The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Business Review Federal Reserve Bank of Philadelphia january * February 199 4 __________________ISSN 0 0 0 7 -7 0 1 1 How Efficient Are Third District Banks? Loretta J. M ester New Indexes Track the State of the States Business Review The BUSINESS REVIEW is published by the Department of Research six times a year. It is edited by Sarah Burke. Artwork is designed and produced by Dianne Hallowell under the direction of Ronald B. Williams. The views expressed here are not necessarily those of this Reserve Bank or of the Federal Reserve System. SUBSCRIPTIONS. Single-copy subscriptions for individuals arc available without charge. Insti tutional subscribers may order up to 5 copies. BACK ISSUES. Back issues are available free o f charge, but quantities are limited: educators may order up to 50 copies by submitting requests on institutional letterhead; other orders are limited to 1 copy per request. Microform copies are available for purchase from University Microfilms, 300 N. Zeeb Road, Ann Arbor, MI 48106. REPRODUCTION. Perm ission must be obtained to reprint portions o f articles or whole articles. Permission to photocopy is unrestricted. Please send subscription orders, back orders, changes o f address, and requests to reprint to Publications, Federal Reserve Bank o f Philadelphia, Department o f Research and Statistics, Ten Independence Mall, Philadelphia, PA 19106-1574, or telephone (215) 574-6428. Please direct editorial communications to the same address, or telephone (215) 574-3805. JANUARY/FEBRUARY 1994 HOW EFFICIENT ARE THIRD DISTRICT BANKS? Loretta J. Mester Although banks are still the main financial interm ediaries in the United States, whether they remain so in the face of increased competition depends on how efficiently they operate. How can we measure a bank's efficiency? And what specifically can we expect for banks oper ating in the Third District? Loretta Mester examines the issue of efficiency and ap plies her findings to banks in the Third Federal Reserve District. NEW INDEXES TRACK THE STATE OF THE STATES Theodore M. Crone For many years the Department of Com merce has published an Index of Coinci dent Indicators to track the U.S. economy. Recently, two econom ists, one from Harvard and one from Northwestern University, developed a new index of coincident indicators for the nation. To gether these two indexes provide evi dence of the direction of the national economy. But what about regional econo mies? They do not always follow the national pattern, and there are no compa rable indexes to track their progress. Ted Crone discusses indexes available at the national level, then details the develop ment of indexes for the states in the Third District. FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? Loretta ]. Mester * I n recent years banks have had to operate in an increasingly competitive environment. Com petitors have come from both within and out side the banking industry. Deregulation has allowed commercial banks to expand beyond their own state's borders; thus banks face com petition from other commercial banks entering their market for the first time. Investment banks have also become competitors for some of the commercial bank's most creditworthy * Loretta J. Mester is a Research Officer in the Research Department, Federal Reserve Bank of Philadelphia, and Adjunct Assistant Professor of Finance, The Wharton School, University of Pennsylvania. customers, who have been able to turn to the commercial paper market as a cheaper funding source than bank loans. Similarly, savers have been funneling their money into mutual funds as opposed to bank deposits in a search for a higher rate of return in the current low-depositrate environment. Although banks are still the main financial intermediaries in the United States, providing funding to firms and other borrow ers and deposit services to savers, whether they will remain dominant in the face of increased competition depends on how effi ciently they produce their outputs, that is, their loans and other financial services. Efficient banks will be able to offer more attractive loan and deposit rates to their customers and still 3 BUSINESS REVIEW make a normal rate of return, while inefficient banks won't be able to follow suit and will, therefore, lose business. Inefficiently run banks will have to shape up, or they will be driven out of the market or acquired by other banks; in deed mergers between efficient and inefficient banks have the potential for substantial social gains via cost savings.1 And banks that operate * with lower costs can pass these savings on to their borrowers and depositors. What can we expect for banks operating in the Third Federal Reserve District, which com prises the eastern two-thirds of Pennsylvania, the southern half of New Jersey, and Delaware? Are they operating at a high level of efficiency, or is there room for substantial improvement? Can we expect a lot of restructuring in our District as inefficient banks are driven from the market? Measuring our banks' efficiency will give us some indication of how they are likely to fare in an increasingly competitive environ ment. WHAT DO WE MEAN BY EFFICIENCY? When economists consider efficiency they typically focus on scale and scope efficiency, which concerns a bank's choice of outputs, and Xefficiency, which concerns a bank's use of in puts. There has been substantially more study of scale and scope efficiency in the banking and financial services industry than of X-efficiency. Scale Efficiency. Scale efficiency refers to whether a firm is providing the most costefficient level of outputs. Let's consider a hypo thetical example. Suppose firms are demand ing $500 million of credit in total, and that it 1Although Berger and Humphrey (1992b) found little in the way of cost-efficiency benefits on average from mergers in the 1980s between banks with assets over $1 billion, as they discuss, this is likely because the aim of such mergers was to increase asset growth and geographic market exten sion rather than to increase cost efficiency. This seems to be changing in the 1990s, as merger participants have been setting cost-cutting goals when announcing their mergers. Digitized for 4 FRASER JANUARY/FEBRUARY 1994 costs one bank $25 million to produce this volume of loans and $10 million to produce $250 million in loans. (Producing a loan in volves the credit evaluation the bank must perform to determine the credit quality of the borrower along with funding the loan and monitoring the loan over its length of matu rity.) Then it is more efficient to supply the $500 million of credit to the market by having two banks each produce $250 million of the loans than by having one bank produce all $500 million—the average cost of production, that is, the cost per dollar of loan, is less (4c versus 5c) when each bank produces $250 million of loans than when one bank produces $500 mil lion of loans. Society is better off with two banks producing the output rather than one bank, since the $5 million saved could be used for some other productive activity. And a bank trying to produce all $500 million of loans would find itself at a competitive disadvantage if another bank entered the market producing only $250 million of loans, because the second bank would be able to charge a lower interest rate on its loans, since its per unit production cost would be lower. A bank is said to be producing with constant returns to scale if, for a given mix of products, a proportionate increase in all its outputs would increase its costs in the same proportion; this is also the point where the average cost of produc tion is the least. A bank is operating with scale economies if a proportionate increase in its out puts would lead to a less than proportionate increase in cost—the bank could produce more efficiently by increasing its output level. A bank is operating with scale diseconomies if a proportionate decrease in its outputs would lead to a more than proportionate decrease in costs—the bank could produce more efficiently by reducing its output level. At output levels where there are scale econo mies, an increase in outputs would reduce the average cost of production, since it costs pro portionately less to produce at a larger scale. FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? Loretta /. Mester One reason it might cost less per unit to pro duce at a larger scale is that there may be large fixed costs in the production technology that are independent of the level of output pro duced. For example, in banking, the cost of the computers used to keep track of accounts can be spread over a larger number of accounts as the scale of operations increases, so that the per unit cost of production falls. Another potential source of scale economies is specialization. Larger firms may permit employees to special ize in one task, and this specialization may also lead to more efficient production.2 In most industries, including banking, firms find that at a certain output volume (for a given mix of products), average cost stops declining. For example, at a large enough volume, the fixed costs of production will become insignificant relative to the cost of producing additional units of output, so that scale economies are exhausted. And if scale is increased beyond a certain level, average costs begin to rise. Of course, depending on the production technol ogy, there may be a broad range of output levels (for any product mix) that firms can produce at minimum average cost. In other words, banks of different asset sizes may be equally competi tive with one another, since their average costs are similar. Scope Efficiency. Scope efficiency refers to whether a firm is producing the most costefficient combination of products. Banks pro duce more than one product—for example, most commercial banks produce a variety of different loans, like commercial and industrial loans, commercial real estate loans, residential mortgages, student loans, etc. To the extent that different types of loans have different default rates or other characteristics and to the extent that they aren't used to fund the same activities, they constitute different outputs of the bank.3 Thus, in addition to choosing the most cost-efficient scale of operations, the bank must also choose the combination of products it will produce. For a given level of outputs, the per unit cost of production may be smaller if the bank produces all of the products rather than specializing in just a few of them, or it might be more efficient to specialize. There are scope economies if the cost of producing a given level of outputs is lower when a bank produces all the products than if the products are divided up into specialized banks. There are scope diseconomies if the costs are lower when special ized banks produce the various outputs. There are several potential sources of scope economies.4* One is the sharing of inputs to produce several outputs. For example, the same group of tellers might handle both check ing and savings accounts, or information on a firm produced in a credit evaluation for a mortgage can be used if the firm wants a busi ness loan as well. Therefore, it would be cheaper for the same bank to handle both types of accounts and to extend both loans than to duplicate the tellers and credit check at another bank. Similarly, excess capacity on the bank's computer may allow it to increase the scope of products it produces as well as its scale. Thus, there is an interconnection between scale and scope economies—the fact that the bank is able to process various types of loans on its com puter (many products) enables it to increase its scale and take advantage of any scale econo mies. Of course, there may be a point at which producing many products will increase the bank's unit costs. For example, it may take a more elaborate hierarchical management struc ture to produce different product lines (some 2Mester (1987) discusses the sources of scale economies in more detail. 4Mester (1987) describes the sources of scope economies in more detail. 3Large banks also engage in many off-balance-sheet activities, like underwriting, letters of credit, and loan guar antees. 5 BUSINESS REVIEW JANUARY/FEBRUARY 1994 times this is mandated by regulation—e.g., equities underwriting and commercial lending must be done in separate subsidiaries of a bank holding company), and this hierarchical struc ture can increase production costs.5 X-Efficiency. If all firms in the industry are producing the level and combination of out puts that minimize the average cost of produc tion, the total cost of producing the industry's output is minimized, and the industry is pro ducing an efficient combination and level of products, provided each firm is using its inputs efficiently. X-efficiency refers to whether a firm is using its inputs, like labor and capital, in a cost-effective manner— that is, for a given level and mix of outputs, is a bank producing them in the cheapest way possible? If not, the bank is either wasting some of the inputs it has purchased, or it is using the wrong combina tion of inputs to produce its outputs. Technical inefficiency refers to using proportionately too much of all inputs and is just pure waste. For example, the bank may be using too many tellers and too many branches to produce its products—it might be able to scale back its inputs and produce the same amount of ser vice. A bank that is technically inefficient is said to be operating within its "production possibil ity frontier." (The production possibility fron tier indicates the maximum amount of output that can be produced with a given amount of inputs.) But wasting resources is not the only way to inefficiently use inputs. A bank might be able to produce a given amount of loans and other financial services by combining its inputs— including labor, physical capital, and deposits— in different proportions than it cur rently is doing. For example, a large bank might be able to supply its output more cheaply by substituting ATMs for tellers—while the fixed costs of setting up an ATM are high, the cost per transaction for an ATM is lower than that for a human teller—so larger banks might benefit by using more ATMs than tellers. Allocative ineffi ciency refers to using the wrong combination of inputs to produce a given output level and product mix—an allocatively inefficient bank is operating on its production possibility frontier—that is, given the inputs it has chosen, it is producing as much output as possible—but the bank could lower its costs of producing that output by selecting a different input mix. Of course, a bank can be both technically and allocatively inefficient. A bank that is operating in an inefficient manner might be doing so because its manager isn't on top of things, but managerial inability isn't the only source of X-inefficiency. It's possible that a bank manager has goals that differ from those of the bank's shareholders. Shareholders want to maximize the stock mar ket value of the bank, and so its long-run profits. Thus, shareholders want the bank to minimize its cost of production. But bank managers might be interested in something other than cost-minimization. For example, managers might desire a larger staff because they think it gives them more prestige within the banking community. Thus, a bank might use an inefficient combination of inputs (more labor than is necessary) to produce its services. Such "expense-preference" behavior on the part of managers has been found in studies of com mercial banks and savings and loans.6* The bank's choice of the products it wishes to pro duce might also be driven less by cost consid erations than by managerial desires to run a particular type of bank. Survival. One might question how ineffi cient banks are able to continue operating. In the usual economic models of competitive markets, competitive forces are thought to drive 5See Mester (1991) for further discussion of diseconomies of scope in hierarchies. 6Mester (1989) discusses the conflicts between owners and managers in financial firms and the empirical evidence. 6 FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? such inefficient banks out of the market. More efficient banks are able to produce at a lower cost. In a competitive market, the efficient bank would share its cost savings with its customers in the form of lower interest rates on loans and / or higher deposit rates. This would attract borrowers and depositors away from ineffi cient banks, since the inefficient banks couldn't match the lower prices without making a loss. The inefficient banks would eventually be forced out of the market. But banking has been a regulated industry; competitive pressures have not been as strong as they might have been. For example, prior to the 1980s, regulations restricted bank holding companies from establishing banks in more than one state, and there are still restrictions on banks establishing branches across state lines. Such restrictions reduce the number of poten tial competitors, making it easier for inefficient banks to survive.7 Similarly, laws that restrict hostile takeovers make it less easy for more efficient banks to gain control of their less efficient counterparts. On the customer side, there is empirical evidence that bank customers have found it costly to switch banks (see Calem and Mester, 1993); thus, it has been difficult for efficient banks to attract customers with lower prices. But inefficient banks will find it less easy to survive in the future as entry barriers fall. States began passing laws in the 1980s that authorize interstate banking for bank holding companies. All but two states (Hawaii and Montana) now allow bank holding companies from at least some other states to acquire in state banks. In April 1992, the Office of Thrift Supervision adopted a rule allowing full na tionwide branching for healthy federally char tered savings and loans. According to the A m erican B an ker (A u gu st 2, 1993) four 7Calem (1993) discusses the benefits of allowing banks to branch across state lines. Loretta f. Mester states—New York, North Carolina, Oregon, and Alaska—have passed reciprocal interstate branching laws permitting a state-chartered commercial bank that is not a member of the Federal Reserve System to become a branch of a bank in any other state that has an identical law. Although interstate branching hasn't been authorized for national banks as yet, Congress has considered several proposals to permit it, and the topic is likely to remain on the agenda. In addition, the Federal Reserve has taken the position that it will treat hostile bids no differ ently from friendly bids in assessing whether to permit a takeover. Nonbank competition is also picking up. According to the flow of funds accounts, banks' share of total U.S. financial assets has shrunk to less than 25 percent from over 35 percent in 1977.8 And foreign bank competition is heating up too; the North Ameri can Free Trade Agreement (NAFTA) should also increase competition. Increased competi tive pressures will make it more difficult for inefficient banks to survive, as will anything that reduces the costs customers face in switch ing to low-cost banks. For example, the Truth in Savings Act, part of the Federal Deposit Insurance Corporation Improvement Act of 1991, requires banks to report the terms of their deposit accounts in all advertisements for these accounts, making it easier for customers to shop for the best rates. Inefficient banks will find it more difficult to keep well-informed customers. MEASURING EFFICIENCY: THE METHODOLOGY Outputs and Inputs. Studies of bank effi ciency are based on an analysis of banks' cost structure, that is, the relationship between 8The flow of funds accounts, published by the Board of Governors of the Federal Reserve System, provide data on financial assets and liabilities outstanding by sectors of the economy and by type of transaction. 7 BUSINESS REVIEW JANUARY/FEBRUARY 1994 banks' costs and their output levels, given the input prices they face. Thus, the first step in measuring efficiency in banking—scale and scope efficiency and X-efficiency—is to deter mine a bank's outputs and inputs. There is some disagreement in the literature over what a commercial bank is actually producing. Two general approaches have been taken: the "pro duction" approach and the "intermediation" approach (also called the "asset" approach). The production approach focuses on the bank's operating costs, that is, the costs of labor (em ployees) and physical capital (plant and equip ment). The bank's outputs are measured by the number of each type of account, like commercial and industrial loans, mortgages, deposits, be cause it is thought that most of the operating costs are incurred by processing account docu ments and debiting and crediting accounts; inputs are labor and physical capital. The intermediation approach considers a financial firm's production process to be one of financial intermediation (the borrowing of funds and the subsequent lending of those funds). Thus, the focus is on total costs, including both interest and operating expenses. Outputs are measured by the dollar volume of each of the bank's different types of loans, and inputs are labor, physical capital, and deposits and other bor rowed funds.9*Luckily, the empirical results on scale and scope efficiency do not seem to be very sensitive to which approach is taken. Theoretically, to compare one bank's effi ciency to another's, we would like to compare each bank's cost of producing the same outputs. For banks, significant characteristics of loans are their quality, which reflects the amount of monitoring the bank does to keep the loan performing, and their riskiness. Unless these characteristics are controlled for, one might conclude a bank was producing in a very effi cient manner if it were spending far less to produce a given output level, but its output might be highly risky and of a lower quality than that of another bank. It would be wrong to say a bank was efficient if it were scrimping on the credit evaluation needed to produce sound loans. Although previous efficiency stud ies have failed to compare the costs of produc ing outputs of equal quality and risk, the study of Third District banks described below does so. Scale and Scope Efficiency Studies. Most of the studies interested in measuring scale and scope efficiency for a particular sample of banks have estimated an average practice cost function, which relates a bank's cost to its output levels and input prices. The technique implicitly assumes that all banks in the sample are using their inputs efficiently, that is, there is no Xinefficiency, and they are using the same pro duction technology. Of course, it recognizes that data are typically measured with error and that there might have been unpredicted factors 9A slight variation on the intermediation approach, which has been used in some studies, is to distinguish between transactions deposits, which are treated as an output, since they can serve as a measure of the amount of transactions services the bank produces, and purchased or borrowed funds (like federal funds or large CDs purchased from another bank), which are treated as inputs, since the bank does not produce services in obtaining these funds. The strict intermediation approach would consider the transactions services produced by the bank as an intermedi ate output, something that must be produced along the way toward the bank's final output of earning assets. Hughes and Mester (1993) empirically tested whether deposits should be treated as an input or output and found that they should be treated as an input in their study. Another approach that has been taken less often is the "value-added" approach, which considers all liabilities and assets of the bank to have at least some of the characteristics of an output. See Berger and Humphrey (1992a) for further discussion. Still another approach, taken in Mester (1992), is to consider the bank's output to be its loan origination and loan monitoring services. See Humphrey (1985) and Berger and Humphrey (1992a) for further discussion of the different approaches to mea suring bank output. 8 FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? that affected a bank's cost over the period when the data were collected, like an unusually large amount of computer down time or up time (bad and good luck) or extraordinary sick leave. Thus, no bank is expected to lie precisely on the estimated cost function; instead the function indicates what, on average, it costs a bank facing a particular set of input prices to produce a particular bundle of outputs. Some banks will produce the given output at a slightly higher cost and others at a slightly lower cost than is indicated by the estimated cost function. Most studies have focused on smaller banks, with assets less than $1 billion. These studies, others that included banks of all sizes, and another study that included all banks with assets greater than $100 million found that the average cost curve is relatively flat, with scale economies exhausted somewhere between $75 million and $300 million in assets.1 This is a 0 relatively small size when you consider the size distribution of U.S. banks. While in 1992 about 90 percent of the 11,461 FDIC-insured commer cial banks in the United States had less than $300 million in assets, these banks held only 20 percent of total bank assets. Fifty-one banks had assets over $10 billion, and the largest, Citibank, had over $150 billion. Thus, the studies of scale economies suggest that only small banks are operating with unexploited economies of scale and could become more efficient producers by expanding their output size. Moreover, the measured scale economies for these small banks are usually fairly small: a 1 percent increase in all output levels typically leads to about a 0.95 percent increase in total cost, which means a 0.05 percent decrease in the average cost of producing the bundle of out puts. A handful of studies have focused solely on large banks with assets over $1 billion. Some 10Berger and Humphrey (1992b), Evanof f and Israilevich (1991), Clark (1988), and Mester (1987) summarize the re sults of the studies. Loretta ]. Mester found scale economies at very large banks—the minimum of the average cost curve usually was found to lie between $2 billion and $10 billion in assets. But here again, measured economies were not very large. On the whole, these studies concluded that there wasn't much in the way of cost gains to be made by changing the scale of operations at the typical bank.1 1 Similarly, although there are exceptions, most studies have found little evidence of economies or diseconomies of scope between the products banks currently produce. Hence, there is little evidence that changing the typical bank's prod uct mix would significantly influence its cost of production.1 1 2 3 X-Efficiency Studies. More recent studies have focused on measuring not only scale and scope economies but also the degree of Xinefficiency in banking. As with scale and scope efficiency, we start with a set of banks that are using the same production technology for creating output. The technique is to esti mate a best practice cost function—that is, the predicted cost function of banks that are Xefficient—and then measure the degree of inef ficiency relative to this best practice technol ogy. Two common methodologies are data envelopment analysis (DEA) and stochastic econo metric cost frontier analysis.1. 3 11This isn't to say that banks operating at a significant distance from optimal scale couldn't become more efficient by changing their operating scale. See Evanof f and Israilevich (1991) for more discussion on this point. 12This is not to say that deregulation that permits banks to expand the types of products they can offer (e.g., equities underwriting) could not enable banks to take advantage of potential scope economies. 13There are other techniques for deriving efficiency mea sures, including so-called "thick frontier" analysis and "shadow price" models. Evanoff and Israilevich (1991) describe these techniques. A simpler method to compare the efficiency of banks is to use peer-group analysis. Certain cost ratios are com- 9 BUSINESS REVIEW DEA uses the data on costs, outputs, and input prices for a sample of banks and deter mines, for each output bundle and set of input prices, the bank in the sample that spends the least to produce the output bundle at the given input prices— this is the "best practice" (that is, most efficient) bank for that output /input price combination. (If no bank in the sample pro duces a particular combination, then a "best practice" bank for the combination is approxi mated based on "best practice" banks produc ing similar combinations that do show up in the sample.) A bank's relative inefficiency is then measured by the ratio of its own cost compared with the cost of the "best practice" bank that faces the same input prices and produces the same output bundle. The technique is called data envelopment analysis because the data on best practice banks "envelop" the data from the rest of the banks in the sample. One benefit of DEA is that it doesn't posit a particular func tional form for the best practice banks' cost function—it is more flexible. But a serious drawback of the technique is that it does not allow for any error in the data—banks that have been lucky or w hose costs have been undermeasured will be labeled as most effi cient and other banks will look relatively less efficient in comparison. Similarly any unfavor able influence beyond the bank's control will be attributed to inefficiency. pared for banks that are considered to be similar in the types of customers they serve and products they produce. These ratios might include operating expenses per dollar volume of assets, number of employees per dollar volume of loans, or expenses attributed to commercial loans per volume of commercial loans. The Functional Cost Analysis (FCA) data collected by the Federal Reserve System permit such an analysis. The drawback of the cost ratio approach is that it cannot control for differences in banks' product mix or in the input prices banks face, which influence bank costs, and it cannot give an overall measure of efficiency. Also, the FCA program is voluntary and the sample is skewed toward smaller banks. And a bank's allocation of cost into various lines of business may require some arbitrary division of fixed or shared costs. Digitized for10 FRASER JANUARY/FEBRUARY 1994 Cost frontier analysis does not have to as sume data are measured without error. In stead, a bank is labeled as inefficient if: (1) its costs are higher than the costs predicted for an efficient bank producing the same outputs and facing the same input prices and (2) the differ ence cannot be explained by statistical noise, e.g., measurement error or luck.1 To obtain the 4 cost frontier, that is, the relationship between costs, outputs, and input prices for the efficient banks, statistical techniques (that is, regression analysis) are used to obtain the best fitting curve through the data, just as they are used to obtain the average practice cost function usu ally employed in the scale and scope economies studies. The difference is that the cost frontier indicates what, on average, it costs an efficient bank facing a particular set of input prices to produce a particular bundle of outputs, while the average practice function applies to all banks. A particular bank's cost will deviate from that predicted by the cost frontier for two reasons: first, there will be statistical noise, or unpredicted factors, that affected the bank's costs—either positively or negatively—com pared with an efficient bank's costs; second, the bank may not be X-efficient—hence its costs will be higher than those of efficient banks. The statistical technique used to obtain the cost frontier also provides information on these two types of deviations in the sample. The second deviation is always positive, since inefficient banks' costs are always higher than efficient banks' costs. This "one-sided" deviation can be used to obtain measures of any particular bank's inefficiency or the average level of inefficiency in the sample of banks. (As with the average practice cost function, no efficient bank is ex pected to lie precisely on the estimated cost frontier. Hence, the point estimate of ineffi ciency for these banks will be small but not 14Again, the banks in the sample are assumed to be using the same production technology in producing their outputs. FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? Loretta /. Mester zero.) Once the cost frontier is estimated, one estimates of inefficiency are higher from these can also estimate scale and scope economies for studies—on the order of 20 to 50 percent. These results suggest that there is substantial room banks operating efficiently.1 5 One drawback of cost frontier analysis com for improvement at the average bank in the pared with DEA is that it does require the United States, and the average bank will have researcher to make more assumptions about to cut costs considerably or will have to leave the form of the frontier and the errors; hence, it the industry via merger or failure as competi is less flexible. However, this is a less serious tive pressures increase. Is the same true of problem than DEA's inability to allow for any Third District banks? noise in the data.1 Therefore, I use frontier 6 analysis to analyze efficiency of banks in the EFFICIENCY OF THIRD DISTRICT BANKS I used the cost frontier approach to study the Third Federal Reserve District. Another poten tial problem with frontier analysis is that if the efficiency in 1992 of commercial banks operat researcher misspecifies the cost function to be ing in the Third Federal Reserve District, which estimated or omits factors that affect cost, this comprises the eastern two-thirds of Pennsylva may be attributed incorrectly to inefficiency. nia, the southern half of New Jersey, and the Current research is expanding on the method entire state of Delaware. Since I wanted to ology by trying to actually model the ineffi estimate the cost frontier of standard commer ciency rather than rely on deviations from the cial banks that are using the same production frontier to capture inefficiency. This has great technology, some banks were omitted from the 7 potential, since it would more readily indicate sample.1 The sample of 214 banks included all the causes of inefficiency. (SeeFaulhaber, 1993.) the Third District banks except the special pur The handful of frontier studies (including pose banks in Delaware (legislated under the stochastic econometric and thick frontier meth Financial Center Development Act and the odologies), in general, used data from the 1970s Consumer Credit Bank Act—thus, we excluded and 1980s, and have found X-inefficiency on Delaware's credit card banks), de novo banks the average of about 20 to 30 percent in banking (that is, banks less than five years old, which (see Evanoff and Israilevich, 1991). That is, have start-up costs that more mature banks do elimination of X-inefficiency at the average not have), and three very large banks (which bank could produce about a 20 to 30 percent very likely use different production techniques 8 cost savings, making this a much more serious than the other banks).1 The median asset size source of inefficiency than scale and scope inefficiency. Not surprisingly, since DEA at tributes any statistical noise to inefficiency, the 17Since efficiency is measured relative to the cost fron 15A more technical explanation of the frontier methodol ogy is contained in Mester (1994). 16Moreover, there are ways of relaxing some of the maintained assumptions of stochastic frontier analysis and achieving more flexibility, depending on the available data. For example, using panel data— that is, data from several periods (years, quarters, etc.) on the same sample of banks— allows some of the assumptions regarding the error struc ture to be relaxed. See Schmidt and Sickles (1984). tier, it is important that all banks in the sample have access to the same frontier; hence they should be using the same technology. (Whether one technology is better than another is a separate issue.) One advantage to restricting the sample to the Third District rather than using a U.S. sample is that banks in the Third District are likely to have more in com mon with each other, thus making it more likely they are using the same production technology. It should be remem bered that the results presented below apply only to the 1992 period. Since branching restrictions have only recently been eliminated in Pennsylvania—branching throughout the state became totally unrestricted only on March 4, 1990—more years of data were not included in the study. 11 BUSINESS REVIEW of banks included was $144 million, and the average asset size was $325 million. The intermediation approach was used to determine bank outputs and inputs. Three outputs were included: real estate loans, com mercial and industrial plus other loans, and loans to individuals. Each of these was mea sured by the average dollar volume that the bank held in 1992. These three outputs account for just about all of a bank's nonsecurities earning assets. The average volume of each of these three outputs at banks in the sample was about $120 million, $52 million, and $31 mil lion, respectively. Thus, about 60 percent of the average bank's loan portfolio is in real estate, about 25 percent is business loans, and the rest is loans to individuals. The inputs (whose prices are used to esti mate the cost frontier) are labor, physical capi tal, and borrowed money (including deposits, federal funds, and other borrowed money) used to fund the outputs. To account for the quality of the banks' outputs and bank risk (and so to avoid labeling as efficient banks that are not monitoring their loans), a bank's volume of nonperforming loans and the volume of its financial capital are included as arguments in the cost fu n c tio n .1 1 8 9 The volum e of nonperforming loans relative to the level of 18If the banks in the District are ordered by asset size, the sizes grow relatively smoothly from about $13 million to about $3.8 billion; then there is a jump to $7.8 billion, then to $9.3 billion, and then to $16 billion. Since there is empirical evidence that very large banks use a different production technology than other banks (e.g., findings of scale econo mies differ for small and large banks), and large banks also produce different outputs from small banks (e.g., they have more off-balance-sheet business), these three largest banks were not included in the sample. 19The translog functional form was assumed for the cost function; the two-sided error representing statistical noise was assumed to have a normal distribution; the one-sided error representing X-inefficiency was assumed to have a half-normal distribution. Interested readers may consult Mester (1994) for further details on the study's setup. Digitized 12 FRASER for JANUARY/FEBRUARY 1994 bank output is inversely related to quality: the higher the bank's nonperforming loans for a given volume of loans, the less resources the bank likely spent on monitoring its loan portfo lio.2 The higher the bank's level of financial 0 capital relative to the level of output, the lower the bank's probability of failure and so the bank's interest costs. Financial capital is also included because capital can be used as a fund ing source for loans. Scale and Scope Economies at Efficient Third District Banks. The estimated average cost frontier for Third District banks seems to be quite flat. The efficient bank producing the average level of each output and facing the average input prices is producing with con stant returns to scale. That is, a 1 percent increase in the level of all outputs would lead to about a 1 percent increase in costs. (See the Table. The first line of the table's top panel, Average Inefficiency Measures, shows the aver age bank's point estimate of the scale econo mies measure, indicating the percentage in crease in cost from a 1 percent increase in all outputs, holding quality and risk constant; it is statistically insignificant from one.) Moreover, over the entire size range of banks operating in the District, efficient banks are operating with constant returns to scale. The first line of the table's middle panel, Scale and Scope Economies over Different Sized Banks, shows the scale econo mies measures for the average efficient bank in each of four size categories. (Although the point estimates suggest decreasing average costs, the scale economies measures are suffi ciently close to one that a flat average cost curve cannot be ruled out statistically.) Therefore, there do not seem to be many cost efficiency 20Nonperforming loans are loans that are 30 or more days past due but still accruing interest plus loans that are not accruing interest. While the macroeconomy can affect nonperforming loans, the effect is felt equally across banks. It is the differences in nonperforming loans across banks that capture differences in quality across banks. FEDERAL RESERVE BANK OF PHILADELPHIA TABLE Average Inefficiency Measures Scale Economies3 Scope Economies1 5 X-Inefficiencyc 0.95% 0.37% 7.90% Scale and Scope Economies over Different Sized Banks Banks with Assets Under $72 Million (53 banks) Banks with Assets Between $72 Million and $144 Million (54 banks) Banks with Assets Between $144 Million and $280 Million (53 banks) Banks with Assets Over $280 million (54 banks) Scale Economies3 0.89% 0.92% 0.94% 0.99% Scope Economies1 5 0.006% 0.22% 0.50% 1.10% Bank-Specific X-Inefficiency Measuresd Range of X-Inefficiency over All Banks in Each Subsample Average X-Inefficiency over All Banks in Each Subsample6 Pennsylvania (182 banks) 2.94% to 19.15% 7.74% New Jersey (26 banks) 3.71% to 22.97% 9.34% 3.69% to 8.58% 6.32% Delaware (6 banks) aThe scale economies measure is (din C /dln y x )+(dln C /dln y2)+(dln C /dln y3)+(dln C /dln k)+(dln C /dln q) where C is the predicted cost of producing the average output bundle (in the specified bank size category) at the average input prices, y. is the volume of output i, k is the level of financial capital, and q is the volume of nonperforming loans. The measure indicates the percentage increase in costs from a 1 percent increase in each output level, holding risk and quality constant. Constant returns to scale is indicated if the measure is insignificantly different from one; decreasing returns to scale is indicated if the measure is significantly greater than one; increasing returns to scale is indicated if the measure is significantly less than one. None of the scale economies measures is significantly different from one, so there is no evidence of scale economies or diseconomies; that is, there are constant returns to scale. bThe scope economies measure is {[C(y1,y™y^4C(y1)y2,y3 C(y™ y2 n V 'fy3)]-C(yJ,y2,y3)|/C(y1 ,y2,y3) where y. is the volume of output i, yj’is the least amount of output i produced by any bank in the sample, and C(») is the predicted cost of producing an output bundle at the average input prices. The scope measure gives the percentage increase in cost of dividing the bank's products among three banks, each of which is relatively specialized in one of the three outputs. A statistically positive scope measure indicates there are economies of scope between the three outputs; a statistically negative scope measure indicates there are diseconomies of scope between the three products. None of the scope measures is significantly different from zero, so there is no evidence of scope economies or diseconomies. cThe X-inefficiency measure is significantly different from zero (at the 10 percent level). This measure is E(u.) where u. is the one-sided component of the composed error term in the frontier regression. See Mester (forthcoming). dThe bank-specific inefficiency measure is E(u. Ie.) where u. is the positive component of the composed error term £. of the frontier regression. See Mester (forthcoming). degression http://fraser.stlouisfed.org/ results indicate that while the average point estimates differ across states, once bank characteristics are Federal Reservecontrolled for there is no statistical difference in inefficiency across states. Bank of St. Louis BUSINESS REVIEW gains to be made from Third District banks' changing their sizes, and these results are much like those obtained in studies using U.S. samples.2 1 The scope economies statistics give the per centage increase in cost if the bank's three outputs were divided up and produced in three banks, each of which is relatively specialized in one of the outputs.2 These measures indicate 2 that there is no evidence of economies or diseconomies of scope at the average efficient bank in the sample nor at banks in different size categories, since the measures are statistically insignificant from zero. (See the Table.) Thus, there do not appear to be many cost efficiency gains to be made by a bank's changing its loan mix (which for the typical bank in the sample is weighted toward real estate loans). X-Inefficiency at Third District Banks. The cost frontier technique allows one to estimate the average level of X-inefficiency for the entire sample of banks and also bank-specific levels of inefficiency. The bank-specific measures can then be averaged by state to indicate the aver age level of inefficiency of banks in each of the three states in the District. As shown in the table's top panel, Average Inefficiency Measures, and in the bottom panel, Bank-Specific X-lnefficiency Measures, X-inefficiency at banks in the Third District runs in the 6 to 9 percent range. In other words, given its particular output level and output mix, if the average bank were to use its inputs as efficiently as possible, it could reduce its production cost by roughly 6 to 9 percent. The average annual cost of output production at banks in the sample was about $12 million, so a 6 percent reduction in cost 21The average scale measure for the sample indicates that a 1 percent increase in output would yield a 0.95 percent increase in cost, which translates into a trivial potential increase in the average bank's return on assets. 22This is the "within-sample degree of scope economies" as defined in Mester (1991). 14 JANUARY/FEBRUARY 1994 could potentially add about $720,000 to bank profits, which, given the average bank's size of $325 million in assets, constitutes a potential increase of 0.2 percent in before-tax return on assets, or about 0.15 percent in after-tax ROA. This isn't a trivial amount, as the average bank in the District had an after-tax ROA of 1 percent in 1992. In competitive markets not all of this gain would be retained by the bank— the sav ings would be passed on to customers in the form of lower loan rates and higher deposit rates. R egard less of who receiv es the savings—banks or their customers—society gains, since the savings created by increased efficiency can be used for other productive purposes. Of course, not all banks are the "average" bank. The figure, Third District Inefficiency Distribution, indicates the number of banks in the sample that fall into different inefficiency ranges. As you can see, while the distribution is weighted in the 6 to 9 percent range, some banks are quite efficient but others show a good deal of inefficiency (as high as 23 percent). When compared with results of other studies using U.S. samples that found average X-inef ficiency on the order of 20 to 30 percent, Third District banks seem to be performing better. It is difficult to determine whether this is a statis tically significant difference, however. It might just reflect that the Third District study is based on more recent data, or it might be because banks in the U.S. samples are more diverse, making efficiency measurement more difficult.2 3 In any case, as with U.S. banks in general, it appears that many Third District banks have room for improvement. Characteristics of Inefficient Banks. Ulti mately, we'd like to be able to say what banks can do to increase their efficiency. For each bank in the sample, the cost frontier analysis 23It might also be because Third District banks use a different production technology than other U.S. banks. FEDERAL RESERVE BANK OF PHILADELPHIA Loretta ]. Mester How Efficient Are Third District Banks? FIGURE Third District Inefficiency Distribution (214 Banks in the Sample) Number of Banks in Inefficiency Range 100 Inefficiency Range provides a point estimate of its level of Xinefficiency. Perhaps the best way to determine what banks should do to raise efficiency is to go on site to the banks that are identified as most efficient in the study and see what they are doing differently from the banks that are least efficient. A simpler first step is to see if there are any aspects of the banks that seem to be related to their degree of inefficiency. (Of course, a relationship need not imply causality. That is, we are not saying these characteristics cause 24Another reason to interpret the results as providing information on correlation only instead of causality is that there may be some endogeneity, since the characteristics are for the same period as the inefficiency measures. Causality may run from inefficiency to the characteristics instead of the other way around. For example, inefficient firms may choose to invest in real estate rather than investing in real estate leading to inefficiency. inefficiency, only that they seem to be more prevalent in inefficient banks.24) Simple corre lations between the inefficiency measures and characteristics of the banks can be calculated, and the inefficiency measures can be regressed on bank characteristics to get an idea of how the inefficient and efficient banks in the sample differ.2 * 5 25The regression involved estimating a logistic equation relating the bank-specific inefficiency measure to the fol lowing regressors: charter type (federal vs. state), holding company status (member of a holding company or not), member of the Federal Reserve System or not, number of branches, total assets, location in Pennsylvania, location in New Jersey, location in Delaware, total qualifying capital/ assets, return on assets, volume of uninsured deposits/total deposits, construction and land development loans/total loans, real estate loans/total loans, loans to individuals/ total loans, and year opened. See Mester (1994) for further details. 15 BUSINESS REVIEW The simple correlation, which does not hold constant the other characteristics, and the re gression results, which do hold constant other characteristics of the bank, indicate that ineffi cient banks in the District tend to be younger than more efficient banks. This might be evi dence that banking involves "learning by do ing," or it might indicate that more efficient banks are more likely to survive. (Recall that the de novo banks were not included in the sample, so the result probably doesn't merely reflect younger banks' higher start-up costs, for example, the costs of establishing customer relationships.) Even though the point estimates show dif ferences in inefficiency among banks in the three states, once other bank characteristics are controlled for, there is no statistically signifi cant difference in inefficiency across the states.2 6 Similarly, there is no evidence that larger banks are more or less X-efficient than smaller banks. This result, coupled with our results on scale economies, suggests that banks of all sizes in our District can be equally competitive when it comes to cost efficiency. Among the statistically significant relation ships, one of the more interesting is the nega tive relationship between inefficiency and the 26The simple correlation coefficient indicates that being located in New Jersey is significantly related to being inef ficient, but this is because the New Jersey banks in the sample tend to have lower capital ratios than Pennsylvania and Delaware banks in the sample. Once capital ratio is controlled for (as in the regression), being located in New Jersey is not significantly related to inefficiency. 27There are a few other statistically significant relation ships. For example, inefficient banks tend to have a higher percentage of their loans in construction and land develop ment; national banks appear to be less efficient than state banks that are members of the Federal Reserve System but seem to have the same level of efficiency as state nonmember banks. (Note: all nationally chartered banks are Fed mem ber banks, but their primary regulator is the Office of the Comptroller of the Currency, not the Fed.) 16 JANUARY/FEBRUARY 1994 capital-asset ratio.27 This result should not be interpreted as saying that if a bank increases its capital-asset ratio then its efficiency will in crease. But it may be an indication that higher capital ratios may prevent "moral hazard." As is often cited in discussions of the thrift crisis, as an institution's capital level decreases it has an increasing incentive to "bet the bank," since it stands to gain if the risk pays off and tends to lose only the amount of capital it has invested in the bank if the bet loses. Similarly, the managers of banks with lower capital levels might have more of an incentive to engage in perk-taking, and they face less shareholder scrutiny than banks with higher capital ratios. (If the owners' stake, that is, capital, is low, owners have less incentive to make sure the bank is run efficiently.28) Therefore, higher bank capital may not only provide a cushion for the deposit insurance fund, it might also pro vide appropriate incentives to bank managers to avoid waste. The capital-asset ratio might also be significantly related to inefficiency be cause inefficient banks have lower profits, which might lead to lower capital-asset ratios in the future.2 9 CONCLUSION Banks in the Third District appear to be operating at cost-efficient output sizes and prod uct mixes, but there appears to be a significant level of X-inefficiency at our banks. Some banks apparently are not using their labor, plant and equipment, and funds in the most efficient way possible, and case studies that focus on the more efficient banks in the District 28Mester (1990) discusses the incentive effects of bank capital in mitigating bank risk-taking. 29But this is probably not the entire reason, since the relationship between capital assets and inefficiency holds even when return-on-assets is held constant, and while return-on-assets and capital assets are correlated, they are not collinear. FEDERAL RESERVE BANK OF PHILADELPHIA How Efficient Are Third District Banks? might shed light on how greater efficiency can be achieved. Theoretical advances may enable us to better identify the sources of the ineffi ciencies and verify that measured differences in inefficiency are true differences and do not result just from omitting factors that affect cost or misspecifying the cost function. In terms of coping with the increased com petitive pressures, inefficientbanks in the Third Loretta ]. Mester District have more to fear from banks that are efficient producers than from banks that are producing a particular output volume or prod uct mix. There is less to be gained in terms of cost savings from changing output size or mix than from using inputs more cost-effectively. Inefficient banks will have to get costs under control or else be prepared to be driven from an increasingly competitive marketplace. Berger, Allen, and David Humphrey. "Measurement and Efficiency Issues in Commercial Banking," in Z. Griliches, ed., Measurement Issues in the Service Sectors. Chicago: National Bureau of Economic Research and Chicago University Press, 1992a. Berger, Allen, and David Humphrey. "Megamergers in Banking and the Use of Cost Efficiency as an Antitrust Defense," Antitrust Bulletin, 37 (Fall 1992b), pp. 541-600. Calem, Paul S. and Loretta J. Mester. "Search, Switching Costs, and the Stickiness of Credit Card Interest Rates," Working Paper 92-24/R, Federal Reserve Bank of Philadelphia, revised January 1993. Calem, Paul S. "The Proconsumer Argument for Interstate Branching," this Business Review (May/June 1993), pp. 15-29. Clark, Jeffrey A. "Economies of Scale and Scope at Depository Financial Institutions: A Review of the Literature," Economic Review, Federal Reserve Bank of Kansas City (September/October 1988), pp. 16-33. Evanoff, Douglas D., and Philip R. Israilevich. "Productive Efficiency in Banking," Economic Perspectives, Federal Reserve Bank of Chicago, (July/August 1991), pp. 11-32. Faulhaber, Gerald R. "Profitability and Bank Size: An Empirical Analysis," The Wharton School, University of Pennsylvania, mimeo, 1993. Hughes, Joseph P., and Loretta J. Mester. "A Quality and Risk-Adjusted Cost Function for Banks: Evidence on the 'Too-Big-To-FaiT Doctrine," Journal of Productivity Analysis, 4 (September 1993), pp. 293-315. Humphrey, David. "Costs and Scale Economies in Bank Intermediation," in R. Aspinwall and R. Eisenbeis, eds., Handbookfor Banking Strategy. New York: Wiley and Sons, 1985. Kraus, James R. "The Whole World Is Watching Asset Quality, Rate of Return," American Banker, July 29,1993, p. 2A. Mester, Loretta J. "Efficient Production of Financial Services: Scale and Scope Economies," this Business Review (January/February 1987), pp. 15-25. 17 SUGGESTED READINGS BUSINESS REVIEW JANUARY/FEBRUARY 1994 Mester, Loretta J. "Owners Versus Managers: Who Controls the Bank?" this Business Review (May/June 1989), pp. 13-23. Mester, Loretta J. "Curing Our Ailing Deposit-Insurance System," this Business Review (September/October 1990), pp. 13-24. Mester, Loretta J. "Agency Costs Among Savings and Loans," Journal of Financial Interme diation, 3 (June 1991), pp. 257-78. Mester, Loretta J. "Traditional and Nontraditional Banking: An Information-Theoretic Approach," Journal of Banking and Finance, 16 (June 1992), pp. 545-66. Mester, Loretta J. "Efficiency of Banks in the Third Federal Reserve District," Federal Reserve Bank of Philadelphia Working Paper 94-1 (1994). O'Hara, Terrence. "Interstate Branching Yet to Sound Alarms," American Banker, August 2, 1993. Schmidt, Peter, and Robin C. Sickles. "Production Frontiers and Panel Data," Journal of Business and Economic Statistics, 2 (October 1984), pp. 367-74. 18 FEDERAL RESERVE BANK OF PHILADELPHIA New Indexes Track the State Of the States Theodore M. Crone* H a v e recessions lasted longer in Pennsylvania than in the nation? When did the most recent recession begin in New Jersey? Did Delaware avoid the recessions of the early 1980s altogether? Questions about how busi ness cycles differ from state to state are raised frequently in the popular press and in business commentaries. The answers are seldom clear, but the questions are not idle ones. Some industries, such as construction and retail trade, *Ted Crone is Assistant Vice President in charge of the Regional Economics section in the Philadelphia Fed's Re search Department. Ted would like to thank James Stock and Mark Watson for a copy of the programs that estimate their new composite index of coincident indicators. He also thanks Keith Sill for help in adjusting Stock and Watson's programs to calculate the new state indexes. are particularly sensitive to the local business cycle. Since people tend to live close to their jobs and shop close to where they live, sales of new homes, cars, and many consumer items depend on the prospects for jobs and income in the local area. If the local economy is weaken ing, these prospects are poor; if the economy is strengthening, the prospects are better. So knowledge about where the region is in the business cycle can be critical for managers in many businesses. But economic data are often ambiguous and sometimes contradictory; one indicator may be showing improvement while another shows decline. For example, the unem ployment rate may be up at the same time job levels are increasing. Composite indexes, con structed from a number of individual indica tors, can help clear up the ambiguity. There are 19 BUSINESS REVIEW JANUARY/FEBRUARY 1994 by a recovery phase in which economic mea sures return to their previous peaks and then an expansion phase in which the measures reach new peaks. The description by Bums and Mitchell mentions three criteria for dating the various phases of the business cycle. The con tractions and expansions must be broad-based, that is, they must occur in many sectors and be reflected in several indicators (diffusion). They must last a sufficient length of time (duration). And the change from peak to trough or from COMPOSITE INDEXES CAN HELP TRACK trough to peak must be sufficiently large (ampli tude). All three criteria must be satisfied in BUSINESS CYCLES Between 1970 and 1990 real output in the order to define a contraction or expansion. A U.S. grew at an average annual rate of 2.7 sharp decline in one sector, such as agriculture, percent, and employment increased 2.2 percent would not qualify as a recession if it did not spill a year. But output and employment fluctuated over into other sectors of the economy; the widely around these trends as the economy decline would not be broad enough. On the went through several business cycles. Eco other hand, a broad-based decline that was nomic trends vary from region to region, but all very brief, one quarter, for example, probably regions are affected by national business cycles, would not qualify as a recession; it would be too and some regions have exhibited cycles of their short. Likewise, two quarters of 0.1 percent decline in output might not qualify as a reces own. Almost 50 years ago, Arthur Burns and sion; the decline would not be deep enough. Since business cycles are broad-based, they Wesley Mitchell fashioned the commonly ac tend to generate their own momentum. Down cepted description of a business cycle: turns in the economy can be set in motion by a Business cycles are a type o f fluctuation found variety of factors, such as a sharp increase in the in the aggregate economic activity o f nations price of a major resource like oil or a sudden that organize their work mainly in business large reduction in government spending. Once enterprises: a cycle consists o f expansions a general downturn begins, firms begin to lay occurring at about the same time in many off workers. This loss of jobs as well as the economic activities,followed by similarly gen uncertainty among people who are still em eral recessions, contractions, and revivals ployed leads consumers to cancel or postpone which merge into the expansion phase o f the purchases, which results in more layoffs. Not next cycle...; in duration business cycles vary all sectors of the economy are equally vulner from more than one year to ten or twelve years; able to a downturn; consumers may be reluc they are not divisible into shorter cycles . . . tant to delay seeing a doctor if they are ill, but [that exhibit swings in economic activity o f they might readily put off the purchase of a new similar] amplitudes} A complete cycle from one peak to the next car. In general, manufacturing industries are consists of a recession or contraction followed1 more sensitive to business cycles than service * industries. A downward spiral in the economy might be halted by a change in consumer expec 1Arthur F. Bums and Wesley C. Mitchell, Measuring tations that raises confidence, by an increase in Business Cycles (National Bureau of Economic Research, disposable income through a reduction in taxes, 1946). commonly accepted composite indexes for the national economy, such as the index of leading indicators, but composite indexes are not readily available at the regional level, making it more difficult to track regional business cycles. This article introduces new composite indexes for the three states in the Third Federal Reserve District that make use of statistical techniques previously used for national indexes but not regional ones. Digitized 20 FRASER for FEDERAL RESERVE BANK OF PHILADELPHIA New Indexes Track the State of the States Theodore M. Crone or by a rise in government spending for goods have been 18 business-cycle turning points in and services— any of which would increase the U.S. economy. With four exceptions, the demand. When this increased demand be highest and lowest levels of the Commerce comes broad enough it creates its own momen Department's index during each cycle have tum toward further expansion. Thus, knowl been within three months of the official dates of edge of whether the economy is entering a the business-cycle peaks and troughs.3 Al recession or beginning a recovery is important though the NBER dating committee considers the coincident index when it sets the dates for to local businessmen. Dating business cycles using the Burns and business cycles, it is not obligated to set the Mitchell criteria is not always straightforward; dates at or near the turning points of the index. it involves some personal judgment. In prac Therefore, the close correspondence between tice, the official dating of recessions and expan the o fficia l dates and the C om m erce sions is done by the Business Cycle Dating Department's index suggests that the index is Committee of the National Bureau of Economic coincident with the business cycle. The Commerce Department's Composite Research (NBER), which is composed of pro fessional economists. To avoid any possibility Index of Coincident Indicators is constructed of political considerations in setting these dates, from four monthly data series—the number of none of these economists is a government offi jobs in nonagricultural establishments, personal cial. The committee considers a number of income (minus transfer payments) adjusted for economic indicators in dating business cycles. inflation, the index of industrial production, Composite indexes, however, play a special and manufacturing and trade sales adjusted for role in these decisions because they combine inflation. Month-to-month percent changes are information from several sources to indicate calculated for each of these series, and the the general state of the economy. They include changes are standardized based on the longnot only data for the overall economy, such as run average absolute monthly change in the employment and personal income, but also series. For example, the average absolute per data from individual sectors, such as retail centage change in monthly employment be sales or industrial production. tween 1948 and 1985 was 0.32 percent. Thus, if The Department of Commerce publishes the change in nonfarm employment were 0.64 three com posite indexes for the national percent this month, the standardized change economy—indexes of leading, lagging, and co for this indicator would be 2 (i.e., 0.64 / 0.32). incident indicators.2 Of these three, the com A preliminary coincident index is formed based posite index of coincident indicators is the most on the average of the standardized changes in important for dating business cycles. A good the components that make up the index. To index of coincident indicators should decline at obtain the Department's official composite in or near the beginning of recessions and should dex, this preliminary index is adjusted to grow rise at or near the end. In the last 45 years there over time at the same rate as real gross national 2This article was prepared before the most recent revisions of the Commerce Department's composite indexes. All references in this article refer to the unrevised indexes. SeeGeorgeR. Green and Barry A. Beckman, "Business Cycle Indicators: Upcoming Revision of the Composite Indexes," Survey o f Current Business, Vol. 73,10 (Oct. 1993), pp. 44-51. 3A s the name suggests, an index of leading indicators should peak several months before the economy goes into recession and should reach its cyclical low before the recession ends. The timing should be reversed for an index of lagging indicators. The leads and lags in the Commerce Department's leading and lagging indexes have varied from as long as 23 months to as short as 1 month. 21 BUSINESS REVIEW product and is set to 100 in 1982.4 Except for the adjustment to account for differences in the average monthly changes in the four indica tors, each indicator is given the same weight in forming the composite index. But some indica tors like the total number of jobs may better reflect the overall state of the economy than other indicators like manufacturing and trade sales. Thus, while the Commerce Department's index has tracked national business cycles very well, it has been criticized for not being derived from a formal m athem atical or statistical model.5 In order to support the theory of business cycles and aid in the dating of recessions and expansions, James Stock of Harvard University and Mark Watson of Northwestern University have constructed a new index of coincident indicators.6Using time-series econometric tech niques, they formalized the notion that the business cycle is best measured by the common movements across several economic data se ries. Each monthly indicator is thought of as having two components. The first is the general "state of the economy," which affects all the monthly indicators. It is not observed directly but only in the common movement of the indi cators that are observed. The second compo nent is an idiosyncratic element that might cause any one indicator to move in ways not associated with the general state of the economy. Stock and Watson's coincident index is an esti 4See "Composite Indexes of Leading, Coincident, and Lagging Indicators," Survey o f Current Business, U.S. Depart ment of Commerce, November 1987. 5The entire attempt to define business cycles was criti cized from the beginning as an exercise in measurement without theory. See Tjalling C. Koopmans, "Measurement Without Theory," Review ofEconomics and S tatistics, 29 (1947), pp. 161-72. 6James H. Stock and Mark W. Watson, "New Indexes of Coincident and Leading Economic Indicators," NBER Macroeconomics Annual (1989), pp.351-94. 22 JANUARY/FEBRUARY 1994 mate of the common component. The move ment of this unobserved state of the economy is reflected in varying degrees in each of the published monthly series used to estimate the composite index. Moreover, for some series, changes in the general economy could be re flected not only in the current month but also in succeeding months, and for other series, changes in the general economy could be foreshadowed in preceding months (see A Formal Model o f the New Coincident Index). In effect, the Stock and Watson index is a weighted average of current and past values of the individual indicators, with the weights determined by the degree of common movement in the indicators. In constructing their coincident index Stock and Watson used the same data series as the Department of Commerce, with one exception: they su bstitu ted em ployee h ou rs in nonagricultural establishments for the number of nonagricultural jobs because economic out put depends not only on how many people are working but also on how long they work. Stock and Watson's new index is available from 1959, and over that period it has coincided with the official business cycles even more closely than has the Commerce Department's Index of Co incident Indicators. The cyclical highs and lows in the Stock and Watson index coincide exactly with the official business-cycle turning points except in 1969 when the new index peaks two months prior to the official turning point.7 CONSTRUCTING STATE INDEXES The success of the Stock and Watson method in constructing a national coincident index that tracks the official business cycles so closely suggests that this method could be used suc cessfully to construct an index for state econo mies. But the construction of a comparable 7Of course, in developing their index Stock and Watson were attempting to trace the official business cycles prior to 1990, and the NBER dating committee had the new Stock and Watson Index when it dated the most recent recession. FEDERAL RESERVE BANK OF PHILADELPHIA New Indexes Track the State of the States Z 0) rz £ 5O Theodore M. Crone The basic notion that a change in a monthly indicator reflects a change in the underlying state of the economy is captured in the following equation: AIf = a + b ASt+ut (1) where: AIt = the change in the observed monthly indicator between time t-1 and time t, and ASt= the change in the unobserved state of the economy between time t-1 and time t.a Since the purpose of this model is to form a composite index, this equation is applied to a number of monthly indicators. For example, Stock and Watson use four monthly indicators so there are four equations similar to equation (1) in their model. The coefficients (a and b) will vary with each equation, but the unobserved variable (ASt) is the same. In addition, the error term inequation (1) and the unobserved variable are assumed to follow an autoregressive process, so that Ut = gl Ut-l + g 2 Ut-2 + et ( 2) and ASt = c + f,AStl + f2ASt 2 + zt (3) where et and zt are error terms. Equations (2) and (3) are the transition equations in the system. This system of equations (1) through (3) can be estimated using maximum likelihood techniques to produce an estimate of the change in the unobserved state of the economy (ASt).b If we then index the unobserved variable St to equal 100 at some point in time, we can construct a time-series of the so-called "state of the economy," or a coincident index.c aIf the monthly indicator also reflects prior changes in the state of the economy, the estimating equation becomes AI = a + bQ + bj AS( ] . . . + u . If the monthly indicator partially foreshadows a change in the AS general state of the economy, the lagged values of the unobserved state of the economy are replaced by leads. bIn the actual estimating equations, Stock and Watson use the log difference of the monthly indicators. The change in the log of the monthly indicator is normalized by subtracting the historical mean and dividing by the standard deviation. Thus the constants a and c do not have to be estimated, and the unobserved variable that is estimated is the normalized change in the log of S . cStock and Watson set their national index at 100 in July 1967, and we set our state indexes at 100 in July 1987. state index is not a simple matter of estimating Stock and Watson's model using state data. The monthly indicators used by Stock and Watson are not available at the state level. Moreover, there is no direct way to determine whether a composite index using other indica tors at the state level would coincide with the business cycle because there are no official dates for state business cycles. Indeed, this was one reason for developing state indexes. To address the problem of finding an appropriate set of indicators to construct state indexes we identified a set of monthly indicators that are available at both the national and state levels. We selected those variables that were useful in dating national business cycles and assumed they would also be useful in identifying cycles in the state economies. 23 BUSINESS REVIEW JANUARY/FEBRUARY 1994 This indirect method resulted in identifying four variables to be used in our state indexes of monthly indicators— the total number of jobs in nonagricultural establishments, real retail sales, average weekly hours in manufacturing, and the unemployment rate.8 These variables differ somewhat from those used in other national ind exes. The to tal n um ber of jo bs in nonagricultural firms is used in the Commerce Department's Index of Coincident Indicators and was used in an earlier version of the Stock and Watson index.9 The sales data used in our indexes are less comprehensive than those used by the Commerce Department and by Stock 8The first three variables enter our model in log differ ence form. The unemployment rate enters in first difference form and is modeled to reflect the current value and three lags of the variable that reflects the state of the economy. 9James H. Stock and Mark W. Watson, "A Probability Model of the Coincident Economic Indicators," in Geoffrey Moore and K. Lahiri, eds., The Leading Economic Indicators: New Approaches and Forecasting Records (Cambridge Univer sity Press, 1990). and Watson. Both use a series that includes sales by manufacturers and wholesalers as well as by retailers. Two of the variables we selected to construct our indexes, average hours in manufacturing and the unemployment rate, have not traditionally been counted among the coincident indicators. But we included them in our index because doing so improved the cor respondence between the index and the official business cycles, compared to an index using only employment and retail sales. Using these four variables we developed a national index and examined how closely it coincides with the official dates of national business cycles and with other composite in dexes for the nation. Based on data since 1972, the pattern of the new national index follows closely the p attern of the C om m erce Department's coincident index and Stock and Watson's coincident index (Figure l).1 * There 0 10We started the index in 1972 because some of the data series used in the index are not available at the state level prior to 1972. FIGURE 1 Coincident Indicators- Index duly 1987=100) -u.s. 120 110 New National Index 1Stock & Watson 1Commerce Departme 100 90 80 70 The shaded areas represent the official recessions as determined by the NBER Dating Committee. New Indexes Track the State of the States Theodore M. Crone have been four national business cycles since 1972. Only the Stock and Watson index coin cides precisely with the official peaks and troughs of all of them. But the new national index developed with data series that are also available at the state level traces the four na tional business cycles closely. With two excep tions the peaks and troughs of this new com posite index are within one month of the official peaks and troughs of the U.S. business cycles since the early 1970s (Table l).1 The Commerce 1 Department's Composite Index of Coincident Indicators was also off by several months at the same two turning points. Thus, the timing of the new index compares favorably with the 11The two exceptions are the peak preceding the 1980 recession when the index led the economy by seven months and the trough of the most recent recession when the index lagged the economy by 15 months. Prior to the 1980 reces sion, the Stock and W atson index, the Commerce Department's index, and the new national index were basi cally flat for almost a year. Although the new index and the Commerce Department's index peaked several months be fore the beginning of that recession, they changed very little in the intervening months. After the official end of the most recent recession in March 1991, the cyclical lows for the Commerce Department's index and the new national index lagged by several months. The steep declines in the two indexes ended, however, about the same time as the official end of the recession, and the two indexes improved tempo rarily shortly after the official end of the recession. timing of the Commerce Department's index, and it can be considered a coincident index.1 2 Using the same monthly indicators as in the new national index, we constructed coincident indexes for each of the three states in the Third Federal Reserve District — Pennsylvania, New Jersey, and Delaware (see New National and State Indexes). Since retail sales data are not available for Delaware, that state's index in cluded only three of the four indicators.1 These 3 12The average monthly increase in the new national index between 1972 and 1992 was 0.13 percent, compared with 0.16 percent for the Commerce Department's index and the 0.19 percent for the Stock and Watson index. The variance in the monthly change for the new index is also smaller than the variance for the other two indexes. The correlation between monthly changes in the new index and the Commerce Department's index is 0.54, and the correla tion between the new index and the Stock and Watson index is 0.55. Both correlation coefficients are significantly differ ent from 0 and from 1 at the 0.01 level. The correlation between the Commerce Department and the Stock and Watson indexes is considerably higher at 0.93, because with one minor exception these two indexes are constructed from the same monthly indicators. 13A national index constructed from the three variables used in the Delaware index tracks the national business cycles slightly less accurately than the new national index constructed from all four variables. In some cases, e.g., the 1981-82 recession, the timing of the peaks and troughs of the two national indexes are identical. TABLE 1 Leads and Lags of the New National Index at Business Cycle Peaks and Troughs (leads and lags in months) BUSINESS CYCLE PEAKS November 1973 January 1980 July 1981 July 1990 lead (+)/ lag (-) 0 +7 0 +1 BUSINESS CYCLE TROUGHS March 1975 July 1980 November 1982 March 1991 lead (+)/ lag (-) -1 0 -1 -15 25 JANUARY/FEBRUARY 1994 BUSINESS REVIEW New National and State Indexes Except for Delaware there are four measurement equations in each system used to estimate the new national and state indexes: (1) Aempt=Pe ASt + ute (2) Ahrst = PhAS, + uft (3) Arst = pr AS, + (4) AURt = Pu AS, + Pu AS,, + pu AS, 2 + Pu ASt.3 + u,u 0 l 2 3 where Aemp = the standardized change in the log of nonfarm employment Ahrs = the standardized change in the log of average hours worked in manufacturing Ars = the standardized change in the log of real retail sales AUR = the standardized change in the unemployment rate. Since retail sales are not available for Delaware, equation (3) is omitted in the system of equations for the Delaware index. Lagged values of the unobserved state of the economy are entered in the unemployment rate equation because including the lags produced a national index that coincided better with the official NBER recession dates. Moreover, the unemployment rate is often a lagging indicator reflecting the state of the economy in previous months. The estimated coefficients for each of the four systems is given in the following table: Estimates of Coefficients Used to Construct Indexes of Coincident Indicators US INDEX PA INDEX NJ INDEX DE INDEX 0.715 (.051) 0.530 (.081) 0.823 (.065) 0.701 (.134) 0.175 (.032) 0.175 (.041) 0.159 (.050) 0.185 (.053) 0.156 (.026) 0.128 (.034) 0.046 (.049) -0.428 (.052) -0.213 (.052) 0.033 (.045) 0.026 (.045) -0.044 (.092) -0.240 (.102) -0.161 (.103) 0.217 (.110) -0.202 (.058) -0.102 (.058) -0.003 (.061) -0.010 (.055) EMPLOYMENT EQ Pe HOURS EQ Ph RETAIL SALES EQ Pr UNEMPLOYMENT EQ P U0 Pul Pua Pu3 -0.637 (.130) -0.136 (.088) 0.071 (.089) -0.010 (.063) ( ) = standard error of the estimate 26 FEDERAL RESERVE BANK OF PHILADELPHIA Theodore M. Crone Neiv Indexes Track the State of the States models produced estimates of an unobserved "state of the economy," or a coincident index, for each of the three states. BUSINESS CYCLES IN THE STATES The coincident indexes for the three states in the Third Federal Reserve District define busi ness cycles that correspond generally to the four national business cycles since 1972. But the cycles in each state have differed in their timing and duration. These differences can be seen by comparing the peaks and troughs of the state indexes with the official NBER dates and with the peaks and troughs of the new national index (Figures 2 through 4). Since there are clear differences between state and national business cycles, we need to apply some crite rion to the new indexes to identify recessions and expansions at the state level. The experi ence of the NBER dating committee illustrates that there is no simple rule that will always identify peaks and troughs in the business cycle, but there should be some minimum de cline in the index in order to characterize a given period as a recession. We found that a cumulative decline four times the average ab solute monthly change in the index clearly defined four recessions in the new national index since 1972, and these recessions corre sponded closely with the four officially recog nized national recessions over that time period. We used the same rule of thumb to identify recessions at the state level. The peak of the cycle can be dated by the high point in the index just prior to the cumulative decline. Likewise, the trough of a cycle can be dated by the low point in the index prior to a cumulative increase that is four times the average absolute monthly change.1 This identification of recessions at the 4 state level allows us to compare cycles in the Third District states to national cycles. The peaks and troughs of the state indexes are shown in Table 2. 14Other simple rules could be used to date the beginning and end of a recession, such as three or four consecutive decreases or increases in the index. While the use of such consecutive decrease or increase rules would move the peak or trough closer to the NBER date for some recessions, for other recessions they would move the peak or trough further away from the official dates. These rules were not clearly superior to using the absolute high point and low point of the index as the business cycle turning points, and they have no compelling theoretical justification. TABLE 2 Peaks and Troughs of State Indexes PA INDEX NJ INDEX DE INDEX PEAK TROUGH November 1973 May 1975 November 1973 May 1975 February 1973 April 1975 PEAK TROUGH June 1979 September 1980 February 1980 July 1980 February 1980 April 1980 PEAK TROUGH March 1981 February 1983 September 1981 November 1982 July 1981 January 1982 PEAK TROUGH March 1990 July 1991 March 1989 September 1992 June 1990 April 1991 27 BUSINESS REVIEW JANUARY/FEBRUARY 1994 Pennsylvania's economy comprises some what less than 5 percent of the U.S. economy as measured by the number of jobs in the state and by gross product. In terms of the mix of industries, the cyclically sensitive manufactur ing sector represents a larger percentage of the Pennsylvania economy than it does of the na tional economy, so one might expect Pennsyl vania to suffer more recessions or longer reces sions than the nation.1 While there have not 5 been more recessions in the state, recessions have generally lasted longer in Pennsylvania than in the nation (Figure 2). In every recession since 1972 the new coincident index for Penn sylvania has recorded a longer downturn than indicated by the official dates for the national 15Over the period 1972 to 1992 manufacturing employ ment has averaged 25 percent of Pennsylvania's total em ployment but only 21 percent of U.S. employment. recession. And except for the last recession, the declines in the Pennsylvania index have also lasted longer than the declines in the compa rable national index. And generally recoveries in Pennsylvania have been less vigorous than in the nation as a whole. The current recovery is a striking example. At the end of the 1990-91 recession the Pennsylvania index technically reached its cyclical low 11 months before the new national index, but the state's index was little changed for more than a year after reach ing that low point and was not signaling a recovery. The index reflected the popular im pression of a lingering recession in the state. New Jersey's economy represents slightly more than 3 percent of the U.S. economy in terms of jobs and gross product. The structure of the New Jersey economy has changed over the past 20 years from a greater than average dependence on manufacturing to a less than average dependence. Financial and business FIGURE 2 Index of Monthly Indicators— PA Index (July 1987=100) 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 The shaded areas represent the official recessions as determined by the NBER Dating Committee. The U.S. index is the new national index constructed from variables also available at the state level. 28 FEDERAL RESERVE BANK OF PHILADELPHIA Theodore M. Crone New Indexes Track the State of the States services have become a more important part of the state's economy. According to the new coincident index for New Jersey, some reces sions in the state have been longer than the U.S. average (1973-75,1990-91), and some have been shorter (1980, 1981-82). That pattern holds whether we measure national recessions by the official NBER dates or by the comparable na tional index (Figure 3). The most recent reces sion in New Jersey has been especially pro longed in part because this recession affected the service-producing sectors more than previ ous ones. Based on the peak and trough in the state's coincident index, the latest recession in New Jersey lasted from early 1989 to mid-1992, much longer than it did in the other states of the Third Federal Reserve District. There were some temporary improvements in the index over this three-year period, but none of the improvements were strong enough to qualify as a recovery.1 6 Delaware's economy is less than one-half of 1 percent of the U.S. economy in terms of jobs in the state and in terms of gross state product. The state's economy is more heavily concen trated in manufacturing than the U.S. economy, a fact that should tend to make it more cyclical. The very rapid growth in financial and business services since the early 1980s, however, has helped the state weather the last few recessions relatively well. Clear counterparts to three of the four national recessions since 1972 are ap parent in the history of Delaware's new coinci dent index (Figure 4). The one national reces sion that has no counterpart in the Delaware index is the short-lived one in 1980. The decline from peak to trough in Delaware's monthly 16That is, the total increase in the index during these temporary improvements did not equal four times the aver age monthly change. FIGURE 3 Index of Monthly Indicators— NJ Index (July 1987=100) The shaded areas represent the official recessions as determined by the NBER Dating Committee. The U.S. index is the new national index constructed from variables also available at the state level. 29 BUSINESS REVIEW JANUARY/FEBRUARY 1994 FIGURE 4 Index of Monthly Indicators— DE Index (July 1987=100) Delaware U.S. The shaded areas represent the official recessions as determined by the NBER Dating Committee. The U.S. index is the new national index constructed from variables also available at the state level for many states. index in the first half of 1980 was very brief (two months), and the total change in the index in those two months was less than three times the average monthly change. Based on the normal criteria for national recessions the brief 1980 downturn in Delaware would not qualify as a recession. The subsequent recession in 1981-82 is clearly discernible in the Delaware index, which registered a cumulative decline well over four times the monthly average, but this recession ended much earlier in the state than in the nation. Passage of legislation in 1981 encouraging the establishment of credit card banks in the state aided Delaware's economy. While the new index indicates that Delaware weathered recessions much better than the nation in the 1980s, it also indicates that the state suffered more in the 1970s. The 1973-75 recession began much earlier in Delaware than in the nation or in the other two states in the 30 Third District. Moreover, the new coincident index suggests that Delaware suffered a local recession between February 1976 and February 1977—a downturn not matched at the national level. The state index declined a total of 5.9 percent, more than six times the average monthly change. The weakness in the state's economy was concentrated in the manufactur ing and construction industries. GETTING ANSWERS ABOUT STATE BUSINESS CYCLES The ability to construct a composite index of monthly indicators that are available at the state level helps answer some of the questions frequently raised about regional business cycles. The new coincident index for Pennsylvania indicates that recessions have generally lasted longer in that state than in the nation as a whole. A set of indexes for all 50 states would undoubt FEDERAL RESERVE BANK OF PHILADELPHIA New Indexes Track the State of the States edly uncover other states that tend to have longer recessions and help us identify some reasons. The new index for New Jersey indi cates that the most recent downturn in that state began more than a year before the onset of the national recession and continued for more than a year after the end of the national reces sion. The index confirms that this recession was much longer in New Jersey than in the other states in the region. The new index for Delaware indicates that the state suffered only one recession in the early 1980s, and that one was briefer than the national downturn. But Delaware suffered a more extended recession than the nation in the early 1970s. Moreover, Theodore M. Crone Delaware's index provides evidence of a local recession in the second half of the 1970s. The expansion of financial and business services in Delaware seems to have made the state's economy less cyclical. These new composite indexes for the states provide another tool to monitor and analyze a region's economy. They can help us compare the timing of business cycles among the states and between any state and the nation. A full set of such indexes for all the states would help answer even more questions about regional business cycles and the structure of regional economies. 31 January/February Leonard I. Nakamura, “Information Externalities: Why Lending May Sometimes Need a Jump Start' D. Keith Sill, “Predicting Stock-Market Volatility" March/April Laurence Ball, “What Causes Inflation?" Satyajit Chatterjee, “Leaning Against the Wind: Is There a Case for Seasonal Smoothing of Interest Rates? May/June Edward G. Boehne, “Testimony on the Third District Economy and Monetary Policy" Paul S. Calem, “The Proconsumer Argument for Interstate Branching" July/August Francis X. Diebolcf, “Are Long Expansions Followed by Short Contractions?" Shaghil Ahmed, "Does Money Affect Output?" September/October B. Douglas Bernheim and John Karl Scholz, “Do Americans Save Too Little?" Gerald A. Carlino, “Highways and Education: The Road to Productivity?" November/December Dean Croushore, “Introducing: The Survey of Professional Forecasters" Laurence Ball, “How Costly Is Disinflation? The Historical Evidence" FEDERAL RESERVE BANK OF PHILADELPHIA Business Review Ten Independence Mall, Philadelphia, PA 19106-1574