w here failure is defined as the date on which the bank was ordered closed by the state banking com missioner. For the remaining banks, no failure date is observed if a bank had not failed by the end of our observation period (1928) or if it liq uidated voluntarily, m erged with another bank or switched to a federal charter. These observa tions are considered censored; inform ation about these banks is available for part of their lives, but we do not observe them failing. Cen soring is common in duration data of all types and must be addressed within the statistical model used to examine the data. Figure 3 illustrates the types o f censoring that may occu r in duration data. Assume that the in terval over which banks are observed runs from time tt to f,. The horizontal lines in the figure represent the time betw een the ch arter date and the date of failure for individual banks. Given the observation period (tt, f2), the observa tion for Bank A will be both left- and rightc e n s o r e d . For this bank, neither the ch arter date nor the failure date occu r within the ob ser vation interval. The observation for Bank B will be left-censored; the ch arter date does not oc cu r within the observation interval, but the failure date does. For Bank C, both the ch arter and failure dates occu r betw een ti and tz, and so the observation is uncensored. Finally, the ob servation for Bank D will be right-censored; the ch arter date occurs within the observation in ter val, but the failure date occurs after f,. 26While deriving a direct relationship between the parameters of a duration model and a discrete choice model would be difficult, in principle one could integrate the hazard func tion estimated from a duration model to obtain the proba bility of failure within a given interval of time. EXPLANATORY VARIARLES AND ESTIMATION RESULTS O ther researchers have employed hazard and discrete choice models to identify characteristics that distinguish failing and surviving banks in a variety of settings. W hite (1984), for example, es timates a probit model to distinguish failing from nonfailing banks during the Banking Panic of 1930. W heelock (1992a) uses a similar model to study Kansas bank failures betw een 1920 and 1926. Both studies found that banks were more likely to fail, the lower their capital/assets, sur plus/loans, bonds/assets, reserves/deposits, or deposits/assets ratios.27 Banks were m ore likely to fail, the higher their loans/assets or sh ort term borrowed funds/assets ratios.28 Many Kansas banks experienced significant loan losses following the collapse of agricultural prices and incom es in 1920-21, and banks with low capital/assets ratios were less well-cushioned against declines in the value of their assets. Banks with little cash and oth er reserve assets were less able to m eet deposit withdrawals, and the smaller a bank's reserves/deposits ratio, the more likely it was to close due to illiquidity. Often a lack of cash was the first sign that a bank was in trouble, and would prompt closure by state banking authorities. Just as a low level of reserves signaled trou ble, so too did a heavy reliance on borrowed funds such as rediscounts of loans with other banks or with the Federal Reserve. Banks that relied heavily on borrow ed funds to finance their operations, or that had to resort to b o r rowing because of loan losses or deposit w ith drawals, appear to have been relatively more likely to fail. Loans are generally the most risky and least liquid of bank assets, and the loan portfolios of the rural unit banks of Kansas were undoubted ly not well-diversified. Accordingly, the higher a bank’s loans/assets ratio, the greater the likeli hood that it would fail. On the other hand, banks with substantial bond holdings might 28Borrowed funds consisted largely of rediscounted loans with the Federal Reserve or other banks, 27Surplus refers to paid-in capital beyond the par value of a bank’s stock plus undistributed profits. Reserves refer to cash, cash items and the liabilities of other banks. MAY/JUNE 1994 66 Figure 3 Possible Types of Censoring Time have been less likely to fail, especially since U.S. Government bonds and bonds of the state of Kansas and of Kansas municipalities probably com prised most of the bond holdings of Kansas banks in this era.29 full impact of insurance may not be captured by observable variables. The deposit insurance dummy variable might reflect the incentive that insurance gives banks to hold riskier loans and investments than they otherw ise would. W heelock (1992b) includes bank size and a dummy variable indicating w hether or not a bank was a m em ber of the state deposit insur ance system as additional explanatory variables. If larger banks w ere better diversified, or could capture econom ies of scale, they might have been less likely to fail. W heelock found, how ever, no significant relationship betw een size and failure. Deposit insurance, on the other hand, did significantly affect the probability of failure. Even though the capital/assets ratio and other m easures of risk-taking should reflect w hether or not a bank had insured deposits, the W heelock (1992b) did not test for interaction effects betw een deposit insurance and the finan cial ratios. One might expect, however, that the effect of a change in a financial ratio on the likelihood of failure would depend in part on w hether or not the bank had deposit insurance. For example, the depositors of an insured bank might have been less concerned w ith a decline in the capital/assets ratio of their bank and, hence, less likely to demand a higher deposit in terest rate than depositors of an uninsured bank. The scope for risk-taking and, thus, the probability of failure, resulting from a change 29The state banking commissioner accepted only U.S. Government, state of Kansas and Kansas municipal bonds to guarantee payment of deposit insurance premiums. Unfortunately, we do not have information on the composi tion of each bank’s bond holdings. Digitized forFEDERAL RESERVE BANK OF ST. LOUIS FRASER 67 in a financial ratio might therefore depend on w hether or not a bank was insured. We test this hypothesis here. Estimation Results Our data consist of a panel of Kansas banks for which we have collected balance sheets and other inform ation as of August 31 of each evennum bered year from 1910 to 1926 (except 1912 and 1916, w hen these data w ere not pub lished).30 Our sample includes 259 banks (ap proximately one-fourth the total operating in 1914).3' Of these, 4 7 (18 percent) had failed by Septem ber 1, 1928. Banks that merged with other banks, liquidated voluntarily or switched to a federal ch arter are treated as censored on the date of m erger or change in charter. Banks that did not fail or otherw ise ceased operating prior to August 31, 1928, are treated as cen sored on that date. In addition to the independent variables used by W heelock (1992b), we include dummy varia bles for each interval of 1920-22, 1922-24, 1924-26 and 1926-28 to investigate w hether the probability o f failure differed across periods for a given set o f bank attributes. Only two banks in our sample failed before 1920 and, hence, we do not include dummies for those years. In one specification we also include interaction term s of deposit insurance and the financial ratios.32 Table 2 reports estim ates of the failure model that include alternative com binations of explana tory variables. In column one, the coefficient on the capital/assets, bonds/assets and reserves/ deposits ratios indicate that the higher each of these ratios was, the less likely a bank was to fail. Better capitalized banks, and those with substantial bond holdings and significant reserves, could b etter absorb the shock of loan losses and deposit withdrawals accompanying the agricultural downturn in Kansas. Banks that had substantial borrow ed funds relative to assets had a greater chance of failing while, contrary to expectations, it appears that the higher a b ank’s loan/assets ratio, the less likely it was to fail. This finding appears due to multicollinearity, however. T he loans/assets ratio is highly co r related with the reserves/deposits ratio. If the latter is omitted, as in the specification reported in column two, the sign of the coefficient on the loans/assets ratio is positive, though not statisti cally significant. The coefficient on deposit insurance is not statistically significant, suggesting that any effect that insurance had on the probability of failure is captured by its relationship with the financial ratios also included in the model. If the dum mies for the biennial observation intervals are omitted, the coefficient on insurance is larger and statistically significant. It may be that the strain on the portfolios of all banks caused by the collapse of commodity prices overwhelmed the effect of deposit insurance on the unob served portfolio risk of insured banks, which could explain why the coefficient on insurance is not significant w hen the time dummies are in cluded. Not surprisingly, for given values of the financial ratios, banks were m ore likely to fail after the collapse of commodity prices and on set of severe agricultural distress in 1920. Final ly, none of the coefficients on the interaction term s of deposit insurance and the financial ra tios is statistically significant. Again, it appears that any impact of deposit insurance on the likelihood of failure is captured by differences in the financial ratios betw een insured and unin sured banks. CONCLUSION Researchers have blamed federal deposit insur ance for contributing to the high num bers of bank and th rift failures and large deposit insur ance payoffs since 1980. Unless insurance premiums increase proportionately with risk, banks will be encouraged to take greater risks than they otherw ise would. This article presents 30The source of our data is Kansas (various years). 3,We dropped seven banks because of missing data. Others fall out of the panel after failing, closing voluntarily, merg ing with other banks, or switching to a national charter. 32Few state-chartered Kansas banks were members of the Federal Reserve System during this era. None of the failed banks in our sample was a member, and so differences in supervisory agency or regulation, except those pertaining to deposit insurance, cannot explain variation in failure probabilities across banks. MAY/JUNE 1994 68 Table 2 Failure Model Estimates Variable 0.17 (0.43) 0.35 (0.90) -0.23 (0.03) Capital/Assets -12.10 (2.76)*** -13.42 (3.02)*** -17.48 (2.52) Bonds/Assets -7.98 (1.84)* -1.56 (0.35) -14.84 (1.19) Loans/Assets -6.05 (2.18)** 2.12 (0.96) -4.82 (0.53) Reserves/Deposits -9.80 (3.79)*** — -9.83 (1.20) Borrowings/Assets 8.10 (5.16)*** Insurance Total Assets 8.34 (5.45)*** 5.31 (1.80) -0.78 (0.84) -1.00 (1.00) -0.77 (0.83) 1920-22 1.10 (1.18) 1.33 (1.43) 1.07 (1.14) 1922-24 2.30 (2.68)*** 2.66 (3.12)*** 2.29 (2.65) 1924-26 2.46 (2.75)*** 3.10 (3.57)*** 2.47 (2.77) 1926-28 3.08 (3.44)*** 3.67 (4.15)*** 3.11 (3.46)' Ins x capital/assets — — 6.66 (0.74) Ins x bonds/assets — — 8.32 (0.63) Ins x loans/assets — — -1.53 (0.16) Ins x reserves/deposits — — -0.26 (0.03) Ins x borrowings/assets — — 3.79 (1.09) Psuedo-R2 Log-likelihood .29 -138.25 .26 -144.61 .30 -137.16 Notes: Absolute values of f-statistics are in parentheses; ***, ** and * indicate statistically significant at the .01, .05 and .10 levels. some historical evidence of how deposit insur ance can alter bank behavior and increase the likelihood that a bank will fail. As in the 1980s, when falling incom es in agricultural and energyproducing states caused large loan losses and led to many bank and thrift failures, a sharp decline in agricultural incom es in the early 1920s caused the failure of many com m ercial banks in rural areas. Not all banks failed, RESERVE BANK OF ST. LOUIS FEDERAL however; in fact, most survived the collapse. Banks that carried deposit insurance had a higher rate of failure than oth er banks. Our findings, along with those of similar historical studies, show that insured banks were less well capitalized and less liquid than other banks. Es timates of a model of time-to-failure indicate that among banks in our sample, those with high ratios of capital to assets, reserves to 69 deposits, large bond holdings relative to their to tal assets, or that relied little on borrowed funds, were less likely to fail. In short, conserva tively managed banks w ere less likely to fail and, at the same time, banks that carried deposit insurance were m ore risky and, hence, more likely to fail than their uninsured com petitors. Johansen, Soren. “ An Extension of Cox’s Regression Model,” International Statistical Review (August 1983), pp. 165-74. Kalbfleisch, J. D. and R. L. Prentice. The Statistical Analysis of Failure Time Data. John Wiley and Sons, 1980. Kane, Edward J. The S&L Insurance Mess: How Did It Hap pen? The Urban Institute Press, 1989. Kansas. Biennial Report of the Bank Commissioner (various years). REFEREN CES Kareken, John H., and Neil Wallace. “ Deposit Insurance and Bank Regulation: A Partial-Equilibrium Exposition,” Journal of Business (July 1978), pp. 413-38. Alston, Lee J., Wayne A. Grove and David C. Wheelock. “ Why Do Banks Fail? Evidence from the 1920s,” Explora tions in Economic History (1994, forthcoming). Keeley, Michael C. “ Deposit Insurance, Risk, and Market Power in Banking,” The American Economic Review (December 1990), pp. 1183-1200. American Bankers Association. The Guaranty of Bank Deposits (1933). Kiefer, Nicholas M. “ Economic Duration Data and Hazard Functions,” Journal of Economic Literature (June 1988), pp. 646-79. Anderson, Per K., and Richard D. Gill. “ Cox’s Regression Model for Counting Processes,” Annals of Statistics (December 1982), pp. 1100-20. Association of Reserve City Bankers. The Guaranty of Bank Deposits (1933). Bankers Encyclopedia Company. Polk’s Bankers Encyclopedia (March 1921). Board of Governors of the Federal Reserve System. All Bank Statistics, 1896-1955 (1959). ________Banking and Monetary Statistics, 1914-1941 (1943). _______ . Annual Report (1937). Calomiris, Charles W. “ Deposit Insurance: Lessons from the Record,” Federal Reserve Bank of Chicago Economic Per spectives (May/June 1989), pp. 10-30. Cooke, Thornton. “ The Insurance of Bank Deposits in the West,” Quarterly Journal of Economics (Nov. 1909), pp. 85-108. Cox, David R. “ Partial Likelihood,” Biometrika (May/August, 1975), pp. 269-276. _______ . “ Regression Models and Life Tables,” Journal of the Royal Statistical Association (1972), pp. 187-220. Efron, Bradley. “ The Efficiency of Cox’s Likelihood Function For Censored Data,” Journal of the American Statistical As sociation (September 1977), pp. 557-565. Federal Deposit Insurance Corporation. Annual Report (1956). Flood, Mark D. “ The Great Deposit Insurance Debate,” this Review (July/August 1992), pp. 51-77. Golembe, Carter. “ The Deposit Insurance Legislation of 1933: An Examination of its Antecedents and its Purposes,” Politi cal Science Quarterly (June 1960), pp. 181-200. Lancaster, Tony, The Analysis of Transition Data. Cambridge University Press, Inc., 1990. Merton, Robert C. “ An Analytic Derivation of the Cost of Deposit Insurance and Loan Guarantees,” Journal of Bank ing and Finance (June 1977), pp. 3-11. Mishkin, Frederic S. “ An Evaluation of the Treasury Plan for Banking Reform,” Journal of Economic Perspectives (Winter 1992), pp. 133-53. O’Driscoll, Gerald P. “ Bank Failures: The Deposit Insurance Connection,” Contemporary Policy Issues (April 1988), pp. 1-12. Robb, Thomas B. The Guaranty of Bank Deposits. Houghton Mifflin Co., 1921. Warburton, Clark. Deposit Guaranty in Kansas, unpublished manuscript, Federal Deposit Insurance Corporation, 1958. Wheelock, David C. “ Is the Banking Industry in Decline? Recent Trends and Future Prospects From a Historical Perspective,” this Review (September/October 1993), pp. 3-22. _______ . “ Regulation and Bank Failures: New Evidence from the Agricultural Collapse of the 1920s,” The Journal of Eco nomic History (December 1992a), pp. 806-25. _______ . “ Deposit Insurance and Bank Failures: New Evi dence from the 1920s,” Economic Inquiry (July 1992b), pp. 530-43. _______ , and Paul W. Wilson. “ Explaining Bank Failure: Deposit Insurance, Regulation and Efficiency,” working paper no. 93-002A, Federal Reserve Bank of St. Louis (1993). Grossman, Richard S. “ Deposit Insurance, Regulation and Moral Hazard in the Thrift Industry: Evidence from the 1930s,” The American Economic Review (September 1992), pp. 800-21. _______ , and Subal C. Kumbhakar. ‘“ The Slack Banker Dances’: Deposit Insurance and Risk-Taking in the Bank ing Collapse of the 1920s,” Explorations in Economic His tory (1994 forthcoming). Harger, Charles Moreau. “ An Experiment that Failed: What Kansas Learned by Trying to Guarantee Bank Deposits,” Outlook (June 23, 1926), pp. 278-79. White, Eugene N. “ A Reinterpretation of the Banking Crisis of 1930,” Journal of Economic History (March 1984), pp. 119-38. MAY/JUNE 1994 70 A p pen dix The P ro p o rtio n a l H azard M odel This appendix describes the proportional haz ard model estimated in this article in some de tail for interested readers who are unfam iliar with duration models, but have some und er standing of econom etrics or statistics. We wish to estim ate the effect of deposit insurance and other variables on the probability of failure at particular tim es for the n banks in our sample. Let T , i = 1, ..., n, represent the failure time for the zth bank in our sample; w here T is not ob served, we say that the observation is censored. Time is m easured relative to the individual bank's ch arter date, with a zero value at the ch arter date. Hence, for each bank in the sam ple, the corresponding time scales will have different values for a given calendar time. If T is a continuous random variable with a continuous probability distribution/(f), w here f is a realiza tion of T, then the cumulative probability is ration of banks. Furtherm ore, all four functions are related. From equations 2 and 3, it follows that Also, rearranging term s in equation 3 yields (5)/(f) = S(f)A(f). Another useful function is the integrated hazard function, (6) A(f) = f0A(u) du. Then, from equations 4 and 6 the survival fu n c tion may be w ritten as (7) S(t) = e~A (”, and from equation 7 we have (1) P robiT < f) = F(t) = \ f{u ) du. 0 ‘ The function F(t) gives the probability that a bank fails b efore time f (subscripts are omitted w here no ambiguity results). Alternatively, the same inform ation may be expressed in term s of the survival function (2) S(t) = 1 - F(f), which is m erely P rob (T > f). Given that a bank has survived until time f, what is the probability that it will fail during the next short interval of time, A? The function characterizing this aspect of the problem is the hazard rate, given by (3) A(f) = lim Prob(t < T < t + A|T > f) ~ A0 — = lim A0 — A F(t + A - Fit) ) 7~7T AS(t) = M . S(t) The hazard function gives the instantaneous rate of failure per unit time period at time f. The density function f(t), the cumulative densi ty function F(t), the survival function S(f) and the hazard function A(f) each characterize the du FEDERAL RESERVE BANK OF ST. LOUIS (8) AU) = -lo g S(f). Estimation of the failure time relationship re quires specifying a functional form for either /(f), F(f), S(t), A(f), or A(f). Note that a functional form need only be specified for one of these functions; the relations in 1 and 4-8 will imply a functional form for the rem aining functions. We use the proportional hazard relationship developed by Cox (1972), w here (9) A(f|*,p) = A 0(f)exP , w here ^ is a row vector of m easured covariates and ft is a column vector of param eters with the appropriate dimensions. This model assumes a baseline hazard, A 0(f), w hich in principle amounts to an unidentified param eter for each bank in the sample. Thus, A accounts for any unob 0(f) served heterogeneity among the banks that might otherw ise bias the param eter estimates. The covariates in x influence the overall hazard for each bank through the exponential term s in equation 9 (the choice of an exponential form here is common throughout the literature on hazard estimation and simplifies the estimation problem relative to choices of other functional forms). The model is sem iparam etric since the exponential in 9 is a param etric form, while the baseline hazard involves an unspecified form 71 and, hence, is nonparam etric. Consequently, the model is m ore flexible than models in which the failure time distribution is assumed known ex cept perhaps for a few scalar param eters. Given the hazard specification in equation 9, the corresponding survivor function (which gives the probability of survival up to time t) may be w ritten as (10) S(f|*,/}) = exp - (X (u)e^ du For uncensored observations with failure at time T, the contribution to the likelihood is f(T\?c)-, for observations censored at time T, the contribu tion to the likelihood is S(T|x), that is, the proba bility of survival until time T. Cox (1972, 1975) suggests a partial-likelihood approach w hich can be used to estim ate the param eters of the hazard function in 9. Assume, for the m oment, that no observations are cen sored, and that the observations are ordered by their completed, untied durations such that tt < t., < ... < tn. The conditional probability that observation 1 fails at time tt, given that any of the n observations could have failed at time tt, is A jjtj,/? U ) (1 1 ) Attjx,,/?) The equality results from the assumption of the proportional hazard in 9; the baseline hazard A(f) cancels out of the expression on the left in 11. The expression in 11 gives the contribution of the first observation to the partial likelihood. Analogously, the contribution of the yth observa tion to the partial likelihood is eV1 1 S . <-j . e'.» The partial likelihood is given by the product of the individual contributions and, hence, its log is (12) L(p) = E U,/J - log E j-> Andersen and Gill (1982) and Johansen (1983) 'A atively, o e co specify a param fo fo the ltern n uld etric rm r baseline hazard inequation 9and m ize the co axim rresponding likelih o fu ctio . A o ghthe p o d n n lth u artial-likelih o od approach avoids the needfo an arbitrary param r etric specification o the baseline hazard, there is a loss o f f efficiency inthe resulting estim relative tothose o ates b show that the partial likelihood can be treated as an ordinary likelihood concentrated with respect to A 0.33 The model represented by equation 12 can be easily adjusted to accom modate censoring in the data. For the data used in this study, each bank i in the sample is observed at J different times ti l, < ti2 < ... < t.„ with either failure or ceniJ ’ soring occurring at time tu. Note that tim es here refer not to calendar time, but to time relative to the date of ch arter for bank i so that ti0 = 0 w here tiB is the date of ch arter for the ith bank. The balance sheet inform ation used in ^ co r responding to time t.., j = 1, ..., { J .- l ) , are as sumed to reflect the position of bank i over the interval [f.., tjij+1)). The model estimated in this paper is time-varying in the sense that covariates in are assumed constant for intervals of time [f , tllj+t)), but may vary across different in tervals. Thus, for the ith bank there are ( J .- l ) observed intervals; the first ( J.-2 ) are both leftand right-censored, and the last is left-censored and also right-censored if failure tim e is not ob served for the ith bank. To accommodate the censoring in the data, let *.{/), i = l , n, j = l , J denote the vector of covariates for bank i during period j. Covariates are fixed within a given period, but may vary over different periods. Let d equal 1 for banks that are observed to fail at some time within the entire observation period, and zero otherwise. Assume that banks are ordered by increasing date of failure. Then, the log-partial likelihood becom es S I k-i e 'V Kiefer (1988) suggests that the intuition behind the partial-likelihood approach used here is that, in the absence of any inform ation about the baseline hazard, only the order of the durations provides inform ation about the unknown param e ters of the model. In both 12 and 13, the instan taneous probability of failure is normalized by the sum of instantaneous probabilities of failure for all other banks that could have failed at the same time as the ith bank. tained b m izin the fu likelih o . See E n(1 7 ) y axim g ll od fro 9 7 fo a discussion o this efficiency lo r f ss, MAY/JUNE 1994 F e d e ra l R e se rv e B a n k o f St. Louis Post Office Box 442 St. Louis, Missouri 63166 The R ev iew is published six times p e r year by the Research and P ublic Inform ation Department o f the Federal Reserve Bank o f St. Louis. Single-copy subscriptions are available to the public fr e e o f charge. Mail requests f o r subscriptions, back issues, o r address changes to: Research and P ublic Inform ation Department, Federal Reserve Bank o f St. Louis, P.O. Box 442, St. Louis, Missouri 63166. The views expressed are those o f the individual authors and do not necessarily reflect o fficia l positions o f the Federal Reserve Bank o f St. Louis o r the Federal Reserve System. Articles herein may be reprinted provided the source is credited. Please provide the Bank’s Research and Public Inform ation Department with a copy o f reprinted material.