The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Economic Review Federal Reserve Bank of Cleveland Fall 1984 R isk in Large-Dollar Transfer Systems .................................... 2 Settlement risk is the risk that a bank will be unable to repay other banks for daylight credit, meaning the net amount (payments minus receipts) owed for a day’s transactions on a payment system. Payments volumes have been growing rapidly on Fedwire and even more rapidly on privately operated systems. This growth highlights the need for man agement of settlement risk exposure. One vital ingredient of settlement risk manage ment is the ability to control daylight credit extended or used by a bank across the various payment systems. Equally vital is the need for banks and their customers to recognize who is at risk in making and receiving largedollar payments. Sources of Change in Rates of Return on Capital: 1952-82 ............. 17 This article reconsiders the problem of a de clining rate of return on capital. The rate of return for noncorporate businesses and corporations fell after 1965. A model designed to examine sources of the decline indicates that the major forces at work were inflation ary distortions in the relative price of capital and falling capital productivity. Other con tributors were a one-time-only price-cost dis tortion in the late 1960s and periodic cyclical weakness. None of these effects, perhaps ex cepting falling capital productivity, can be considered a permanent or long-term erosion in the profitability of capital. Economic Review is published quarterly by the Research Department of the Federal Reserve Bank of Cleveland, P.O. Box 6387, Cleveland, OH 44101. Telephone: 216/579-2000. Editor: Pat Wren. Assistant editor: Meredith Holmes. Design: Jamie Feldman. Typesetting: Lucy Balazek. Economic Review Opinions stated in are those of the authors and not necessarily those of the Fed eral Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted provided that the source is credited. Please send copies of reprinted materi als to the editor. ISSN 0013-0281 A n assistant vice president with the Federal Reserve Bank of Cleveland, the author is re sponsible for Fed eral Open Market Committee briefings and other monetary policy advice. The author has benefited from the comments of Donald Hester, Ernest Patrikis, David Humphrey, James Hoehn, Roger Hinderlitcr, Kim Kowalewski, John Carlson, and Mark Sniderman. Risk in Large-Dollar Transfer Systems by E.J. Stevens Risk in large-dollar transfer systems was an unfamiliar topic until quite recently. Even today, people who are not banking profession als probably have little notion of what these systems are, what risks they involve, and why those risks should be investigated. In a nutshell, large-dollar transfer sys tems in the United States are telecommuni cations networks (currently including Fed wire, CHIPS, CashWire, and CHESS) almost entirely dedicated to same-day handling of multimil lion dollar payments among banks (see box 1). These payments may be for a bank itself, as it buys and sells money, or for the accounts of the bank’s customers, especially financial institutions, active in world money and capi tal markets, and nonfinancial corporations. The risk being discussed is settlement risk: that, at the end of a day, a participating bank will not be able to pay the net amount (pay ments minus receipts) it owes to the other banks on one of the systems for that day’s transactions. How to deal with this risk may be viewed differently by banks, their custom ers, and the Federal Reserve System. NevBox 1 Large-Dollar Transfer Systems CashWire is operated by the Payment and Administra tive Communication Corporation owned by a consortium of 180 U.S. banks. Service is currently provided to 17 of these banks, which send approximately 350 payments daily with an average value of about $700,000. CHESS, or Clearing House Electronic Settlement Sys tem, is operated by the Chicago Clearing House Associa tion with service available to members, who must be in the 7th Federal Reserve District. Service is currently pro vided to six banks, which send approximately 450 mes sages daily with an average value of about $1.0 million. CHIPS, or Clearing House Interbank Payments System, is operated by the New York (City) Clearing House Asso ciation with service available to its members. Service is currently provided to over 120 institutions, which send approximately 75,000 payments daily with an average value of about $3.0 million. Fedwire is operated by the 12 Federal Reserve Banks and their branches with service available to any depository institution with an account relationship with a Reserve Bank. Service is currently provided to over 7,000 insti tutions, which send approximately 150,000 payments daily with an average value of about $2.4 million. 2 Federal Reserve Bank of Cleveland 1. For example, see “Risks in Elec tronic Payments Systems, ’’ Report of the Risk Task Force, Association of Reserve City Bank ers, October 1983. . See Board of Gov ernors of the Fed eral Reserve System, “Proposals to Re duce Risk on LargeDollar Transfer Systems,” Docket No. R-0515. 2 3. Automated Fedwire payments, for example, involve a fee of 55C to each of the sending and receiving banks. The banks them selves would have to charge substan tially more than this to cover the 55C fee plus inhouse costs. Non-pecuniary costs may loom just as large. Specifically, it is sometimes ar gued that small-value wire transfers are more likely to go astray. A bank that fails to receive a transfer, standing to gain only the over night return on the value of the payment, has little incentive to initiate a search fora m issing$ 1,000 payment; other par ties to the transaction may be relied on to do the job. The incen tive is a thousand fold greater on a million-dollar trans fer, which is there fore likely, if lost, to be sought assiduously until found. 3 ertheless, there is a common concern that the risk be recognized and methods of risk man agement be understood.1To that end, the Board of Governors of the Federal Reserve System has formally asked for industry comment on a number of proposed methods of risk man agement and control.2 The purpose of this article is to examine the concept of settlement risk (section II). This requires, in addition, a background description of the institutional basis for making largedollar payments in the United States (section I) and an explanation of why concern about set tlement risk has only recently come to the fore (section III). I. Large-Dollar Payments and Settlement Payment Systems The nature of large-dollar transfer systems in the United States is best understood by con trasting them with small-dollar systems on the one hand and, on the other, with nonpay ment message systems. Small-dollar transfer systems—cash, checks, automated clearing houses (e.g., for direct deposit of Social Secu rity or salary payments)—are familiar to most people. Whereas an average cash transaction is for considerably less than $100, and even the average check (including business checks) is for less than $1,000, the average wire trans fer (the generic name for payments made on large-dollar systems) is in the million-dollar range. This is not to imply that wire transfers are somehow restricted to such large sums: small payments can be made as readily as large. However, routine wire transfer of small payments is usually un-economic because the value of gaining interest for a day or so by completing a same-day payment is not worth the additional cost of a wire transfer.3 Wire transfers typically are completed on a telecommunications system, although tele Economic Review • Fall 1984 phone and other devices can be employed. Fledgling debit card and home banking sys tems exist, some automated clearinghouse transactions are now made on a computer-tocomputer basis, and electronic check collec tion is contemplated. Nonetheless, small pay ments typically are not completed on tele communications systems. Large-dollar transfer systems should also be contrasted to inter-bank message systems that, standing alone, do not necessarily effect payments. SWIFT and Bankwire II, for exam ple, are technologically sophisticated tele communications systems; however, they are only capable of conveying messages. Those messages may be—frequently are—instruc tions authorizing a bank to remove funds from the account of one depositor and place them in the account of another. When both deposi tors maintain accounts at the bank receiving the message, only follow-up internal book keeping is required to complete the process. However, if the account to which the payment is directed is at any other bank, then no pay ment has been made until a follow-up inter bank payment has been initiated via one of the large-dollar transfer systems. In this sense SWIFT and Bankwire II, despite restricted access, standardized formats, and a sophisti cated telecommunications vehicle for trans mission, are closer to telex, telephone, and mail service than to large- or small-dollar transfer systems. What distinguishes largedollar transfer systems from these message systems is the ability of one participant to transfer cash in the form of immediately available bank balances to any other partici pant by a single message. In the jargon of payments, these large-dollar transfer systems make payments because they include settle ment—the irrevocable transfer of ownership of bank balances from one participant to another. This transfer is analogous to pay ment in cash in the form of legal tender cur rency and is brought about by the transfer of deposit balances at Federal Reserve Banks from the account of the paying bank to that of the receiving bank. Settlement Risk The risk that is the major concern in largedollar transfer systems is settlement risk, reflecting the possibility that the recipient’s bank will not be paid by the payor’s bank. This risk should be distinguished from the more familiar risks confronted in ordinary payments, including the simple risk of non payment of a debt and the risk of being paid with a bad check (see box 2). Settlement risk is the risk that a bank, despite a paying customer’s adequate balance or credit line, will be unable to cover payments it has initi ated on various payment systems during the banking day. Box 2 Three K inds of R isk A and B are banks. A' and B' are customers of those banks. 1. Nonpayment A'owes money, but may fail to do any thing about it. B' is exposed to a credit risk: A' may be a deadbeat. 2. Bad Check: A' pays B' by check, but has insufficient funds in his account with A to cover the check. A'gives the check to B\ who deposits it in B. B sends the check to A for payment, receiving provisional credit in B’s ac count at the Federal Reserve Bank, where A's account is provisionally debited. A then finds A'has insufficient funds and returns the check to B via the Federal Reserve, which reverses the previous entries. 3. Settlement Risk: Upon instruction from A\ A wires funds to B for credit to B'. At the end of the day, A has insufficient funds to pay B the amount it owes for this (and other) payments, even after subtracting other payments made by B to A. Federal Reserve Bank of Cleveland The circumstances under which this might happen are extreme. Inability of a bank to settle would be the symptom not just of a bank failure but of an unexpected bank fail ure. Indeed, it might be argued that, at least for very large banks, such an event could not happen. After all, in the event of trouble, wouldn’t the supervisory authorities be sum moned immediately to arrange a last-resort loan from its Reserve Bank? In this way, every other participant in a large-dollar transfer system would be paid; no settlement failure would occur. This argument ignores two vital matters. First, as explained in the next para graph, settlement risk would not disappear; it would simply be transformed into credit risk of the Federal Reserve or other creditors of the troubled institution. Second, as explained in section II, the supervisory authorities and the Federal Reserve might be well-advised to adopt a “hands-off” policy in the case of an incipient and isolated settlement failure, allowing other participants in a network to absorb the losses occasioned by settlement failure. Settlement risk is, therefore, the risk that no other bank, supervisory authority, or government lender of last resort will lend enough to a bank to cover its debit position vis-a-vis another institution before the end of the day. Exposure to settlement risk may be borne by the recipient of a payment or by his bank, depending on which payment system is util ized. Settlement risk for Fedwire payments is borne by the Reserve Banks, which stand as the bankers (B in box 2) for the depository institutions receiving payments (B' in box 2). Notification of a payment sent by a Reserve Bank to a recipient bank is an irrevocable notice that immediately available funds have been credited to its account. If the Reserve Bank discovers at the end of the day that the paying bank has insufficient funds to cover its Fedwire payments, the Reserve Bank can 4. These matters are spelled out in the Federal Reserve’s Regulation / Sec tion 210.36, Final Payment and Use of Funds: (a) Final Payment. A transfer item is finally paid when the transfer ee’s Reserve Bank sends the transfer item or sends or tele phones the advice of credit for the item to the transferee, whichever occurs first, (b) Right to Use Funds. Credit given by a Reserve Bank for a transfer of funds becomes avail able for use when the transfer item is finally paid, subject to the Reserve Bank s right to apply the transferred funds to an obligation owed to it by the trans feree. See “Regula tion J: Collection of Checks and Other Items and Wire Tra nsfers of Fu nds, Federal Reserve Regulatory Service, vol. Ill, section 210, as amended, effec tive April 2, 1984, p. 7-043. not recover the funds from the recipients.4 The location of settlement risk exposure in CHIPS is different from Fedwire. This largedollar system is not a bank, but a clearing house with net settlement. Participating banks receive notices of payment during the day from each of the other banks and send notices of payments to the other banks. The clearinghouse records all of these payments, but no transfer of assets between partici pants takes place. At day’s end, each bank then is in a net debit or credit position with respect to each other bank, reflecting the net balance of payments to, relative to payments from, each of those other banks. Of course, the sum of all these bilateral net debit and credit positions is always zero because what each bank owes is what each other bank is owed. This accounting identity makes clear why settlement of the day’s payments can be such a simple matter. Each bank in net debit position with respect to the aggregate of all other participants might pay what it owes into a pool from which each bank in net credit position could be paid. In this way 120 partic ipants would be able to settle 7,140 bilateral positions [(120 X 119) -r- 2] with a total of only 120 payments, some into and some out of the clearinghouse settlement account (see box 3). The facts of CHIPS settlement are both more and less simple than these few sen tences suggest. CHIPS participants include both settling and non-settling banks. Non settling banks both send and receive pay ments in their own names. However, at the end of the day, they pay or receive the net balance due from or to them through one of the settling banks rather than directly with the settlement account at the Federal Reserve Bank of New York. A settling bank agrees to receive the net credits or pay the net debits attributable both to its own activity as well as to that of each of its non-settling respondents. Settlement then proceeds in several steps at the end of the day. First, net positions are Economic Review • Fall 1984 reported so that each participant has an oppor tunity to verify all of its bilateral net debit and credit positions and its aggregate net debit or credit (“net net”) position. Then, settling banks must indicate their readiness to settle, before which they have had an opportunity to confirm that their non-settling respondent participants in debit positions either have a sufficient balance with the settling bank, or are able to borrow a sufficient amount from the settling bank, to cover the debit. When all settling banks have signaled their readiness to settle, those in debit positions send the amounts they owe from their Federal Reserve Bank deposit accounts via Fedwire to the CHIPS settlement account at the Federal Re serve Bank of New York. When all payments are received, the balance in the settlement account is dispersed via Fedwire to the Fed eral Reserve Bank deposit accounts of the settling banks in credit positions. Settlement risk in CHIPS, then, refers to the possibility of two different kinds of failure to pay. One would be the inability of a non settling participant to cover its debit position, either by funding its account with the settling bank or by a loan from that bank. The other would be the inability of a settling bank to fund its account at the Federal Reserve Bank. Locating the risk of loss in these hypothet ical situations involves identifying how settle ment is handled if one of the participants is unable to make settlement. CHIPS’ Rule 13 specifies the series of steps for operating under such circumstances. First, more time can be granted, in effect extending the end of the day beyond the appointed settlement time (nor mally 5:45 pm). This extension may be granted by the clearinghouse because a settling bank notifies the clearinghouse of its unwilling ness to settle for one of the non-settling par- 5. Deletion of trans actions does not re lieve the participant of its obligation to make those pay ments, according to Rule 13. ticipants for whom it would otherwise be obligated to settle, or because a settling bank notifies the clearinghouse that it is unable to cover its own debit position by settlement time. The extra time would allow a non-settling participant to seek funding and/or an alter native settling bank, or would allow a settling participant to secure funds. If, despite extra time, a participant is still unable to settle, the day’s transactions of that participant (both payments and receipts) can be deleted from the settlement.5 This means that the aggregate net debit or credit position of each remaining participant will change by the amount of its bilateral net credit or debit for the day’s transactions with the deleted insti tution. The remaining participants might then settle, if those with increased net debit positions have or can acquire sufficient funds to cover their new, larger settlement obliga tions. The deleted institution would then be come the legal quarry of a host of unhappy customers, other banks and their customers, the clearinghouse, and regulatory authorities. Box 3 The Benefits of a Net-Settlement System Suppose 120 banks agree to send and receive payment messages among themselves during the day and to settle only the net amounts due to and due from one another at the end of the day, using deposit balances at Reserve Banks as the medium of settlement. At the end of the day, each of the 120 banks either owes or is owed by each of the 119 other banks. Thus, there will be 7,140 (half of 120 X 119) pairs of net credit and debit positions, or bilateral net positions, to be settled. In the absence of a net-settlement system, settlement of these 7,140 positions would involve a payment by each of the debit position banks to each of the corresponding credit position banks. With a net-settlement agreement, settlement requires only 120 payments, some into and some out of the set tlement account at a Reserve Bank. Those owing more than they are owed (at least one bank) pay that “netnet” amount to the settlement account; those that are owed more than they owe are paid that “net-net” amount from the settlement account. When settlement is com plete, the settlement account is exhausted because the sum of all banks’ positions is precisely zero: what each bank owed was what another bank was owed. 6 Federal Reserve Bank of Cleveland The possibility of deleting payments helps to clarify the difference between provisional and final payment. Payments made via inter bank net settlement networks for large-dollar payments are provisional until settlement is complete. This comes at the end of the day, when funding of the settlement account by net debit position banks is complete and pay ments have been sent from the settlement account to the banks in net credit positions. Between the morning opening of a network and evening settlement, payments are provi sional in the sense that the receiving bank may not actually receive credit to its deposit account at a Reserve Bank. Those who use the proceeds of such payments without any other guarantee are exposed to the risk that the payment will not become final because of settlement failure. Settlement risk is more narrowly focused in CashWire and CHESS. All participants in these systems are settling banks using their own accounts at Reserve Banks to pay to or receive from the settlement account. Also, at least as an interim matter, all CashWire participants guarantee irrevocable availabil ity of funds to their receiving customers. Set tlement risk is thus absorbed by the partic ipating banks in CashWire, while it is shared with the customers of participating banks in CHIPS. II. The Nature of Risk Exposure Bilateral Risk Exposure Identifying how losses might be realized in large-dollar transfer systems is not the same as evaluating the extent of risk exposure of any individual participant in those systems. Bilateral settlement risk exposure may be thought of as the expected value of the cost to a participant of a settlement failure by a par 6 . The discussion here proceeds as though credit is being extended by the receiving bank to the paying bank , rather than to the customers of the receiving bank. This initial presumption is relaxed in the next section. ticular counterparty bank. Aggregate settle ment risk exposure of a bank is the sum of the expected values of the cost of settlement failure evaluated across all counterparties in a large-dollar transfer system. Conceptually, then, a bank would measure its bilateral set tlement risk exposure as the expected value of the unrecoverable portion of its bilateral set tlement position (net credit) with respect to the *th other bank. This risk exposure of one bank to another, ignoring any interdependence among all banks, has two ingredients.6 One is the actual dollar value of the bilateral net credit extended by B to the ith bank in a par ticular payment system during a day; the other is the expected value of the percentage of this position that will not be recovered. Consider first the dollar value of the bilat eral net credit extended by B to i. This amount will vary each day and during the day with the ebb and flow of payments to and from one another, and can be represented symbolic ally as $Bi. Next, consider the unrecovered percen tage of the bilateral net credit extended by one bank to another. That a bank is unable to settle at the end of a day would impose a cost on its net creditors in the form of interest foregone on the unsettled position for the period of time the position remained unset tled. Of course, this cost might be recovered, in effect writing an ex post loan and then being repaid, but costs of negotiating and litigation would remain. Alternatively, if the bank had failed, there might be some delay before its obligations were settled, involving both admin istrative and waiting costs. Finally, the un settled position might turn out to be unrecov erable in part, in total, or (including litigation and waiting) more than total. All of this potential write-down is what is imagined to underlie each participant’s esti mate or expectation of the percentage unre covered of each net bilateral credit position Economic Review • Fall 1984 linking it to another bank. Further, it seems plausible that these estimates are not made with certainty, but are better imagined as being distributed across a potentially wide spectrum of values, ranging from just above zero (complete recovery) to more than 100 per cent (no recovery plus costs of waiting and litigation). The probability of any particular percentage, j, actually being realized from the ith bank could then be stated as A whole spectrum of such probabilities of per centage unrecovered would exist, with their distribution tightly clustered near zero for counterparty banks of good repute and long relationship, and less tightly clustered, more evenly spread above zero for less reputable, less well-known counterparties. Combining the bilateral net credit position and the probability of percentage unrecovered, settlement-risk exposure of bank B to the ith counterparty bank can be represented by the weighted average of all possible outcomes, E (X B i) = j> 0 Lest this characterization of risk exposure seem too remote from the real world, recognize that something like this risk-exposure calcu lation must be employed by any bank that makes loans. The expected return on a loan must incorporate both the explicit yield and the likelihood of costs on nonrepayment. In the large-dollar transfer system case, there is credit extended (the net bilateral credit position, $Bi) but no explicit yield because the credit is to be repaid before the end of the day and, with few exceptions, interest is charged only on loans with maturities of overnight or longer. For this reason the net bilateral credit posi tion is often referred to as daylight credit extended by B to i, or as Vs daylight over draft with B. Sharing Risk exposure of B to i is affected by some thing in addition to the bilateral credit posi tion and the creditworthiness of i as per ceived by B. This additional matter involves the degree to which others share B ’s exposure to i. The more widely shared an exposure is, the better for B in two distinct ways. The first would be the extent to which funds, $b are at risk. As in a case in which B is receiv ing payments from i for the accounts of a num ber of customers, B ’s exposure is reduced to the extent that its customers might both share litigation and collection costs as well as bear the burden of amounts that are ultimately unrecoverable. In a more limited way, some cost sharing might be possible among a num ber of banks all of whom were in net credit position when i failed to settle. The second way would skew the spectrum of probabilities of percentage unrecovered toward zero. More widespread sharing of exposure may benefit B in that more numer ous sources of independent scrutiny of fs creditworthiness may reduce the chances of i s pursuing imprudent courses of action. This is no more than the fooling-all-of-the-peopleall-of-the-time impossibility theorem at work. In this application, the theorem reflects the increasing likelihood that (true) adverse infor mation will become available to counterpar ties as the number of counterparties increases. The power of information in constraining behavior and reducing risk exposure depends in part on who gets the information. A single customer of a bank, receiving adverse infor mation about another bank from which pay ments might be received, can simply refuse to do business with customers of the suspect bank or request payment in some safer means. While this may be a prudent course of action for the payee, it has only limited power to force constructive change in the suspect bank, which may be able to ignore the qualms of Federal Reserve Bank of Cleveland one or a few isolated alarmists among its many customers, only some of whom may switch to a more sound bank. Less likely is a bank to be able to ignore the qualms of one or a few counterparty banks representing the interests of many customers. Banks are in a better position to profit from, and there fore would likely be more aggressive in seeking, information about the creditworthiness of counterparty banks. Similarly, the power of information to constrain risky behavior is likely to be greater when in the hands of a bank than in the hands of a customer. It is the credit judgments of other banks that provide the foundation of a bank’s liquidity, deter mining its ability to obtain short-term fund ing in the interbank money market. Counter party banks would be more likely to maintain a truly less risky posture, including smaller aggregate daylight overdrafts, when other banks have continuing daily incentive to dis cover their creditworthiness. Systemic Risk All of the foregoing discussion of risk expo sure implicitly has assumed that a settlement failure is an isolated event; that losses because of a settlement failure would be small enough that banks could absorb the resulting unex pected loss of funds without themselves being unable to settle. Upon investigation, this is an overly strong assumption that ignores the interdependence of participants in a payment system and the resulting vulnerability of many banks directly or indirectly to a single counterparty’s failure to settle. Vulnerability to counterparty failure re flects both the speed with which a bank must gain access to cash and the likely scar city of cash. An unexpected settlement failure would occur at the end of a day when most banks have finished matching their sources with uses of cash through purchases and sales of funds in the money market. Unexpected non-repayment of a daylight credit then removes one expected source of cash, which must be replaced before the close of Fedwire prevents further transfers of cash that day. Even though the Fedwire facility might be kept open beyond the normal 6 pm close if there were trouble, emergency financing arrange ments would have to be made in an hour or so, or before the next morning. Speedy access to cash might be sought in a bank’s own balance at a Reserve Bank. This would require no action, because the cash already would be on deposit and, for reserve requirement purposes, the unexpected drain might be offset by larger cash holdings over future days. However, for many of the nation’s largest money center banks, the size of daylight overdrafts is many times larger than their own or any counterparty bank’s normal overnight reserve deposit position. For this reason, cash balances would be inad equate in preventing vulnerability to a coun terparty failure. If insufficient cash is avail able in the bank’s own account, then steps might be taken to sell liquid assets. However, many banks would have difficulty doing this because liquid assets would already have been serving as collateral for other purposes and therefore be unavailable. Their alternative would then be to borrow. Whether borrow ing or, in exceptional cases, selling unpledged liquid assets, the challenge is both to find other banks with the required amount of cash as well as to convince them to make the cash available. Which of these hurdles—finding cash or acquiring it—would be the more trou blesome is a debatable matter, for neither would be easy. Finding banks with cash may be difficult, given the limited stock of cash on which the U.S. financial system operates. The rele vant concept of cash is, again, in the form of deposit account balances at Federal Reserve Banks (because they can be transferred via Fed wire before the end of the day). Vault cash Economic Review • Fall 1984 is unlikely to help, being of limited amount, cumbersome denomination, dispersed location, and probably sealed in time-locked vaults. Deposit balances at correspondent banks would be useful to a single bank, but their use would merely relocate the cash shortage to the correspondent bank. In this sense, the search for end-of-day cash is one of musical chairs in which the cash needs (players) are no larger than the available stock (chairs), but matching need with available stock will be accompanied by pandemonium, and the pan demonium may last longer than the few hours available to settle. Clearly, there is almost no excess cash in the system to be mobilized on short notice if by excess we mean excess reserve deposits. Excess reserves are widely distributed as “small change” across the entire banking system; if they were available to be mobilized into sellable quantities, the federal funds rate would already have induced their owners to bring them to market before the end of the day. The total reserve deposits of banks with cash may be available to lend to those w ith out cash to the extent that reduced holdings one day can be offset by enlarged holdings (or reduced requirements) later in a two-week reserve maintenance period or by carry-over of deficiencies into the next period. Further, just as one bank must cover an unrepaid daylight credit in the event of a settlement failure, so, too, other banks will have un expected cash surpluses because of unsettled payments to the failed institution. On balance, the surpluses are less than the deficits only in the amount of the funding gap that triggered the initial bank failure. In general, then, suf ficient cash will exist in the aggregate, but its redistribution must be accomplished very quickly, after the time at which most m ar kets have begun to close, and subject to the perceived creditworthiness of deficit banks. The difficulty arises, therefore, not just in the scarcity of total cash but in the distri bution of unexpected cash deficits and sur pluses across banks. If a bank with an unex pected deficit does not have established credit lines with a surplus bank or the surplus banks do not have management authorization to | sell funds to the deficit bank, then a speedy redistribution of cash would be very difficult— perhaps impossible—by transactions among these banks late in the day in the money mar ket. Intermediary banks, having authority to lend to the deficit bank and to whom the surplus banks will lend, would then be required to complete the process that would prevent a wave of systemic failures. The reliability of these market mechanisms would presumably depend importantly on the degree to which confidence in normal credit judgments about deficit banks could overcome a prudent incli nation of surplus banks to seek safety in cash in the midst of waves of market talk occa sioned by an unexpected bank failure. Of course, this is just the setting—a “liquid ity crisis”—in which a central bank lender of last resort might play a constructive role. Systemic risk—of a cascade of settlement fail ures—can be eliminated by isolating the ini tiating settlement failure, preventing it from forcing unexpected cash deficits on other banks, which in turn could force unexpected deficits on still other banks, and so on. The lender of last resort might isolate the prob lem in one of several ways. It might lend to a bank that would otherwise initiate a settle ment failure, or to second-round banks with unexpected deficits produced by the settlement failure, or to banks that would be willing to lend to the second-round banks. However, the central bank may face constraints on its lend ing. For example, it may be prevented from making loans to foreign institutions or to banks without adequate collateral or to in solvent banks. Such constraints would influ ence which set of banks became the focus of last-resort-lending activity in isolating settle ment failure and eliminating systemic fail ures. Beyond these institutional considerations, however, is an important issue to be faced in designing settlement-risk-management pol Federal Reserve Bank of Cleveland icy, involving the degree to which risk expo sure is concentrated on banks in the pay ment system. Concentration of Risk There is a fundamental tension between two means of managing settlement risk in a payment system. One would seek to concen trate risk exposure on banks; the other would seek to distribute exposure more broadly over banks, their customers, and the lender of last resort. Concentrating risk exposure on banks is promoted, as we have seen, when banks guar antee irrevocable availability of funds to their customers. The result should be incentives for banks to monitor carefully the condition of counterparty banks and to manage day light credit extensions to those counterpar ties. At the same time, banks would have incentives to monitor their own market rep utations and to protect those reputations through prudent banking practices, including management of their use of daylight credit. Concentrating risk exposure could also be promoted by the lender of last resort through a “hands-off” attitude toward systemic risk. This would require a credible policy of unwill ingness to be pressured into eleventh-hour lending even to otherwise sound institutions suffering the repercussions of a settlement failure. Knowledge that the lender of last resort would not intervene should create incentives for sound banks to manage care fully their bilateral net credit extensions across all payment networks as well as their net debit positions across networks. Scrupu lous management of these daylight credit extensions would aim at narrowing the size of possible unexpected end-of-day financing needs to levels commensurate with access to cash available in the system to meet them. This management might involve minimizing daylight credit extensions by careful control of the timing of payments relative to receipts, by lengthening the m aturity of bank financ ing to reduce daily payments and receipts, and by increased holdings of cash to act as a buffer stock. Distributing, rather than concentrating, risk exposure can be promoted if banks hold customers liable for funds provisionally cred ited to their accounts but not delivered be cause of settlement failure. Perhaps more important, distributing risk exposure can be promoted if the lender of last resort follows a policy of lending freely to banks in circum stances that might otherwise threaten sys temic settlement failures. There are important differences between concentrating and distributing risk as a means of managing settlement risks in a payment network, although both may be effective in achieving a low incidence of settlement fail ures. Concentration of exposure relies primar ily on market pressures within the banking industry. Banks will wish to conserve their own liquidity by shunning normal reliance on volatile sources of funds, maintaining a good ability to borrow should unexpected settle ment problems arise. This may also serve to economize on their daylight overdrafts by reducing the need to repay overnight funds each day before fresh funds have been received. Liquidity conservation will be encouraged by other banks that will be carefully manag ing daylight credit extensions and the quality of the banks to which such credit might be extended. Distributing exposure relies more heavily on bank customers and supervisory authorities to promote prudent banking prac tices, and ultimately on the intervention of the lender of last resort to prevent settlement failures. Clearly, distributing exposure cre ates a moral hazard, in the sense that a cred ible commitment by the lender of last resort to prevent systemic settlement failures will make redundant many management and mon itoring efforts of banks. This will reduce the liquidity of the banking system and raise chances that imprudent practices will go un Economic Review • Fall 1984 detected and uncorrected. In combination, both effects will increase the probability that the lender of last resort will have to intervene. III. The Current Concern Both banks and the Federal Reserve are con cerned about risk exposure on large-dollar transfer systems. What is not clear is whether this reflects a recent discovery of a longstand ing and stable exposure, or an awareness of recent growth in that exposure. The evidence presented in what follows argues that it is the latter, the awareness of significant growth of settlement-risk exposure, that underlies the current concern. Degrees of settlement-risk exposure cannot be measured. Creditworthiness cannot be modeled by an exact science; the expected value of the unrecovered percentage of a bilateral net credit position necessarily lies in the eye and mind of its beholder. Bilateral and aggregate net credit and debit positions on a payment system during a day may be recorded by individual banks, but have only recently begun to be recorded by operators of the payment systems. Therefore, any his torical evaluation of exposure must rely on inferences drawn from indirect evidence bear ing on creditworthiness, bilateral and aggre gate net credit and debit positions of indi vidual banks, and systemic risk. Individual Bank Risk The dollar volume of daylight credit extended apparently has been growing rapidly since at least 1970. The dollar volume of largedollar payments grew at a 24 percent com pound annual rate from 1970 through 1983, almost 2M> times the rate of growth of the value of national income and output over the same period. All but 3 percentage points of this volume expansion was in the number of transactions rather than in their average value. This was quite different from national income and output, where inflation-adjusted output grew only 3 percent, while the aver age value (price) of output grew 7 percent. In itself, growth in the number and value of transactions does not necessarily require growth in daylight credit extended. It is pos sible that, with increasingly careful control of the timing of payments relative to receipts, banks could have accommodated a larger value of payments with no increase in day light credit—that is, without using larger bilat Fig. 1 Large-Dollar Transfers M ade or Settled through Reserve Deposit Accounts Billions of dollars bUU 1 1 Fedwire f i j j l CHIPS 2 500 --- 1 CashWire 1____ | and CHESS 1 mmmm Reserve deposit H H account balances 400 30 0 20 0 100 0 1970 1983 SOURCE: Board of Governors of the Federal Reserve System. Federal Reserve Bank of Cleveland eral net credit extensions during a day to make payments. Sequencing—delaying payments until re ceipts arrive—is an alternative to daylight overdrafts in covering payments. It is an approximation of a “realtime” accounting system in which payments could only be made if a sufficient balance were on hand at the instant the payment were ordered. Pay ments accounting is not typically “realtime,” but “batched” overnight. Daylight overdrafts arise if payments are ordered before the re ceipts come in that will leave a positive cash balance. In principle, the total stock of cash in the banking system could be as little as a single dollar, as long as that dollar were used and reused over 500 billion times during the day. In practice, with a total stock of deposit balances at Federal Reserve Banks of a little more than $20 billion, the average dollar would have to turn over only about 26 times daily to be used and reused in completing over $500 billion of large-dollar transfers. This is up from only IV times daily as recently as 1970 (see figure 1). Nonetheless, the feat of turning over the stock of cash 26 times daily would indeed be prodigious if each payment were made against an adequate cash balance in “realtime,” so that dollars were actually used and reused 26 times. Sequencing would require that initial payments be made up to the limit of the opening cash balance, but further payments be delayed until sufficient receipts accumulated to fund them. Alterna tively, the feat of 26 times per day turnover is not at all prodigious if daylight overdrafts can be used. Payments far in excess of open ing cash balances could be made with the expectation of covering them with offsetting receipts later in the day before settlement. Increasingly powerful computer and tele communications technology might have made it possible for banks to sequence payments carefully as the volume of transfers grew, 7. The Federal Re serve Bank of New York maintains an on-line monitor of positions of Edge Act customers and restricts their day light overdrafts. Starting in 1984, an ex post monitor was put in place for all Fedwire users. 13 thereby avoiding increased daylight overdrafts. Corporate cash management became a so phisticated science during this period. Indeed, the spread of cash management had as a by-product some of the rapid growth in wire transfer volume. Moreover, computerized telecommunications for wire transfer, inte grated with realtime accounting systems, did make it possible to build into banks’ systems preset limits on daylight credit extensions to customers. This had a counterpart in the “store and forward” environment of the CHIPS system. Banks could store payment messages in the system until such time as they were prepared to authorize a payment, when the payment would be released and the transaction completed. Beyond this, how ever, there is little to suggest the application of sophisticated technology to the problem of sequencing inter-bank payments traffic to enable banks or networks to control their net bilateral or aggregate daylight credit or debit positions. Lack of evidence of attempts to sequence inter-bank wire payments is not surpris ing. Investing resources in managing the sequencing of payments would be likely only if there were some clear incentive to do so. In fact, incentives to manage daylight credit positions have been weak in the large-dollar transfer systems. Fedwire, until recently, had no systemwide mechanism for monitoring daylight credit extended to banks.7 With no restrictions on their daylight overdrafts, efficient manage ment of commercial bank operations had every incentive to use large daylight overdrafts to reduce other costs. For example, large banks routinely finance themselves by buying over night funds from many other banks and fi nancial institutions (aggregating over $100 bil lion daily in mid-1984). Just as routinely, the banks can prepare the necessary list of repay ment messages at the end of a day for auto mated telecommunications transm ittal on the following day. The first job each morning, Economic Review • Fall 1984 before competing uses of people and equip ment intervened, would be to activate the system to transm it the string of repayments. The result would be efficient work flow and a routine drawing down of the bank’s Federal Reserve deposit account balance at the open ing of business each morning. If overnight sources of funds normally exceeded reserve deposit balances (reserve deposits of all banks amounted to about $20 billion in mid-1984), then the result would be an immediate day light overdraft that would last until fresh over night funding had been bought and received. The incentive structure in CHIPS has been somewhat different. Settling banks have had an interest in the amount of daylight credit provisionally extended to their respondent banks participating in CHIPS. Indeed, the “store and forward” mode of operation made it possible to set dollar limits on the net credit extended to a customer by a bank, whether the customer were a large nonfinancial cor poration making trade payments or a respon dent bank paying for money or foreign ex change market purchases. Another indication of settling participants’ interest in control ling risk exposure may be found in the change in CHIPS finality from 1:00 pm of the next day, first to 10:00 am of the next day (in 1979) and then to 6:30 pm of the same day (in 1981). Shortening this time gap had the effect of substantially reducing the duration of settlement-risk exposures. These efforts by settling banks to control their own potential credit-risk exposure to customers might also be viewed as efforts to control settlement risk. Any limitation on credit extended through payments made for a customer is also a limit on daylight indebted ness of the paying bank to other participants in the network. Until recently, however, these incentives were not reflected in any network limit on bilateral credit or debit positions, or in a bank’s ability to set a limit on bilateral or aggregate credit extended on the system. CashWire and CHESS do have both bilateral net credit limits and sender net debit caps, but these cannot be attributed to any incen tives operating within the group of partic ipants, because they are an interim require ment of the Federal Reserve for net settlement of the networks. CashWire also requires cus tomer guarantees. The conclusion that one draws is that the bilateral and aggregate net credit positions generated on large-dollar payment systems have been growing rapidly over the years since at least 1970. Payments volume grew rapidly; the cash position of the banking system using these payment networks did not grow at all; incentives for participating banks to sequence their own transactions to avoid reliance on daylight credit were weak, at best. If risk exposures were not growing, it could only have been because the creditworthi ness of participants was improving enough to offset growing daylight credit usage. The creditworthiness of system partici pants probably was not improving. In addi tion to a general weakening in capital posi tions of the U.S. banking system, two specific developments suggest this conclusion. One has to do with the range of institutions eligi ble for direct access to Fedwire and the second with the burgeoning role of foreign partici pants in the daily flow of inter-bank largedollar payments. Prior to 1980, access to Fedwire was re stricted to member banks, a (declining) sub set of depository institutions that included all national banks plus those state banks that chose to become members. With passage of the Depository Institutions Deregulation and Monetary Control Act of 1980, access was broadened to include all commercial banks, thrift institutions, and credit unions eligible for federal deposit insurance. Without argu ing that any particular newly eligible institu tions were less creditworthy than any mem ber banks, it is still possible to argue that Fed eral Reserve risk exposure increased. More Federal Reserve Bank of Cleveland institutions, supervised by a more diverse set of regulatory authorities, entering new mar kets and (for some) entering the twilight of viable operation during 1981-82, surely could have reduced the average creditworthiness of the set of institutions eligible to use Fed wire. How much Federal Reserve risk expo sure actually increased as a result is impos sible to quantify. Actually, it may not have been a substantial increase. Most of the newly eligible institutions did not avail themselves of access to Fedwire or any of the other largedollar payment systems. A more important increase in risk exposure was the result of growing international inter bank volume. The 1970s first saw U.S. banks follow U.S. multinational corporations abroad and then become immersed in recycling petro dollars through the world banking system. CHIPS dollar volume was growing at a 35 per cent annual rate starting in 1970. Between mid-1978, when current data reporting began, and the end of 1983, U.S. banks’ own claims on foreigners were growing at a 30 percent annual rate; overnight Eurodollar deposit holdings were growing at a 50 percent annual rate starting in 1977. The burgeoning daily flow of payments through CHIPS involving New York-based subsidiaries of foreign banks was a likely source of increased risk exposure. Foreign-owned institutions handling the trans actions of foreign-based banking establish ments subject to unfamiliar legal and regula tory systems necessarily injected uncertainty and new risks into banking relationships. Systemic Risk The conclusion to be drawn must be that, both on account of growing daylight credit exten sions and on account of the widening set of banking institutions whose creditworthiness is relevant, individual risk exposures of par ticipants in large-dollar transfer systems has been growing. Systemic risk exposure is not necessarily driven by these individual expo sures, however. Mechanisms to isolate the impact of a settlement failure might have been strengthened or weakened, offsetting or aug menting the effects of growing daylight credit extensions and the changing creditworthi ness of individual banks. Nonetheless, it would appear that systemic risk has also been grow ing in recent years. Systemic risk is absent from Fedwire because of the finality of Fedwire payments. Since 1970, the percentage of the dollar volume of large-dollar-value payments made on Fedwire has declined from 88 percent to 54 percent, suggesting substantial growth of systemic risk exposure. Moreover, the shift from next-day to same-day CHIPS settlement completed in 1981 might be interpreted as a further in crease in systemic exposure in one important sense. Participants did reduce the duration of their bilateral exposures to settlement risk, and the shift to same-day settlement was de signed for that purpose. However, systemic risk exposure may have been increased because of the shortened time available during which banks, in the event of a settlement failure, could develop alternative financing arrange ments. This would be the case if a participant’s ultimate inability to settle were known or strongly suspected before normal settlement hour, so that the period between the end of a day’s activity and settlement could be used by otherwise sound banks to find alterna tive funds in the cash markets. Whether individual bank participants would have recognized the potentially offsetting in crease in systemic risk exposure when their individual exposures were reduced cannot be determined. Their own systemic risk expo sure would depend as well on the extent to which Federal Reserve Banks’ lending could Economic Review • Fall 1984 be expected to isolate settlement failure and control systemic risk. That systemic risk has increased seems clear. Whether the exposure is that of private financial institutions and their customers or that of the Federal Reserve Banks is not so clear. The distribution could only be deter mined by experience with failures, which does not exist and no one wants, or by a binding commitment by the Federal Reserve about how it would act in the event of a failure, which is fraught with difficulty. A lender of last resort could state its commit ment not to lend in the event of a settlement failure, thereby attempting to concentrate systemic risk exposure squarely on private institutions. That such a commitment could be credible is itself incredible, given the panicavoiding objective of a central bank lender of last resort. Ambiguity might be the most that could be gained from such an approach, cre ating a minimal sense of systemic risk expo sure in private institution managements and some minimal risk-controlling behavior. On the other hand, a stated commitment to isolate all institutions from the impacts of a settle ment failure, while more credible than a pledge of inaction, carries with it the moral hazard that private institutions will have weak incen tives to consider systemic risk in their con trol practices. Recent Proposals Silence about settlement-risk exposure was a reasonable Federal Reserve policy when cash balances equaled or exceeded all of a day’s payments on Fedwire, when Fedwire was the dominant large-dollar transfer system, and when supervisory oversight of member banks provided a monitor of the creditworthiness current interest of bankers and the Federal Reserve in assuring prudent management of settlement-risk exposure by private finan cial institutions and their customers and by the Federal Reserve. Two matters seem central to prudent man agement of settlement-risk exposures. One is the ability of participants to maintain limits on their own bilateral net credit and aggre gate debit positions. This involves both infor mation systems that track those positions and control over their size. Improved positionmonitoring capabilities are being developed by the respective large-dollar payment systems that may provide the necessary information; proposals are being considered that would result in debit and credit limits, either selfimposed or derived from uniform rules. The other matter is the recognition of settlement-risk exposure. Supervisory examina tion may assure that banks have settlementrisk-management procedures in place, but risk unperceived will go unmanaged. Ambiguity about who is at risk in making large-dollar payments clouds this perception. The Federal Reserve is at risk in extending daylight credit on Fedwire and must manage its exposure. In the absence of receiver guarantees, banks IV. Conclusion and their customers are both at risk in mak ing payments on net settlement systems. Settlement risk is real, although actual set tlement failures have been nonexistent. Rec Whether either party, but especially custom ers, clearly perceives the full extent of its ords of bilateral daylight credit positions are beginning to be kept, but evaluating risk exposure is hard to determine. Systemic risk exposure cannot be unambiguously located, exposures will always be judgmental. This but must be managed by the Federal Reserve is because the creditworthiness of “borrow jointly to protect the resilience of the bank ers” is a subjective judgment and because ing system to an unexpected settlement fail the incidence of systemic risk depends on the reaction of the lender of last resort to an ure and the discipline of a grudging lender actual settlement failure. It seems clear, none of last resort. theless, that both individual and systemic settlement risk exposure have increased rap idly in recent years. This has prompted the of all participants in Fedwire. Recently, how ever, risk exposure has grown rapidly, both individual and systemic. Uniform procedures for examination have been adopted by the Fed eral Financial Institution Examination Coun cil that seek to assure, among other things, that financial institution managers are aware of settlement-risk exposure. Beyond that, the Federal Reserve has sought comment on three principal risk-management devices (re ceiver guarantees, bilateral net credit limits, and sender net debit caps) that might be used singly or in combination by network partici pants, by network managements, and/or by regulators seeking to assure prudent set tlement-risk management. The Federal Reserve clearly has a large stake in this. Its own bilateral credit risk has been growing rapidly, unshared with its de positor banks because of Fedwire finality of payment. Its systemic risk exposure has also grown rapidly because its lender-of-last-resort function implies an obligation to isolate the impacts of a settlement failure. A prudent central bank, no less than a prudent private bank, must manage its risk exposure. Federal Reserve Bank of Cleveland Economic advisor Roger H. Hinderliter researches capital investment and bus iness finance for the Federal Reserve Bank of Cleveland. 1. Business earnings may be measured relatively with re spect to the value of total output (capi tal’s share of the proceeds from pro duction and sales); the replacement value (current cost) of the capital stock (the rate of return on capital); or net worth (the rate of return on equity). The rate of return on equity is impor tant only if the divi sion of earnings be tween equity hold ers and debt holders is an issue. Thus, while the rate of re turn on equity may be a critical mea sure for evaluating the performance of an individual firm , it is less useful in evaluating the over all performance of capital assets. The other two ratios are closely linked, and most efforts to track the earnings prob lem have concen trated on them. 17 Sources of Change in Rates of Return on Capital: 1952-82 by Roger H. Hinderliter Economic Review • Fall 1984 Business earnings are the reward to capital from current production and sale of goods and services. As such, they reflect the per formance of existing capital in an uncertain economic environment. Business earnings also are a major influence on investment deci sions of firms and on the willingness of indi viduals and institutions to help finance invest ment by purchasing debt or equity issued by firms. Funds raised by businesses in capital markets are invested in new capital assets in expectation of future earnings from the assets’ productive use. If past expectations were unfulfilled in current earnings, further invest ment would become less attractive, and, in time, output growth, creation of new jobs, and living standards might suffer. Thus, while measuring the performance of existing capi tal assets, business earnings also might be an important indicator of the nation’s future economic health. Beginning in the late 1960s, the economy of the United States was beset by a number of related problems threatening future prosper ity. These problems included falling labor productivity growth, accelerating inflation, sharp increases in energy costs, and an appar ent decline in relative business earnings. Earnings are measured in relative terms as the rate of return on capital or capital’s share of nominal output.1 In studying the perfor mance of corporate business, economists gen erally have concluded that relative earnings declined sharply in the late 1960s and, if not falling further, at least remained low during the 1970s. Declining relative earnings are not unambiguously a signal of weakness and dis appointed expectations. They could reflect, for example, the disappearance in the 1960s of capital scarcity accumulated during the De pression and war years. Although economists agree that the rate of return and the capital share fell, there is little agreement over the sources and implications of the decline. 2. This brief review of past studies high lights fundamental issues. A more tho rough evaluation of the literature is pro vided by Scanlon (1981). 18 In an early attempt to track the earnings performance of corporations, Okun and Perry (1970) attributed the initial decline in the capi tal share, between 1965 and 1969, to a rise in labor cost relative to output price and to poor labor productivity growth. These forces were viewed from the perspective of 1970 as transitory. Somewhat later, Nordhaus (1974) concluded that a longer-term, more perma nent decline in the capital share had occurred. He suggested the decline reflected the effects of higher levels of investment achieved after World War II. This investment was made possible by a lower cost of capital associated with declining corporate tax rates and a falling risk premium on corporate equity as the De pression became a more distant memory. Nordhaus’ conclusion was supported by additional research results. Kopcke (1978) estimated a “normal” rate of return from a productivity model based on a Cobb-Douglas production function and also found that a longer-term decline was indicated. Lovell (1978) examined a number of relative earnings con cepts and, after accounting for cyclical fluc tuations and productivity growth, concluded that a downward trend characterized earn ings patterns. Other studies, however, pro duced contradictory results. Working with gross and net (of depreciation) rates of return on capital, Feldstein and Summers (1977) found lower earnings in the 1970s probably resulted from a variety of factors—wage-price controls, rising inflation, and larger fluctu ations in capacity utilization—none of which would confirm a long-term downward trend. Runyon (1979) compared the approach of Nord haus with that of Feldstein and Summers and also concluded that a secular decline in relative business earnings was not supported.2 In short, the decline of relative business earnings may have been the result of tempo rary reversible disturbances without last ing effect on investment, a sustainable down ward trend possibly constraining investment over a longer period, or a complex mixture Federal Reserve Bank of Cleveland of short-run and long-run factors with equally mixed implications for investment. The strong recovery in investment that began in 1983 is an encouraging sign of an improved invest ment environment. Many industries, even manufacturing industries widely regarded as inefficient high-cost producers, now are mak ing or planning large investments in tech nologically advanced capital assets. A pro longed and successful investment program depends heavily on whether business earnings justify the effort over a longer period of time. Thus, any judgment on the promise of the current recovery depends on unraveling the mysteries surrounding earnings patterns in the past. This study approaches that task in five stages. A method of evaluating earnings perfor mance is proposed in section I. The rate of return on capital and capital’s share of nom inal output are joined together and the result ing expression defining the rate of return is decomposed into a margin effect (which cap tures interactions among the output price, the replacement price of capital, and unit costs) and a multiplier effect (which captures inter actions between capacity utilization and capital productivity). In section II, estimates of before- and after tax net rates of return and their margin and multiplier components are presented for total business (corporate and noncorporate busi nesses). One benefit of a total business frame work is that GNP, the normal unit of account in discussions of economic performance, be comes the relevant measure of output in the calculations. Thus, the estimates can be related to other issues—the productivity slowdown, economic growth potential, and the trans mission of policy actions, for example—that have an economy-wide focus. Moreover, while the noncorporate sector has always been much smaller than the corporate sector, it still is a significant force in creating jobs and con trolling capital. Because most studies of busi 3. Feldstein and Summers (1977) speculated that coun tervailing differences in tax treatment and risk made the noncorporate rate of return indetermi nate relative to the corporate rate in theory. In one em pirical study that included noncorpo rate business (the nonfarm component), Bosworth (1982) re solved the indeter minacy in favor of lower noncorporate rates of return (in creasingly so in the 1970s). However, Bosworth’s estimates of the capital com ponent of total non corporate earnings (capital plus labor) seem to have placed a disproportionate share of cyclical and secular weakness in the 1970s on capital earnings (see below, appendix 1). 4. Holland and Myers (1979) include capital gains (or losses) in estimates of after-tax rates of return. They fin d that individual years can be greatly af fected, but over a longer period the average rate of re turn is little changed by the capital gains calculation. This is to be expected if relative price move ments are such that gains and losses roughly balance overtime. The esti mates by Holland and Myers excluded land, which severely limits capital-gains potential. ness earnings have focused on the larger cor porate sector, it is unclear to what extent noncorporate businesses shared in any earn ings difficulties that emerged in the postwar period. Noncorporate business has undergone a striking metamorphosis over these years. It has been transformed from a largely agrar ian sector to a largely service-producing sec tor, and these structural dynamics are one element affecting rates of return.3 The before- and after-tax rates of return are re-estimated in section III under the counterfactual hypothesis of a constant relative price of capital. Eisner (1980) argued that rel ative price changes during inflationary peri ods create capital gains for owners of existing capital assets, and these are appropriately treated as income. Rates of return measured with respect to replacement value may under state earnings performance, because the rela tive price of capital does change, especially when land is included in the capital stock, and gains typically are not included in income.4 A model for evaluating changes in rates of return that attempts to identify and quantify specific sources of change is proposed in sec tion IV. Apart from relative price increases and cyclical variation in capital utilization, slower capital productivity growth and a one time distortion of the relationship between output price and unit costs in the late 1960s, as first noted by Okun and Perry, contrib uted to the fall in observed rates of return. The extent to which these forces represent a permanent erosion of business earnings is considered in section V. The emphasis in this study is not one of providing better numeri cal estimates of rates of return. Although there are interesting differences between the esti mates developed here covering total business and prior estimates of corporate rates of return, judgments about the significance of these differences are hazardous and necessar ily speculative. The overall pattern of move ments in rates of return found here is quite similar to evidence from past studies. The major departure of this study is the attempt to identify and quantify the sources of change in this pattern of behavior. 19 Economic Review • Fall 1984 I. The Rate of Return on Capital The rate of return on capital is a measure of relative business earnings derived from capi tal budgeting theory. In principle, it compares the discounted present value of the expected future earnings stream to the current cost of capital assets. Thus, the rate of return is a useful guide in choosing among investment projects. Those projects that offer the highest rates of return are implemented, and those projects with relatively low rates of return are discarded. The rate of return also is useful as an indi cator of the performance of existing capital. As such, it represents an average rather than a marginal rate and is defined as cur rent nominal business earnings divided by the replacement value of the capital stock. This measure can be defined before or after taxes, and on a net or gross basis depending on whether a fully depreciated or an undepre ciated capital stock is entered in the denom inator. The choice between net and gross cap ital stocks is governed by the way capital affects output. If capital becomes less efficient in production over time, if it gradually wears out year-by-year, the net concept is appropriate and, correspondingly, earnings are measured net of depreciation or capital consumption. Measurement of the average rate of return on capital rests on a capital maintenance theory of income. Nominal earnings in the numerator are the residual gain from the current year’s production and sale of goods and services, after deducting labor and other costs and providing for the maintenance of capital assets at their replacement value. Nom inal earnings include profits and rents (with appropriate capital consumption and inven tory valuation adjustments) of the corporate and noncorporate business sectors. Noncorpo rate profits are estimated from total noncor porate earnings by assuming the owners of these businesses are paid the average wage in their industry after correcting for the effects of cyclical and secular changes in labor market conditions. The correction (detailed in appen dix 1) is designed to avoid placing the entire burden of cyclical and secular weakness in the economy on capital earnings. In addition to rents that are embedded in profit figures, rental income exclusive of rents on owneroccupied nonfarm housing is included in the numerator of the rate-of-return ratio. Because the division of gains between equity and debt Fig. 1 holders is not an issue in rate-of-return cal culations, net interest payments are added to nominal earnings.5 The nominal earnings stream —profits plus interest plus rent—is a broad-based measure of the returns to business capital. The denomi nator of the rate of return ratio accordingly is defined to include the broad set of capital assets generating this stream. The relevant Rates of R eturn on C apital Nominal output Less claims on output Equals business earnings Business Earnings Labor Compensation Compensation of employees and noncorporate business owners \ O ther Business Paym ents = Before-tax earnings Indirect business taxes and trans fers, less govern- Less business income taxes m ent subsidies V Corporate profits and capital earn ings of noncorpo rate businesses, rent (excluding overhead), and net interest (ex cluding overhead) ............. > = Aftertax earnings Capital Costs Capital consumption on repro ducible business capital X—.............' Overhead Costs N. Labor compensa tion and other charges in govern ment, households, and nonprofit institutions \ Federal Reserve Bank of Cleveland ' s . „ ,, Divided by value of capital Value of Capital Current cost of reproducible capital (plant, equipment, and rental housing), inventory stocks, and land „ Equals rates of return on capital 5. See, for example, Walton (1981). Lovell (1978) sug gests that including interest also may help correct for the financial distortions of inflation, which otherwise tend to understate earnings by raising interest charges without a corresponding reduc tion in the real value of outstanding debt. 21 capital stock is the sum of business invento ries, plant and equipment, tenant-occupied res idential housing, and land used for business purposes. Among these components, only land presents difficult measurement problems. In past studies of business earnings, a variety of simplifying assumptions have been em ployed to deal with these problems, among them ignoring land altogether. Here, the basic land series for rate of return calculations are the market value estimates from the flow-offunds balance sheets. Real value of land stocks and the corresponding price indexes, which are needed for margin and multiplier calcula tions, are estimated from the sketchy data available on agricultural land, land transi tion ratios, and land ownership (see appen dix 2). The resulting estimates suggest a nearly constant real value of business land, produced by a reduction of farm land at about 0.5 percent per year and an increase in non farm business land at about the same rate be tween 1952 and 1982. Land prices increased rapidly, especially in the 1970s. A schematic view of the derivation of rates of return is shown in figure 1. Starting from nominal GNP, labor and other superior claims are deducted, leaving business earnings as the residual claim. The ratio of this earnings stream (before or after taxes) to the value of the capital stock determines rates of return. Formally, the definition of the before-tax rate of return can be stated as follows: Ei ( 1) Rt =-7T Vt CPFTt + NCPFTt + NINTt + RENTt KPRt (INVt + FlXt + LANDt) Economic Review • Fall 1984 where E = nominal business earnings— sum of corporate profits, non corporate profits, net interest, and rent, V = replacement value of capital— (weighted average) capital price index times sum of real values of inventories, fixed reproducible capital, and land. The alternative measure of relative busi ness earnings examined in the studies cited above is defined as the earnings stream divided by nominal output or GNR This is the share of nominal output generated dur ing the year claimed by the business owners of capital. It is related to the rate of return on capital in the following manner: Et GNPt A X (2 ) Vt Vt GNPt The rate of return is equal to the capital share times the ratio of nominal output to the value of the capital stock. If the components of equation (2) are rearranged slightly, the rate of return becomes: x — (3) A . vt Qt v, — / QPRi - UCS,\ = ( KPRl ) x (QQt * OK). where QPR, KPR = price indexes of output (real GNP) and real cap ital stock, respectively, UCS = total unit cost of pro ducing output, QQ = ratio of actual output to potential output, QK = ratio of potential output to real capital stock. 6 . The QK ratio is a hybrid measure in that it relates poten tial output to actual capital. Perhaps a more appropriate measure of long-term capital productivity would be potential output divided by potential (cyclically adjusted) capital. This would require a separate estimate of potential capi tal, one coyisistent with the chosen mea sure of potential out put. Deviation of actual capital from potential capital then would become a third component of the earnings multiplier. No significant cycli cal variation in the QK ratio was de tected, however, and this refinement was not made. 7. The major source of data used in this study is the national income and product accounts. The ac counts provide esti mates of earnings, prices, costs, out put, and inventory stocks. Fixed repro ducible capital stocks are from the Com merce Department’s tangible wealth esti mates. Land stocks and prices are esti mated from several underlying data sources, as described in appendix2. Poten tial output is from Clark (1983). All data are mid-year values. Capital-stock data reported endof-year are shifted to mid-year by simple averaging. 22 The first component on the right-hand side of the final expression in equation (3) is the earnings margin. This margin includes nomi nal variables relating the spread between out put price and unit costs to the replacement price of capital. Thus, the price of capital rel ative to the price of output is an important determinant of the margin effect on the rate of return. Unit costs are divided into four groups (as identified in figure 1) for before tax estimates of the rate of return. These are labor compensation, other business payments, capital costs, and overhead costs. Labor com pensation is the payment for labor services in the corporate and noncorporate business sec tors. Other business payments include indi rect business taxes and business transfer payments. Capital costs are the depreciation charges (capital consumption with capital consumption adjustment) on the fixed repro ducible portion of the business capital stock. Finally, overhead costs include compensation paid in government, households, and nonprofit institutions, and rents, interest, and other charges on owner-occupied residential hous ing and the fixed capital of nonprofit institu tions. Government, households, and nonprofit institutions produce no earnings, but they perform functions necessary for the ongoing activities of earnings centers. Government provides the legal system, national defense, and rules and regulations for society’s organ ization and well-being. The household is the fundamental organizational unit of society, performing many functions that support the earnings centers. Nonprofit institutions like wise perform a variety of needed functions definitionally outside the domain of earnings centers. Hence, they are treated as the over head component of the national economy. Bus iness income taxes, of course, are deducted along with the above costs to calculate the after-tax rate of return. The second component on the right-hand side of equation (3) is the earnings multiplier. The earnings multiplier includes real vari Federal Reserve Bank of Cleveland ables, measuring capacity utilization and cap ital productivity. These components, which together measure the efficiency of the capital stock in producing output, essentially deter mine how a given earnings margin will be transformed into a rate of return. The ratio of actual output to potential output (the QQ ratio) captures the effects of capacity util ization on the rate of return. High capacity utilization (larger values of the QQ ratio) sup ports a higher rate of return by stretching the margin over a more efficient volume of out put. The ratio of potential output to the real capital stock (the QK ratio) is a measure of capital productivity. Higher capital productiv ity boosts earnings by squeezing larger poten tial (full-capacity) volumes out of a given capital stock? II. Tracking the Rate of Return Before-tax (BXR) and after-tax (AXR) aver age net rates of return on capital, calculated from the definition of equation (1) for the total business sector, are plotted in figure 27 The estimates cover the period from 1952, when any distortions from World War II and its aftermath might be expected to have dissi pated, to 1982, the latest year for which com plete and compatible earnings and capital stock data were available. This is a period of rich economic experience. It includes episodes of stable economic growth and episodes of accel erating inflation; structural changes in out put and business organization and cyclical swings of major proportions; wage-price con trols in the early 1970s and periodic tax changes, the most sweeping of which came in 1981. In the early part of the period, both before tax and after-tax estimates were relatively stable. BXR declined slightly, and AXR changed even less, except for cyclical fluctu 8 . Kopcke (1978) and other researchers argue that after tax estimates of the rate of return are more revealing measures of performance. Be cause taxes are levied on reported earnings rather than economic earnings (with cap ital consumption and inventory valuation adjustments), effective tax rates increase during a period of accelerating infla tion, possibly distort ing before-tax esti mates relative to their after-tax counter parts. When interest was included with earnings, however, AXR and BXR gen erally moved together, and A XR fell only marginally faster than BXR between 1965 and 1978. ations produced by recessions in 1954, 1958, and 1960. In the early 1960s both measures rose sharply, reaching 30-year highs of 11.2 per cent (BXR) and 8.0 percent (AXR) in 1965. After 1965, a sustained erosion in both mea sures is clearly indicated. Cycle peaks in 1973, 1978, and 1981 were progressively lower than previous peaks. In 1981 BXR was 7.5 per cent, about two-thirds of its mid-1960s record high, while AXR had declined to 5.8 percent. In the recession year of 1982, BXR fell to a historic low of 6.5 percent. The after-tax mea sure recorded its record low (5.0 percent) in 1980, prior to the tax legislation of 1981. In deed, AXR increased relative to BXR prior to 1965 and after 1978, especially in 1982 when the effects of the Economic Recovery Tax Act of 1981 became significant.8 Comparison of the rates of return shown in figure 2 with the results of other studies is hazardous because of substantial differences in methods of variable estimation. With this caveat in mind, it can be noted that the pat terns of change in BXR and AXR are similar to those found by other researchers. The bulge Fig. 2 Percent 14 - Rates of R eturn on Capital 12 10 8 6 4 2 0 1952 23 1962 -------AXR 1972 --------BXR 1982 in the mid-1960s and the decline thereafter are pieces of the earnings puzzle common to virtually all estimates of the rate of return. It also appears that the decline is somewhat less pronounced (at least through the late 1970s) in the estimates presented here. For example, Feldstein and Summer’s (1977) before-tax estimates of the net corporate rate of return are higher than BXR throughout the 1950s and 1960s, by 2.4 percentage points near the peak (1964-66). This gap diminishes in the 1970s, and, in several years, particu larly 1973-75, BXR exceeds Feldstein and Summer’s corporate rate. Although these comparisons are highly provisional, it does not appear that a general deterioration of noncorporate rates of return relative to cor porate rates in the 1970s, noted by Bosworth (1982), is supported here. Noncorporate business was in sharp tran sition throughout the postwar period, a tran sition with roots reaching back into the De pression years of the 1930s. In 1952,39 percent of all noncorporate entrepreneurs operated agricultural businesses, and 19 percent were in the service and finance industries. By 1982, the composition was reversed. Over the same period, farm earnings (labor plus capital) fell from about one-third to about one-fifth of total noncorporate earnings. Thus, by the 1970s the marginal farm businesses in the noncor porate sector, probably characterized by low rates of return on capital, were substantially reduced in number and influence on earnings, and services had supplanted agriculture as the dominant force in the noncorporate busi ness structure. Structural transition in the noncorpor ate sector probably contributed stability to capital earnings in the sector. Certainly, non corporate businesses performed less well than corporations in the 1950s and 1960s, because their capital earnings did not match the corporate pace. In the early 1950s (1952-54), noncorporate business profits were about 43 percent of corporate profits. By the late 1960s (1967-69), the profit proportion had de clined to 24 percent, but it fell only slightly 9. I f no adjust ment for labor mar ket conditions had been made (that is, if noncorporate bus iness owners were paid the average wage without quali fication), the profit proportion in the late 1970s would have been about 12 percent rather than 2 2 percent, suggesting a deterio ration as noted by Bos it’o rth (1982). Noncorporate prof its and rates of return in 1980-82 then would have been negative. 10 . See, for example, Eisner (1980) and Cagan and Lipsey (1978). in the 1970s, to 22 percent by the end of the decade (1977-79). More stable profits coupled with traditionally slower growth of capital (at replacement value) would have supported noncorporate rates of return in the 1970s. In 1980-82, however, the noncorporate sector collapsed. Even with considerable allowance being made to insulate capital earnings from slack labor market conditions, the profit proportion fell sharply in these years, to about 9 percent; hence, the extremely low rates of return of this period apparently were concen trated in the noncorporate sector.9 Following the definition of equation (3), the rate of return on capital can be decomposed into the margin effect, which accounts for the influence of prices and unit costs, and the multiplier effect, which accounts for the influ ence of capacity utilization and capital pro ductivity. Before-tax (.BXM) and after-tax (.A XM) earnings margins are shown in figure 3. The earnings multiplier {MULT), cycling about the productivity ratio (QK) as capacity utilization rises and falls, is shown in figure 4. There can be little doubt that a long-term margin squeeze was a substantial contributor to falling rates of return. Even in 1965, the record year for rates of return, the before-tax margin fell short of previous peaks, and sub sequent peaks continued to exhibit deteri oration. The after-tax margin reached an all-time high in 1965, but steadily declined thereafter. The sharpest declines in both margins occurred in the late 1960s (1965-70). Contributions to rates of return from the earnings multiplier can be separated into three periods. The first, from 1952 to 1964, was a period of rapid growth in the multiplier, produced by sharply rising capital productiv ity. Although relatively low capacity utili zation limited the effect on rates of return in many of these years, the rapid growth offset much of the erosion of margins and paved the way for record rates of return in the mid1960s. In the late 1960s the multiplier plateaued at a high level. Capacity utilization was quite high, but capital productivity was not growing. The multiplier in the 1970s was characterized by irregular growth in capital Federal Reserve Bank of Cleveland productivity and large cyclical swings in capacity utilization. III. Constant Relative Price Estimates The persistent decline in both before- and after-tax margins, which would virtually fore close all doubt about trend movements in rates of return were it not for the partially offset ting increases in the mutliplier, raises an important question about rate-of-return mea surement. Rates of return are based on replace ment value of the capital stock, and earnings margins correspondingly are measured rela tive to the replacement price of capital. If the capital price increases faster than the output price, such that the relative price of capital increases, should account be taken of the capital gains accruing to the owners of capital assets? Two types of capital gains can be identi fied. Nominal capital gains on inventories and fixed reproducible capital are removed from earnings by the inventory valuation and capi tal consumption adjustments. These adjust ments are necessary because otherwise earn ings would be distorted by amounts properly reflecting current economic costs of capital maintenance—replacement of inventories, plant, and equipment. During inflationary periods both the inventory valuation adjust ment and the capital consumption adjust ment lower reported earnings, often by sizable amounts. In 1979-81, for example, the peak of the last inflationary spiral, the two adjust ments lowered business earnings by an aver age of $67 billion a year. This produced a before-tax rate of return that averaged 1.5 per centage points less than would have been estimated without the adjustments. Price increases need not be and usually are not equally distributed among all price indexes. Although nominal gains are excluded from earnings, many researchers contend that rel ative price changes create real capital gains, which augment earnings.10Thus, if the replace ment price of capital rises faster than the output price, the relative price change creates a real capital gain for the business owners Fig. 3 E arnings M argins Percent 1952 Fig. 4 1962 ------- AXM 1972 ------- BXM E arnings M ultiplier Ratio 25 Economic Review • Fall 1984 1982 of capital assets. This gain simply reflects the fact that existing capital has risen in value, and because the owners of capital have use of higher-valued assets (at no additional cost), allowance should be made in earnings. Con versely, a decline in the relative price of capital results in a capital loss and a reduction in earnings. With no accounting for real capital gains, rates of return and earnings margins almost certainly would exhibit a downward trend during periods of prolonged acceler ating inflation such as the 1970s. Replacement prices of fixed reproducible capital, and espe cially land prices, tend to outpace the gen eral price level under such conditions. If rel ative price movements are the major factor depressing estimated rates of return through the margin effect, it is at least arguable that no meaningful decline occurred. A number of difficult problems arise in estimating real capital gains. Usually, no sale of capital assets takes place so a true picture of capital gains is unavailable from market data. Eisner (1980) develops a revaluation ap proach to estimating real capital gains. The carryover capital stock (that portion of the stock in use from one period to the next) is valued first at actual replacement prices and then at a simulated price that is limited to the increase in the general price level. The difference between the first and second of these formulations is an estimate of real capital gains. Holland and Myers (1979) esti mate corporate rates of return that include real capital gains from revaluations of repro ducible capital stocks. These gains raise rates of return in the early 1950s (increasing the after-tax rate of return by an average of 2.5 percentage points between 1951 and 1956), but have less effect thereafter. If revaluations were included in the rate of return estimates computed here for the total business sector, both before-tax and after-tax estimates would be higher in the 1970s (through 1979) by an average of 2.5 percentage points. The larger gains in the 1970s found here, compared with Holland and Myers’ estimates (through 1976), reflect the importance of land in computing capital gains. In the 1970s, when gains were accruing to both land and repro ducible capital, the gains on land accounted for about 72 percent of the total. However, the movements in the relative price of reproduc ible capital imply substantially smaller gains in the 1950s, and substantially larger gains Fig. 5 Constant-Relative-Price Rates of R etu rn on C apital P ercent 15 10 5 0 1952 1962 ------- AXRC 1972 1982 ------- BXRC Table 1 C apital Productivity Grow th Period (potential output per unit of capital) Potential output Capital stock 1952-64 1.88 (0.53) 3.61 (0.10) 1.70 (0.47) 1965-70 0.01 (0.29) 0.95 (0.71) 3.44 (0.09) 3.48 (0.10) 3.43 (0.29) 2.51 (0.71) Average rate in percent per annum QK 1971-82 NOTE: Standard deviations are in parentheses. Federal Reserve Bank of Cleveland in the 1970s, than those estimated by Holland and Myers. Apart from differences in the rel evant business sectors, and therefore in the measurement of the variables used to esti mate capital gains, differences between out put prices used in revaluation and methods of estimating gains on inventory stocks seem to account for these results. Eisner’s (1980) net revaluations of reproducible capital were generally larger in the 1970s (through 1977) than those estimated here. Scanlon (1980) points out that an alterna tive approach to revaluation is to assume that capital and output prices move together, thus entirely submerging the capital gains ques tion. This assumption will not do, of course, as long as the relative price of capital clearly is not constant. It may be made operational, as a counterfactual experiment, by limiting capital price increases to increases in the output price, thus creating a synthetic capi tal price that remains constant relative to the output price. Variation in rates of return and earnings margins that remain must result from forces other than changes in relative prices. Before-tax (BXRC) and after-tax (AXRC) rates of return on capital estimated with a constant relative price of capital are shown in figure 5. As expected, constant-relative-price rates of return are higher than the variable price estimates for most of the period 1952-82. The gap widens in the 1970s, indicating that failure to take account of relative price changes and potential capital gains that accompany them is an important force behind the appar ent decline in rates of return after 1965. Wider cycles are evident in BXRC in the 1970s, but the level is about the same as the 1950s. Indeed, the bulge in rates of return during the 1960s stands out in these estimates. The after-tax measure, AXRC, rises in the 1970s despite greater cyclical variations. 11. See, for example, Nordhaus (1974) and Kopcke (1978). 12. The causal fac tors in the capital productivity decline are not known with any precision. Bern stein (1980) suggests “w illful'’and “in escapable” errors in capital-stock m an agement (associated with increasing atten tion given to shortrun profit perfor mance in the first instance, and energy price increases and pollution abatement requirements in the second) produced . the wrong stock in the wrong place at the wrong time.” Baxter (1978) fu r ther speculates that falling research and development expenditures and heightened empha sis on replacement investment may have aided the de cline. No single fac tor seems capable of accounting for the tim ing of the sharp decline in 1965 (prior to energy problems and m an dated capital expen ditures) and the partial reversal of capital productivity growth after 1970. This question, part of the wider mystery surrounding the be havior of produc tivity, is left to fu ture research. 13. For a description of spline functions, see Poirier (1976). 27 IV. Sources of Change in Rates of Return As outlined above, estimating the pattern of long-term change in the rate of return on capital generally has been approached by searching for statistically significant trends in models that also standardize for cyclical variation. Results from these efforts have been mixed and have revealed little about under lying sources of change. The trend coefficients have had no clear interpretation, and sorting out different short-run trends within a longer period often has been an arbitrary process. Advancing beyond simple trend/cycle models depends on an ability to assign trend effects in some systematic way to the margin and multiplier components of the rate of return. This, in turn, depends on how output price is determined as a markup over unit costs, which governs long-term change in the margin, and on the capital-output relationships that char acterize long-term growth paths. Markup rules and capital-output relationships can be mod eled explicitly with considerable rigor.11 In this study, where isolating the sources of change in rates of return is of paramount interest, an approach retaining the simplicity of trend/ cycle models, though enhancing interpretability, is followed. The behavior of the QK ratio shown in figure 4 suggests three different regimes of capital productivity growth: 1952-64,1965-70, and 1971-82. These regimes are identified in table 1. It is evident that slower capital pro ductivity growth in the periods 1965-70 and 1971-82 is associated with faster growth in capital stock, the denominator of the pro ductivity ratio, in those periods compared with the 1950s and early 1960s. Potential out put grew at about the same rate in all three periods, despite the acceleration in capitalstock growth. Other researchers, using dif ferent methodologies, have found broadly similar behavior in the relationship between output and capital. Bosworth (1982) estimated a capital output ratio adjusted for cyclical change, and found it rising in the late 1960s Economic Review • Fall 1984 and 1970s (implying the inverse, comparable with the QK ratio, fell). Bernstein (1983) noted a flat capital-output ratio after 1965 for nonfinancial corporations, while Baxter (1978) measured a declining output-equipment ratio (without adjustment for the cycle) in manu facturing from 1965 on. The major re s u ltfaster capital-stock growth after 1965 did not produce commensurate increases in potential output—is the same, which had to depress rates of return on capital.12 If changes in the rate of growth of capital productivity identify periods of different underlying trend growth, rates of return can be estimated from a linear spline model (see table 2).13 Trend coefficients in the model are interpreted as long-term growth rates, depending on the rate of capital productivity growth in each period, and the margin drift that characterizes the behavior of earnings margins in each period (see appendix 3). Mar gin drift accounts for long-term growth in the earnings margin as determined by the price markup rule that best approximates pricing strategy for the aggregate economy. In gen eral, margin drift is expected to be greater than or equal to zero. Thus, estimated trend coefficients are expected to be greater than or equal to observed capital productivity growth in each period. Estimates of trend coefficients for all rate of return concepts employed here are shown in table 2. The coefficients displayed are max imum likelihood estimates with first- and second-order autocorrelation corrections. The model appears to fit the data veiy well. All individual coefficient estimates are signif icantly different from zero at the 1 percent level. Estimates of /S4 suggest considerable cyc lical sensitivity in rates of return. The elas tic response corresponds to the direct effect from changing output volume (unitary elasti city) and the indirect effect of the cycle on prices and costs in the earnings margin (elas tic). An increase in the QQ ratio from reces sion levels (95 percent) to boom levels (105 per cent) increases rates of return from 30 per cent (AXR) to 33 percent (BXRC). 14. I f margin drift were approximately constant (not neces sarily zero), the sum of the trend shift coefficients (fc, @3) should not differ significantly from the differential in capital produc tivity growth rates between 1952-64 and 1971-82 (appropri ately signed). Based on restricted regres sion V-tests, the dif ference is statistically significant (at 5 per cent) in variablerelative-price equa tions, but not in constant-relativeprice equations. The major points of interest are the trend, coefficients (J3lt 02>P ), and their sums, which are estimates of underlying growth in rates of return for the periods 1952-64, 1965-70, and 1971-82. In 1952-64, the estimates of pl imply rate-of-return growth ranged from about 1.6 percent a year (BXR) to about 3.5 percent a year (AXRC). The sharp downward shift (02) in the late 1960s produced significantly negative growth rates (/^ + 02) in all mea sures of the rate of return in 1965-70. The shift was reversed (03) in the 1970s, but in 1971-82 only constant-relative-price rates of return grew at a pace significantly different from zero (/^ + 02 + 03). The trend coefficients quantify the com bined effects of capital productivity growth and margin drift on rates of return. Although the trend shifts were controlled by observed changes in productivity growth, coefficient estimates also reveal the importance of margin drift in determining the course of rates of return. Quite different growth patterns are implied by the coefficients estimated from variable-relative-price and constant-relativeprice measures of the rate of return. To disen 3 Table 2 M ax im um Likelihood Estimates of Rate of Return G row th From the process described in appendix 3, the rate of return on capital over three regimes of capital productivity growth may be expressed as follows: R, R 0TT(l nYQQU, where t = time, r = rM+ (sum of growth rates of earn ings margin and capital productiv ity ratio), i = index denoting three periods of capital productivity growth. The parameters of this equation—R0, rit and A—are estimated from a linear spline function: = + 1=1 Federal Reserve Bank of Cleveland tangle the productivity and drift components, the estimated trend coefficients can be com pared with mean values of productivity growth rates (table 1). That is, the trend coefficient estimates can be tested against the hypothe sis of zero margin drift. The rate of growth of BXR was significantly lower than productivity growth alone would suggest in each of the three periods, imply ing negative margin drift over the entire 30 years. Setting aside 1965-70 for the moment, the appearance of negative drift in these esti mates reflects only the increasing relative price of capital in both 1952-64 and 1971-82, with a steeper increase, of course, in the latter period. When holding the relative price con stant (BXRC), estimated growth rates in 1952-64 and 1971-82 were significantly greater than could be accounted for by productivity growth alone, indicating positive margin drift and a pricing strategy that does not exclude rapidly rising unit costs from the markup func tion. However, the difference between growth rates in 1952-64 and 1971-82 is accounted for by the capital productivity growth differ ential between the two periods. Thus, margin Ini?, = fio + + fcTS + P3TSS + 04ln QQt + n, where T = linear trend (T = 1, 2 ,..., 31), TS = shift in trend at 1965 (TS = T-12), TSS = shift in trend at 1971 (TSS = T-18), n = random error, and 00 = \nR0, 01 = ln(l + ri), 02 = shift in 1 + r at 1965, 01-< 02 “ ln(l + r2), 03 = shift in 1 + rat 1971, 01 + 02 HK03 “ ln(l + r3), 04 = X. 15. The early stages of the labor produc tivity slowdown ac counted for only a fraction of the in crease in unit labor compensation rela tive to output price in 1965-70 (Okun and Perry 1970). Rela tive increases in the compensation rate, perhaps induced by the long duration of high capacity utili zation, accounted for the remainder. Wage -p rice co ntrols, followed by deep re cessions in 1973-75 and 1981-82, could have curtailed rela tive increases in the compensation rate and partially offset even greater deteri oration of labor prod uctivity in 1971-82. drift in BXRC was positive, which is accept able from the standpoint of pricing strategy, and also was approximately constant from 1952-64 to 1971-82.14Similar results are implied by the after-tax trend coefficients. Estimated growth rates of after-tax rates of return were higher than their before-tax counterparts in both 1952-64 and 1971-82, although the differential narrowed slightly in the 1970s. Declining tax costs in the 1950s at least kept AXR growing apace of BXRC, but this cor respondence was broken in the 1970s. The sharp downward shift and the result ing negative growth rates in all measures of the rate of return in 1965-70 cannot be explained by a combination of changing cap ital productivity growth and constant margin drift. Although capital productivity growth was essentially zero, it did not become neg ative, on average, for the period. Yet even constant-relative-price rates of return exhib ited substantial negative growth rates, sug gesting a restructuring of the spreads between Rate of return Coefficient BXR BXRC AXR AXRC 0o -2.4445* (0.0136) -2.4643* (0.0202) -2.9279* (0.0235) -2.9464* (0.0185) 01 0.0158* (0.0016) 0.0255* (0.0024) 0.0255* (0.0028) 0.0352* (0.0022) 02 -0.0602* (0.0045) -0.0731* (0.0066) -0.0735* (0.0078) -0.0873* (0.0061) -0.0445* (0.0032) -0.0480* (0.0047) -0.0480* (0.0056) -0.0525* (0.0044) 03 0.0433* (0.0050) 0.0685* (0.0073) 0.0540* (0.0087) 0.0810* (0.0068) 01 + 02 + 03 -0.0012 (0.0021) 0.0205* (0.0029) 0.0060 (0.0036) 0.0285* (0.0029) 04 2.7749* (0.1917) 2.8292* (0.2539) 2.6137* (0.3245) 2.7753* (0.2573) + 02 R2 0.995 0.898 0.986 0.989 Pa -0.586 -0.123 -0.818 -0.636 SE 0.032 0.037 0.050 0.045 NOTE: Standard errors are in parentheses. * = Significant at 0.01. a. p = Sum of first- and second-order autocorrelation coefficients. 29 Economic Review • Fall 1984 output price and unit costs depressed earnings. The contribution of price and unit costs to changes in earnings spreads is illustrated in table 3. The average annual percentage change in price is shown in row 1. Share-weighted changes in unit costs and earnings spreads then represent the claims on the price increase. Clearly, the relationships between price and unit costs have not been constant in the post war period. Price increases averaged 1.93 per cent in 1952-64 and accelerated to 7.06 per cent in 1971-82. Claims represented by other business payments (UBC) and overhead costs (UOC) declined, while capital costs (UKC) increased. These changes were relatively steady transformations. Unit labor compen sation (ULC), however, rose sharply from 39 percent of the price increase in 1952-64, to 53.7 percent in 1965-70, before falling back to 47.2 percent of the price increase in 1971-82. Given changes in other unit costs, the reduc tion in labor’s claim on price increases in 1971-82 was sufficient to restore the before tax spread to the relative position in 1952-64. (About 12.8 percent of the average price in crease went into the before-tax earnings spread in each period.) After-tax spreads were rela tively lower in 1971-82 because unit-tax-cost reductions of the 1950s were not duplicated in the 1970s. Thus, the late 1960s was a period when accelerating labor costs were not marked up sufficiently to maintain a spread suffi cient to cover other costs and provide earn ings as well.15 V. Summing Up Rates of return on capital from the early 1950s through the early 1980s were influenced by four sets of factors: changes in capital pro ductivity growth; changes in the relative price of capital; changes in margin drift associ ated with disturbances in the price-markup function; and changes in capacity utilization. An idea of the relative importance of these factors can be obtained by asking what a particular rate of return might have been in 1982 had it grown from its initial value along alternative growth paths where the separate effects are controlled. This experiment is illustrated in table 4 for the before-tax rate of return. The first entry projects the rate of return along a driftless and cycle-free path defined by average capital productivity growth in 1952-64 extended through 1982. The sec ond entry allows average productivity growth to change as observed in the three sub-periods Table 3 considered in the analysis, while maintain ing the zero drift and cycle assumptions of the first entry. In the third entry, margin drift is added but the relative price of capital is held constant. This incorporates the legacy of the markup failure in the late 1960s. The fourth entry allows for a variable relative price and thus illustrates the effects of infla tion-induced increases in the relative price of capital on the rate of return. Finally, in the fifth entry, the estimate of BXR in 1982 picks up the cyclical and random fluctuations of that year. Price, U n it Cost, and E arnings Spreads Spread component 1965-70 1971-82 Percent of Rate of increase, percent per year price increase Rate of increase, Percent of percent per year price increase — 39.0 9.9 6.5 31.8 12.7 3.87 2.08 0.35 0.35 1.15 -0.06 0.02 -0.08 p 1.93 0.75 0.19 0.13 0.61 0.25 -0.02 0.27 t— 1H QPR ULC UBC UKC UOC BXSPRD UXC AXSPRD 1952-64 Percent of Rate of increase, percent per year price increase 13.7 — 53.7 8.9 9.1 29.6 -1.5 0.4 -1.9 7.06 3.33 0.46 0.80 1.57 0.90 0.12 0.78 — 47.2 6.5 11.3 22.3 12.8 1.7 11.0 NOTE: Unit costs and earnings spreads are share-weighted average increases, where weights are the ratio of the individual cost or spread com ponent to price. Table 4 Before-tax Rates of R eturn in 1982 For alternative growth assumptions Growth path 1. Defined by average capital productivity growth 1952-64, extended through 1982 2. Defined by average capital productivity growth 1952-64, 1965-70, and 1971-82 3. Defined by growth coefficients in column 2, table 2 4. Defined by growth coefficients in column 1, table 2 5. Before-tax rate, 1982 Rate of return, percent Federal Reserve Bank of Cleveland 15.5 12.4 11.3 8.0 6.5 It would be stretching this example be yond its bounds to claim that a 15.5 percent before-tax rate of return on capital was an attainable goal in 1982. Nevertheless, the rates illustrated in table 4 indicate the nature and importance of the changes that distinguished later years from earlier years in the postwar period. The difference between the rate of return in line 1 and line 5 is -9 percentage points, split about equally between margin effects and multiplier effects. The smaller mar gin and multiplier effects, dealing with the B H H H H B B H B H H B H IH H a n a M B legacy of an already reserved change in price-cost relationships and cyclical fluctu ations in capacity utilization, clearly are tran sitory. Of the larger effects, the margin com ponent reflects a permanent erosion in the rate of return only if the inflationary expe rience of the 1970s is a permanent feature of the U.S. economic environment. Even under inflationary conditions, if capital gains on the existing capital stock are counted in business earnings, the erosion is offset when these gains are valued on the same terms as cash earnings. In a disinflationary environment dif ferences associated with changes in the rela tive price of capital are narrowed. Declining capital productivity is more difficult to explain. Because the decline in productivity was asso ciated with an acceleration in capital stock growth after 1965, achieving capital suffi ciency or related explanations seem doubtful. The puzzle to solve here is one of determining how the U.S. economy managed to wring so much output growth out of so little capital stock growth in the 1950s and early 1960s. is the relevant one. If owners of noncorporate businesses are to be paid the average wage, they are compensated partly as managers, partly as skilled workers, and partly as un skilled workers, depending on the character istics of the work force embodied in each indus try’s average wage. To extract the labor component from total noncorporate earnings, some account must be taken of cyclical and secular changes affect ing labor markets. In the corporate sector workers are laid off in periods of slack de mand to reduce or limit the increase of the wage bill. Wage rates, however, are affected less by falling demand and usually continue to rise during recessions. Controlling labor costs through layoffs distributes the bur den of recession to labor income as well as capital income. Noncorporate businesses have similar con trol of their hired labor force, but not of their own amalgamated labor services. Owners of noncorporate businesses do not lose their jobs during recessions, except those whose firms fail. Consequently, paying the owners the average wage during recessions would mean Appendix 1 Estimating the that cyclical weakness in the economy falls Capital Component of most heavily on their capital earnings. To Noncorporate Earnings avoid this incidence, owners of noncorporate must take a cut in labor income In the national income and product accounts, enterprises when economic activity and labor market noncorporate earnings are not divided into conditions weaken. On the other hand, they labor and capital components. The entry in may be paid an “entrepreneurial bonus” dur the accounts for proprietors’ income is simply ing boom periods when labor markets are the owners’ total earnings and therefore in tight and labor services in short supply. cludes both a return for their labor services A second problem encountered in estim at and a return on the capital invested in the ing division of total noncorporate earnings businesses. To estimate the labor-capital divi intothe labor and capital components is asso sion of earnings, the noncorporate business ciated with secular changes in labor markets. sector could be given the same return on capi The natural rate of unemployment, or the non tal as corporations, thus estimating labor accelerating inflation rate of unemployment, compensation as a residual. Alternatively, has risen since the 1950s, and most of the owners of the noncorporate businesses could increase occurred in the late 1960s and 1970s. be paid the average wage prevailing in their industry, thus estimating business earnings on capital as a residual. Here, the second option Economic Review • Fall 1984 a. See Phillips (1962). It appears that one force behind the turnabout was growing num.bers of women and the young entering the noncorporate sector, two groups whose employment opportu nities might be more severely curtailed as the natural rate rises (Fain 1980). 32 One estimate, constructed by Clark (1983, table 5) for modeling potential GNP, shows this underlying unemployment rate rising from 4.5 percent in 1954 to 7 percent in 1978, before easing slightly in the early 1980s. As long-term labor market conditions deterio rated, the number of full-time partners and proprietors in the noncorporate sector (re ported in the national income and product accounts) reacted in a curious way. Noncor porate business owners declined from 1952 to 1967 at about 2 percent a year, largely as the result of an exodus from farming that began in the 1930s. Between 1967 and 1972, the number of full-time noncorporate owners was roughly constant, increasing at about 1.5 per cent a year thereafter. This change in non corporate business formation and retention slowed the decline in relative (to employ ment in private industry) noncorporate own ership from about 3.3 percent a year to about 0.5 percent a year on average beginning in the late 1960s. One reason why the noncorporate sector turned around as it did may be that increas ing long-term weakness in labor markets dis couraged people from a career of working for wages. When job availability is diminished on a long-term basis, the noncorporate sector may serve as the employer of last resort. This suggests, however, that labor earnings poten tial in the noncorporate sector may not be as great as the average wage implies. Certainly, the notion of hidden unemployment in the noncorporate sector is not new, even applied to highly developed economies.3 If owners of non corporate enterprises were paid the average wage during periods when long-term labor market conditions are deteriorating, then just as in the case of cyclical fluctuations the bur den of limited earnings capacity would fall most heavily on the capital component. To adjust for the effects of cyclical and secu lar distortions of the composition of noncor porate earnings, we estimated a simple inter active model of the ratio of full-time owners Federal Reserve Bank of Cleveland of noncorporate businesses to full-time em ployees in private industry. Using maximum likelihood techniques, this employment ratio was regressed on a time trend and measures of long- and short-term labor market conditions: (A.l) InRFTE = -0.0551 - 0.0837T (0.1681) (0.0059) + 0.0517LT- 1.2392L (0.0052) (0.1618) + 0.1090S + e. (0.0121) R2 = 0.981p = 0.682S£ = 0.014 (Standard errors are in parentheses.) where RFTE = partners and proprietors devoting substantially full time to their businesses divided by full-time equivalent em ployees in private industry, L = measure of long-term labor market conditions: natural rate of unemployment divided by minimum natural rate achieved in period 1952-82, S = measure of short-term labor market conditions: difference between actual and natural rates of unemployment divided by natural rate. In this model the sum (-0.0551 - 1.2392L) defines the intercept for each level of the nat ural rate of unemployment. The trend com ponent (-0.0837T) captures the long-term decline in the full-time equivalents ratio, and the interaction term (0.0517L7) retards the trend rate of decline as the natural rate rises. The effects of short-term cyclical fluctuations are captured by (0.1090S). To estimate the labor component of total noncorporate earnings, an adjusted full- b. This admittedly is somewhat arbi trary. It suggests that any deviation from the period m in im um natural rate, if uncorrected, dis torts the labor capi tal division of non corporate earnings, but that the m in im um determines an appropriate base line division. Several alternatives were explored, including replacing the m in im um with a trend value of the natural rate, and making no adjustment at all. In 1982, the esti mate of the before tax rate of return was 6.5 percent. If a trend value of the natural rate had been used, the rate would have been lower, 6 . 0 percent; if no adjustment had been made, the estimate would have been 5.4 percent. c. In the tax calcu lations income mea sures were converted to taxable income equivalents. The relationship between personal income in the national income and product accounts and taxable income (adjustedgross in come) is illustrated by Hinrichs (1975). Noncorporate capi tal earnings and rent were approximated on an A G I basis by removing the capital consumption and inventory valuation adjustments and all imputations made in the national income accounts. 33 employment equivalents ratio, ARFTE, was calculated by setting the actual rate of un employment equal to the natural rate, and both equal to the minimum natural rate. That is, all labor market fluctuations about the mini mum natural rate were removed.b The labor component of total noncorporate earnings was calculated from ARFTE as follows: NCLC = (ARFTE * EPI * W)w, where NCLC = noncorporate labor composition, EPI = full-time equivalent employ ees in private industry, W = average wage, weighted by distribution of partners and proprietors by industry, w - markup on wages for other labor compensation (pension and profit sharing). When ARFTE exceeds RFTE, the owners of noncorporate enterprises receive the entre preneurial bonus (that is, they earn more than the average wage for their labor ser vices); when ARFTE is less than RFTE, their wages are cut. In the 1970s, the implied wage cut progressively widened, and by 1982 it amounted to nearly 30 percent of the average wage. Even with this allowance for changes in labor market conditions, capital earnings in the noncorporate sector declined sharply rel ative to labor earnings in the early 1980s. Cap ital earnings generally exceeded 30 percent of total noncorporate earnings prior to 1979. This proportion fell to an average of 13 per cent between 1980 and 1982, and, in the reces sion year of 1982, capital earnings were about 3 percent of total noncorporate earnings. A final problem of estimating the division of total noncorporate earnings into labor and capital components relates to taxes. To calcu late the after-tax rate of return, taxes also must be apportioned to labor and capital earn ings. The first step here was to estimate the tax liability on total noncorporate earnings and Economic Review • Fall 1984 rent. This was done by computing the ratio of total noncorporate earnings plus rent to personal income and multiplying the result by personal taxes paid (federal, state, and local).0 The resulting tax liability then was distrib uted to capital (including rents) and labor according to their proportions in the aggre gate. The effective tax rate on capital earnings was 12.4 percent in 1952 and was relatively constant until the late 1960s. Thereafter, the effective tax rate rose to about 20 percent in the early 1980s. For years when comparisons can be made, these tax rates are similar to esti mates made by other researchers (Kahn 1964). Appendix 2 Estimates of Land Prices and Real Values One of the most challenging problems in con structing estimates of the rate of return on capital is to derive adequate measures of land. Land is an integral part of the capital stock, along with inventories and fixed repro ducible capital, and business earnings must be evaluated relative to its value as well as the value of the reproducible assets. Nom inal land value is the product of market price (which is equal to the replacement price for nonreproducible assets such as land) and the quantity of land in use. The analysis of rates of return developed here requires estimates not only of nominal land value, for computing rates of return, but a separation of this value into price and real value (quantity), for com puting the margin and multiplier effects on rates of return. Detailed statistical information on land prices is restricted to farm land (Goldsmith 1982). Past studies of the rate of return filled the information gap in a variety of ways: by ignoring land altogether (Lovell 1978); by linking nominal land value to the value of structures on the land (Feldstein and Sum mers 1977, following Goldsmith 1962 and Den d. As an example of this calculation, in 1980, 710,390 pri vate single-family housing units were authorized, which also means 710,390 structures were authorized, mostly for owner occupancy. A t the same time, 53,768 two-family units were autho rized, or 26,884 structures, half of which were pre sumably for rental. Three or more fa m ily units contributed 39,895 structures, virtually all for ren tal. Thus, there were 777,169 struc tures authorized, 53,337of which were for rental or busi ness use, a ratio of 6.9 percent. See U.S. Department of Com merce, Bureau of the Census, Housing Units Authorized by Building Perm its and Public Con tracts: Annual 1980, July 1981, table 2 , p. 4. 34 ison 1974); by using book-value (historical price) data from income-tax records (Holland and Myers 1979); and by combining incometax data, benchmark estimates from propertytax-assessment records, and price distribu tion assumptions such as uniform land prices across sectors (Fraumeni and Jorgenson 1980). The few attempts to estimate land prices outside the farm sector (Milgram 1973) have not produced continuing time series beyond the limits of the study. Market value estimates of farm and non farm business land recently have been devel oped in the Board of Governors (1983) flowof-funds national balance sheets. These data provide the variables necessary for estim at ing rates of return and are the starting point for separating values into price and real value components. Acreage in farm production is monitored and reported by the U.S. Depart ment of Agriculture (1983). The data on acreage allow an average dollar price per acre to be computed from the market value estimates of farm land in the flow-of-funds balance sheets. Thus, real value of the farm land (1972 dol lars) and a simple price-relative index can be computed. The price index increases from 0.321 in 1952 to 3.396 in 1982, an annual rate of increase of about 8 percent, compared with about 4 percent for the GNP deflator. The most rapid increase, of course, took place in the 1970s. Between 1970 and 1981, the price index on farm land rose by about 14 percent a year; in 1982, the price index declined. Real value of the farm land stock declined by about 0.5 percent a year between 1952 and 1982. As indicated by the variety of approaches used in other studies to estimate land values, only a rough-and-ready approximation of non farm land price and quantity components can be developed. In this study, it is assumed that all changes in nonfarm business land use were transfers from agricultural use. This is approximately true, judging from the little information available on land transition. An Agriculture Department photographic inter pretation study of land transition in 53 rapidly growing counties throughout the United States between 1961 and 1970 indicated that over Federal Reserve Bank of Cleveland 90 percent of the land entering nonfarm bus iness use, including forests, was withdrawn from agricultural use (Zeimetz 1976). It is not certain, of course, that data drawn from such a small sample is representative of the United States as a whole, but it is likely that land transition has been concentrated in fast-growth areas. From transition matrixes compiled from the photographic evidence, it appears that the transition rate from agricultural to commer cial and industrial use was about 20 percent; that is, for every 100 acres withdrawn from agriculture, 20 acres was shifted into com mercial and industrial use. The highest tran sition rate (58 percent) was into residential use and another 18 percent of the loss in agri cultural land was diverted into transporta tion and recreation uses. Commercial and industrial businesses include a large propor tion of total business enterprises, but do not exhaust the total. Some portion of the land transferred to transportation and recre ation uses may represent a business gain, but the largest exclusion is the rental compo nent of the residential transfer. An examination of housing units authorized since the mid-1960s suggests that an average of about 6.5 percent of new structures was multi-family dwellings.d Multi-unit structures naturally are more land-intensive than single unit structures. Two-family structures may use only marginally more land, while large apartm ent buildings can use several times the land of the typical single-family home. There is no way of precisely determining intensity differentials, but an intensity coefficient of 2 was used here. This assumes two-family struc tures are equal to single-family structures in land use, three- and four-family structures are twice as land-intensive, and five- or more family structures are three times as land intensive. Thus, about 13 percent of the land transferred into residential use should be counted in the business sector. This raises the transition ratio from 20 percent to 28 per cent, and implies a total transition to non farm business use of about 47 million acres between 1952 and 1982. A survey of private landownership in 1978 indicated that nonfarm land in some form of business organization (proprietorship, part nership, or corporation) amounted to about 30 percent of the land in farm use (see USDA 1979). From the farm acreage data cited above, this would imply about 312 million acres in nonfarm business use in 1978. Using the tran sition ratio of 28 percent and the estimates of changes in farm acreage, nonfarm business land acreage can be estimated for 1952-82. From the market-value balance-sheet esti mates, the price index and real value compo nents also can be constructed. In 1972, the estimated average price per acre of nonfarm business land was $886.70, nearly 5 times the average price of farm land. The price index of nonfarm business land increased from 0.416 in 1952 to 2.779 in 1982, implying an annual rate of increase of about 6.5 percent. The real value of nonfarm business land increased by about 0.5 percent a year. Although estimates of nonfarm business land prices are consider ably higher than farm land prices, the rate of increase is less rapid, though still faster than the rate of increase in the GNP deflator. This is reasonable if improvements to the land limit the price inflation. Appendix 3 A Technical Note on Margin Drift Consider rate-of-return behavior as a growth process: (A.2) Rt = Ro(l + r)‘QQ)e, Rt - rate of return in current period, Rq = rate of return in base period, r = “normalized” (invariant with respect to cyclical and random disturbances) growth rate, QQ)t = allowance for cyclical and random movements in R (e > 0 fluctuates randomly about 1). From the definition of equation (3) in the text, the rate of return also is expressed as the Economic Review • Fall 1984 product of the earnings margin (M) and the earnings multiplier (m), and the rate of growth (r) can be viewed as the sum of the growth rates of these two components. (A.3) r = rM+ rm - rM + yqq + yqk - In the sense of a normalized long-term growth rate, the middle term in equation (A.3) is zero. The capacity ratio rises and falls as the economy moves through expansion and re cession phases of the business cycle (these effects on the rate of return are captured by QQKin equation (A.2)), but does not grow over time. In the long term, only contributions from the margin and capital productivity influence rate-of-return growth. This can be shown directly as follows: Let Ro be defined such that QQo = 1 (i.e., the initial value is a fullemployment level). From equations (3) and (4) in the text, the normalized growth rate of the rate of return must be: <A'4) (1 + rV = MrRQoQQKe £ K‘The earnings margin and capital productivity have grown from the base period at normal ized rates of tm and tqk, respectively. In addi tion, the margin fluctuates cyclically and randomly, while capital productivity fluctu ates randomly about its normalized growth path. Thus, (A.5) (1 + r)f = [Mp(l + rMyQQat u]QQt[QKo(l + rQK)‘w] 0 RoQQ/ e = (1 + rM)l(1 + rQKy, where t = uw, k = (1 + a), a > 0. Thus, (A.6) r = rM+ rQK + rMrQK, where rM rQK ~ small second-order effects. The question now becomes one of deter mining the nature of long-term growth in the earnings margin and capital productivity. Once cyclical and random movements in the margin are accounted for, a nonzero drift may remain, depending on the price markup rule that best approximates pricing strategy for the aggregate economy. To illustrate margin drift, consider the case where the relative price of capital is held constant (as in BXRC and AXRC), and output price is determined as a markup on a subset of total unit costs (for example, unit labor costs). Unit costs are normalized for cyclical and random variation in the markup function, and the earnings margin corresponding to the growth rate rM would be defined as follows: (A.7) Mt = (1 +/>)(! + g j ) t U io - { 1 + g t) l U io - ( 1 + g e ) ‘ Ueo (1 +/>)(1 +gi)tUio where p = price markup, Ui = unit costs included in markup function (subscript 0 denotes base period), gi = growth rate of included unit costs, Ue = normalized unit costs excluded from markup function, ge = growth rate of excluded unit costs. The continuous change in the normal mar gin with respect to time is given by (A-8) at r -U eo 1 L (i+ » « J X [ln(l + ge) - ln(l + gi)] X If growth rates of unit costs and the price markup are constant, the following possibili ties characterize change in the normal margin Federal Reserve Bank of Cleveland , v ~dM n If. , ge = gi. (a) jF = 0, dM > if Se < gi, dt — 0, as / increases. (c) dM < 0, if e > g i, dt — -00, as t increases. In case (a), no drift in the earnings margin is indicated; hence, the normalized growth rate (rM) would be zero. In case (b), however, a positive drift is indicated, gradually becom ing smaller over time. When the difference between growth rates of included and excluded unit cost is fairly small, margin drift is also small but narrows very gradually (i.e., approx imately constant for relatively long periods). Finally, if those unit costs excluded from the markup function grow faster than included unit costs, as in case (c), the normal margin would continuously deteriorate over time. This could not be descriptive of pricing strat egy over any extended period, but it could describe a transition period of price-cost re alignment. Negative drift would be possible in variable relative-capital-price margins, even if g i> g e- Increases in the replacement price of capital pull the margin down; the faster the capital price rises relative to the output price, the more likely the possibility of nega tive margin drift. (b) 8 References Baxter, John D. “W hat’s Behind Falling Prod uctivity of Equipment?” Iron Age, Decem ber 11, 1978, pp. 27-30. Bernstein, Peter L. “Capital Stock and Man agement Decisions,” Journal of Post Keynes ian Economics, Fall 1983, pp. 20-38. Board of Governors, Federal Reserve System. Balance Sheets for the U. S. Economy 1945-82. Washington, DC: October 1983. Bosworth, Barry P. “Capital Formation and Economic Policy,” Brookings Papers on Economic Activity, 2:1982, pp. 273-326. Cagan, Phillip, and Robert E. Lipsey. The Fi nancial Effects of Inflation. Cambridge, MA: Ballinger Publishing Company, 1978. Clark, Peter K. “A Kalman Filtering Approach to the Estimation of Potential GNP,” Pro cessed. Revised, November 1983. Denison, Edward F. Accounting for United States Economic Growth 1929-1969. Wash ington, DC: Brookings Institution, 1974. Eisner, Robert. “Capital Gains and Income: Real Changes in Value of Capital in the United States, 1946-77,” in Dan Usher, Ed., The Measurement of Capital. Chicago: Uni versity of Chicago Press, 1980, pp. 175-342. Fain, T. Scott. “Self-employed Americans: Their Number Has Increased,” Monthly Labor Review, November 1980, pp. 3-8. Feldstein, Martin S., and Lawrence Sum mers. “Is the Rate of Profit Falling?” Brookings Papers on Economic Activity, 1:1977, pp. 211-27. Fraumeni, Barbara M., and Dale W. Jorgen son. “The Role of Capital in U.S. Economic Growth, 1948-1976,” in George M. Von Furstenberg, Ed., Capital, Efficiency, and Growth. Cambridge, MA: Ballinger Publish ing Company, 1980, pp. 9-250. Goldsmith, Raymond W. The National Wealth of the United States in the Postwar Period. Princeton: Princeton University Press, 1962. ______ The National Balance Sheet of the United States 1953-1980. Chicago: Univer sity of Chicago Press, 1982. Hinrichs, John C. “The Relationship between Personal Income and Taxable Income,” Survey of Current Business, February 1975. Holland, Daniel M., and Stewart C. Myers. “Trends in Corporate Profitability and Capital Costs,” in Robert Lindsey, Ed., The Nation s Capital Needs: Three Studies. New York: Committee for Economic Develop ment, 1979, pp. 103-88. Kahn, C. Harry. Business and Professional In come Under the Personal Income Tax. Prince ton, NJ: Princeton University Press, 1964. Kopcke, Richard W. “The Decline in Corpo rate Profitability,” New England Economic Review, Federal Reserve Bank of Boston, May/June 1978, pp. 36-60. Economic Review • Fall 1984 Lovell, Michael C. “The Profit Picture: Trends and Cycles,” Brookings Papers on Economic Activity, 3:1978, pp. 769-88. Milgram, Grace. “Estimates of the Value of Land in the United States Held by Various Sectors of the Economy, Annually, 1952 to 1968,” in Raymond W. Goldsmith, Ed., Insti tutional Investors and Corporate Stock—A Background Study, New York: Columbia University Press, 1973, pp. 343-77. Nordhaus, William D. “The Falling Share of Profits,” Brookings Papers on Economic Activity, 1:1974, pp. 169-208. Okun, Arthur M., and George L. Perry. “Notes and Numbers on the Profits Squeeze,” Brookings Papers on Economic Activity, 3:1970, pp. 466-72. Phillips, Joseph D. The Self-employed in the United States. Urbana, IL: University of Illinois Press, 1962. Poirier, Dale J. The Econometrics of Structural Change. Amsterdam: North-Holland Pub lishing Company, 1976. Runyon, Herbert. “Profits: A Declining Share to Capital?” Business Economics, vol. XIV, no. 4 (September 1979), pp. 85-94. Scanlon, Martha S. “Postwar Trends in Cor porate Rates of Return” Public Policy and Capital Formation. Board of Governors of the Federal Reserve System, April 1981, pp. 75-87. U.S. Department of Agriculture, Who Owns the Land? A Preliminary Report of a U.S. Landownership Survey, September 1979. U.S. Department of Agriculture, Economic Indicators of the Farm Sector. Income and Balance Sheet Statistics, 1983. Walton, John. “Capital Maintenance and the Measurement of Corporate Income,” Re view of Income and Wealth, series 27, no. 2 (June 1981), pp. 109-35. Zeimetz, Kathryn A., et al. Dynamics of Land Use in Fast Growth Areas, U.S. Department of Agriculture, April 1976. This working paper complements James G. Hoehn and William C. Gruben, with Thomas B. Fomby, Some Time Series Methods of Forecasting the Texas Economy, Working Paper 8402, Federal Re serve Bank of Dallas, May 1984. James G. Hoehn is an econo mist with the Fed eral Reserve Bank of Cleveland. 1. Anderson, Paul A. "Help for the Re gional Forecaster: Vector Autoregres s i o n Quarterly Review, Federal Re serve Bank of M inne apolis, vol. 3, no. 3 (Summer 1979), pp. 2-7; and Kuprianov, Anatoli, and William Lupoletti. “The Economic Out look for Fifth Dis trict States in 1984: Forecasts from Vec tor Autoregression Models,” Economic Review, Federal Reserve Bank of Richmond, vol. 70/1 (January/February 1984), pp. 12-23. 2. Ashley, Rich ard A., C.W.J. Granger, and R. Schmalensee. ‘Advertising and Aggregate Con sumption: An Analysis of Cau sality,” Econometrica, vol. 48, no. 5 (1980), pp. 1149-67. 38 Working Paper Review James G. Hoehn A Regional Economic Forecasting Procedure A pplied to Texas Working Paper 8402. September 1984. 45 pp. Bibliography. In this paper economist James G. Hoehn pro poses and implements a relatively simple method for building a multivariate autore gressive forecasting model for regional eco nomic time series. The method used is a time series approach requiring little a priori or theoretical knowledge, as in the building of structural econometric models. In this way, the method is similar to the so-called vector auto regression (VAR) models of Anderson and Kuprianov and Lupoletti.1However, the way variables are chosen to be included and the way relationships are estimated involve more hypothesis tests and less prior knowledge. We applied the method to the problem of forecasting quarterly growth rates of seven Texas variables that were seasonally adjusted. Each of the seven variables was related to its own past two observations in a simple re gression equation to determine whether lagged values would aid forecasts. Then, two lagged growth rates of other regional and national variables were added to the equation to con struct Granger causality tests. Such tests sug gested whether inclusion of the other regional and national variables was likely to improve Federal Reserve Bank of Cleveland forecasts. After the series of causality tests was completed, a list of variables was sug gested as likely candidates for inclusion in the forecasting equation for each Texas vari able. This list was trimmed, and lag speci fications were chosen, using the criteria of minimizing the standard error of the equa tion and the principle of parsimony. The re sulting equations, although chosen on statis tical grounds alone, conformed reasonably well to intuitions about the regional economy and its relations with the national economy. Of 14 national variables tried, inclusion of 4 appeared to capture most of the information— the index of leading indicators, the index of coincident indicators, the producer price index, and the federal funds rate. Out of sample, the model was found to forecast consistently better (lower root mean square error) than benchmark univariate autoregressive integrated moving average (ARIMA) models, sometimes substantially bet ter. In some cases, the improvement was sta tistically significant at the 0.05 level, despite the shortness of the forecasting sample, accord ing to a test adapted from Ashley, Granger, and Schmalensee.2 The model was reasonably stable as re-estimated over the out-of-sample period, and its forecasts could not be system atically improved by combining them with ARIMA forecasts. The results of the study are subject to a number of caveats common to statistical studies. Nevertheless, results suggest that the proposed forecasting procedure can pro vide systematically better forecasts than uni variate ARIMAs. Such systematically better forecasts apparently have not been delivered by the complex simultaneous equation models or by other time series methods. The proce dure proposed requires only ordinary least squares regressions. A by-product of model building is insight into the regional economy and the strength of various leading relation ships in the data. The method can easily be applied to other regional economies, and we have begun building a model for Ohio. The working paper series is published by the Research Department of the Federal Reserve Bank of Cleveland to stimulate discus sion and critical comment. Copies of worki ng papers, either future or past issues, are avail able through our Public Inform a tion Department, Federal Reserve Bank of Cleveland, PO. Box 6387, Cleve land, OH 44101; (216) 579-2048. Working Paper Series 8304 Forecasting the Money S upply in Time Series M odels Michael L. Bagshaw and William T. Gavin December 1983, 23 pp. 8101 The W elfare Im plications of A lternative U n em plo y m ent Insurance Plans Mark S. Sniderman April 1981, 20 pp. 8201 M u ltib a n k H olding C om pany O rg an iza tional Structure and Perform ance Gary Whalen March 1982, 30 pp. 8202 S tability in a M odel of StaggeredReserve A ccounting Michael L. Bagshaw and William T. Gavin August 1982, 27 pp. 8203 A M icro View of the Transactions M oney M arket Mark A. Zupan September 1982, 31 pp. 8301 Non-Nested Specification Tests and the In term e diate Target for M onetary Policy Mitsuru Toida and William T. Gavin June 1983, 14 pp. 8302 H olding Com pany O rganizational Form a nd Efficiency Gary Whalen July 1983, 20 pp. 8303 Extension of G ranger C ausality in M ultivariate Time Series M odels Michael L. Bagshaw August 1983, 11 pp. 8401 Economic Estim ates of U rban Infrastructure Needs Paul Gary Wyckoff June 1984, 36 pp. 8402 A Regional Economic Forecasting Procedure A pplied to Texas James G. Hoehn September 1984, 45 pp. 8403 Forecasting Using Contem poraneous Correlations Michael L. Bagshaw September 1984, 17 pp. 8404 D ollar Intervention and the Deutschem ark-D ollar Exchange Rate: A D aily Time-Series Model Owen F. Humpage September 1984, 28 pp. 8405 Velocity: A M ultivariate Time-Series A pproach Michael L. Bagshaw and William November 1984, 29 pp. 8406 M onetary Policy a nd Real Interest Rates: New Evidence from the Money Stock Announcem ents William T. Gavin and Nicholas V. Karamouzis December 1984, 43 pp. 8407 M onetary Policy Regimes: A Synthesis of the M onetary Control a nd R atio nal Expectations Literatures James G. Hoehn December 1984, 60 pp. 39 Economic Review • Fall 1984 T. Gavin Economic Review is published quar terly by the Research Department of the Federal Reserve Bank of Cleveland. Copies of the issues listed here are available through our Public Information Depart ment, 216/579-2048. Economic Review Winter 1983 “Location and Reinvestment: The Youngstown Steel District” Robert H. Schnorbus “Thrifts and the Competitive Analysis of Bank Mergers” Paul R. Watro Working Paper Review: Stability in a Model of Staggered Reserve Accounting Michael L. Bagshaw and William T. Gavin Spring 1983 “Money Demand: Cash Management and Deregulation” John B. Carlson “Divisia Monetary Aggregates: Would They Be More Palatable than the Traditional Simple-Sum Stews?” Mark A. Zupan Fall 1983 “Plant Closings and Worker Dislocation” Daniel A. Littman and Myung-Hoon Lee “Prevailing Wage Laws, the Federal Reserve, and the Service Contract Act” Mark S. Sniderman Working Paper Review: Forecasting the Money Supply in Time Series Models Michael L. Bagshaw and William T. Gavin Winter 1984 “Reflections on Money and Inflation” William T. Gavin “The Outlook for Inflation” K.J. Kowalewski and Michael F Bryan Spring 1984 “Estimating Infrastructure Needs: Methods and Controversies” Paul Gary Wyckoff “Nonbanking Operations of Bank Holding Companies” Gary Whalen Summer 1984 “The Implementation of Industrial Policy” Daniel A. Littman “Voluntary Export Restraints: The Cost of Building Walls” Michael F. Bryan and Owen F. Humpage Working Paper Review: Dollar Intervention and the DeutschemarkDollar Exchange Rate: A Daily Time-Series Model Owen F. Humpage Federal Reserve Bank of Cleveland