The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 30, No. 2 ECONOMIC REVIEW 1994 Quarter 2 U.S. Banking Sector Trends: Assessing Disparities in Industry Performance 2 by Katherine A. Samolyk Competition for Scarce Inputs: The Case of Airport Takeoff and Landing Slots 18 by Ian Gale Regional Wage Convergence and Divergence: Adjusting Wages for Cost-of-Living Differences by Randall W. Eberts and Mark E. Schweitzer FEDERAL RESERVE BANK OF CLEVELAND 26 B U E E MI C REVIEW 1994 Quarter 2 Vol. 30, No. 2 U.S. Banking Sector Trends: Assessing Disparities in Industry Performance 2 by Katherine A. Samolyk While the past decade appears to have been a difficult time for the U.S. banking sector, performance within the industry varied widely. Using statelevel data, the author investigates the extent to which variations in banking conditions were associated with differences in bank size and holding com pany relationships. Controlling for local economic factors, very large banks had more problems with loan quality and poor profitability over the period than did smaller banks; the results, however, do not indicate an emerging relationship between bank size and bank performance. At the same time, smaller banks that affiliate with larger organizations in the form of holding companies appear to benefit from the relationships. Competition for Scarce Inputs: The Case of Airport Takeoff and Landing Slots 18 Economic Review is published quarterly by the Research Depart ment of the Federal Reserve Bank of Cleveland. Copies of the Review are available through our Corporate Communications and Community Affairs Department. Call 1-800-5433489, then immediately key in 1-5-3 on your touch-tone phone to reach the publication request option. If you prefer to fax your order, the number is 216-579-2477. Coordinating Economist: William T. Gavin Advisory Board: Ian Gale Jagadeesh Gokhale Joseph G. Haubrich by Ian Gale Since 1986, airline carriers have exercised the right to buy and sell takeoff and landing slots at airports. Questions remain, however, about the optimal way to allocate these slots. This paper provides a framework for analyzing competition for such scarce inputs, describing the outcome of an auction of slots between two carriers, who may have existing slots, and the possi ble outcomes from a merger or takeover wave. The author finds that the equilibrium allocation of slots is typically asymmetric, but not monopo listic, because as the allocation of slots becomes more concentrated, the price that the leader must pay for the marginal slot rises. This suggests that the concern over monopolization of airports may be misplaced. Regional Wage Convergence and Divergence: Adjusting Wages for Cost-of-Living Differences 26 by Randall W. Eberts and Mark E. Schweitzer After decades of convergence, the economic fortunes of U.S. regions appeared to diverge in the early 1980s as measured by both per capita income and wages. This study examines that phenomenon by looking at the effect of rela tive price-level controls on the convergence/divergence of regional wages. The authors find that once prices are factored in, relative wage rates continue to converge across regions due to rising covariance between price and wage lev els. The results also confirm that the trend in regional wage variation can be traced to declining differences in labor market valuations of worker attributes rather than to shifts in the regional composition of the workforce. Editors: Tess Ferg Robin Ratliff Design: Michael Galka Typography: Liz Hanna Opinions stated in Economic Re view are those of the authors and not necessarily those of the Fed eral Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted pro vided that the source is credited. Please send copies of reprinted material to the editors. ISSN 0013-0281 U.S. Banking Sector Trends: Assessing Disparities in Industry Performance by Katherine A. Sarnolyk Introduction The U.S. banking industry has a long tradition of decentralization as measured by geographic market structure. This feature largely reflects the impact of both inter- and intrastate branch ing restrictions as well as regulatory policies to ward mergers and acquisitions. As a result of these policies, the industry comprises many small banks that operate in relatively localized and stmcturally diverse markets. In states al lowing branching, banks tend to be fewer but larger than in unit banking states. The phenomenon of bank holding companies emerged in the 1950s and 1960s as a response to restrictions on the scale and scope of bank ing activities. By holding banks as affiliates, a holding company can expand the geographic scale of its banking operations and broaden the scope of its nonbank activities to certain per missible lines of financial services. During the 1970s and 1980s, both the number of bank holding companies and the share of banks so affiliated increased, partly as a response to regulatory changes (Savage [1982], Amel and Jacowski [19891). However, the trend also re flects changes in the environment in which http://fraser.stlouisfed.org/these organizations operate. Federal Reserve Bank of St. Louis Katherine A. Samolyk is an econo mist at the Federal Reserve Bank of Cleveland. The author thanks Robert B. Avery, Richard Dreese, Ian Gale, William Gavin, and James Thomson for helpful comments. While the U.S. banking industry has been con solidating into holding companies, it also appears to be shrinking. Domestic nonfinancial-sector debt grew substantially faster than GDP in the past decade, but the share of intermediated funds advanced by banks fell from 50 percent to 36 percent. The number of banks contracted by nearly 21 percent, from more than 14,400 in 1982 to about 11,400 at the end of 1992. The decade also witnessed a dramatic rise in bank failures and a spate of asset quality problems that trans lated into low industry profitability. In assessing these trends, analysts have de voted considerable attention to the regional na ture of the banking industry. Disparities in bank profitability over the past decade have been widely attributed to differences in local economic fortunes. Bank failures were largely concentrated in states experiencing economic difficulties. More recently, the poor performance of banks in New England and California has been associated with the so-called bicoastal recession. Regional banking conditions also reflect the structural diversity across state banking sectors. Historically, bank failures have tended to occur in unit banking states, whereas institutions in branch banking states seem to have fared better during periods of economic adversity.1 In the past several years, however, the problems con centrated in large banks have raised concerns that a “too big to fail” regulatory policy is en couraging excessive risk-taking.2 Differences in economic fortunes and in bank structure across states complicate the as sessment of industry performance. Are certain types of banks performing poorly because they are inherently different from other types of banking organizations, or do they happen to be concentrated in regions where the local economy is faltering? In this paper, I exploit the differences both within and across states in an attempt to evaluate how these factors were related to banking sector performance during the past decade. Performance refers to stan dard measures of banking conditions, including bank profitability, asset quality, capitalization, and lending. Data are compiled from individ ual Federal Financial Institutions Examination Council’s Reports o f Condition a n d Income (call reports) for each year between 1984 and 1992. I disaggregate state-level balance sheets and income statements to construct perform ance measures for banks that differ in size as well as in their holding company relationships. Then, controlling for state-specific economic factors, I examine the extent to which dispari ties in performance have been associated with differences in these bank characteristics. The tone of this analysis is descriptive; the parsimonious number of relationships examined precludes a more causal interpretation. The find ings reveal that the health of the local economy is indeed important in assessing the performance of the local banking sector. However, differences in banking conditions also appear to be associ ated with bank size and holding company affili ation; moreover, the emergent relationships are consistent with microeconomic studies that exam ine individual bank performance (Berger, Hanweck, and Humphrey [1987]). The results indicate that, controlling for local economic factors, rela tively small banks (assets between $100 million and $1 billion in 1987 dollars) turned in the best performance over the past decade. In addition, smaller institutions that were affiliated with 1 This trend is less true in recent years. With the exception of Texas, failures during the past decade were not disproportionately located in unit banking states. Moreover, Wheelock (1993) notes that the choice of unit banking restrictions was popular in states with relatively cyclical economies, such as agricultural states. multibank holding companies had fewer prob lems with asset quality than did other small banks. Alternatively, the largest institutions — almost all of which belong to multibank hold ing companies — were less profitable. Hence, although banking fortunes reflect those of the local economy, performance also appears to be related to local industry structure. Judging at least by the experience of the 1980s, it seems that banks can be too large. I. Industry Structure and Performance: An Overview Assessments of the banking industry frequently describe it as a composite of banks that differ in size and location. For example, the FDIC Quarterly Bulletin presents industry data on banks classified by size and geographic region.3 This focus reflects the view that such factors are important determinants of banking conditions. The term bank structure is frequently used when referring to the characteristics of banking markets as well as those of individual institu tions. Individual bank characteristics, such as the scale and scope of operations, can affect the costs at which banks produce financial serv ices; hence the rationale for the focus on bank size. Market stmcture, measured by the relative size and number of firms, can influence the de gree of local competition and, by extension, the quality, quantity, and price of financial serv ices ultimately available to bank customers. Researchers have studied how both market structure and individual bank characteristics are related to bank performance. One genre of studies looks at how market concentration is related to bank profitability and to the cus tomer’s cost of banking services.4 Most find ings reveal a positive relationship between market concentration and bank profitability. This result has been cited as evidence that more concentrated markets are less competi tive. However, it also has been interpreted as an indication that more efficient firms tend to dominate the marketplace. A second line of re search looks at how the costs associated with producing financial services are related to a ■ ■ 2 See Boyd and Gertler (1993) for a recent evaluation of this http://fraser.stlouisfed.org/ perspective. Federal Reserve Bank of St. Louis ■ 3 Similarly, assessments of changes in the structure of banking markets focus on how the geographic distribution of banks and the atten dant concentration of banking markets have evolved (Amel and Jacowski [1989]). ■ 4 For example, see Berger and Hannan (1989). Structural Characteristics of Bank Cohorts Size Classes (1987 dollars) Very small: Less than $100 million in assets. Small: $100 million to $500 million in assets. Medium: $500 million to $1 billion in assets. Large: $1 billion to $10 billion in assets. Very large: More than $10 billion in assets. Holding Company Affiliations MBHC: Bank holding company holding more than one bank. SBHC: Bank holding company holding only a single bank. Independent: Not affiliated with a bank holding company. Performance Measures of Industry Conditions Lending and Capitalization Capitalization: Bank equity capital as a percentage of total assets. C&I lending: Commercial and industrial loans as a percent age of total assets. CRE lending: Commercial real estate loans as a percentage of total assets. Total bank lending: Total loans (including C&I loans, CRE loans, home mortgages, consumer loans, and other loans) as a percentage of total assets. Bank Profitability and Asset Quality ROA: Return on assets as measured by the ratio of annual net income to total assets. Nonperforming assets: Past due loans (more than 90 days) plus nonaccruing assets plus other real estate owned, as a percentage of total assets. Net loan charge-offs: The ratio of annual net charge-offs for loan losses to total bank loans as defined above. NOTE: All measures use fourth-quarter data from the Federal Financial Insti tutions Examination Council’s Reports o f Condition and Income (call reports). Each performance measure is constructed from the cohort-level balance sheet or income statement. For example, ROA for each size class of banks is measured as the ratio o f net income to total assets for each respective cohort o f banks. bank’s structural characteristics.5 Although the results are mixed, these cross-sectional assess ments of bank efficiency have found evidence of modest economies of scale; the costs of pro viding banking services decline as firm size in creases up to a relatively small size (Berger, Hanweck, and Humphrey [1987]). The potential for the characteristics of banks and banking markets to affect industry perform ance motivates our interest in the phenomenon of bank holding companies. The importance of viewing holding company affiliation as a struc tural characteristic of banks depends on whether a bank in a holding company behaves differently than it would as an unaffiliated entity. At one extreme, holding companies may be passive ve hicles that diversify across a number of banks and allow almost all decisions to be made at the subsidiary level. In this case, holding company affiliation might be unrelated to a bank’s perform ance because it does not affect the bank’s behav ior. At the other extreme, if a bank can draw on its relationship with its holding company (for ex ample, by reducing certain operating costs or increasing portfolio diversification through inter bank loan sales), it may perform more like a larger institution. Here, my focus on the link between the structural characteristics of banks and industry performance at the state level is more macroeconomic than microeconomic in nature. To the extent that banking conditions may impact credit availability, they may also affect eco nomic activity. In a previous study using statelevel data from the past decade (Samolyk [1992]), I found evidence suggesting that the health of the local banking sector plays a role in local economic fortunes. Banking conditions were more strongly related to current real per sonal income growth in states where the health of the banking sector was poor than in states where it was sound. Moreover, this relation ship was not simply mirroring a correlation be tween banking conditions and past income growth. These findings suggest that if local bank characteristics affect local industry per formance, they may have important economic consequences. Both market factors and regulatory policies determine the structural characteristics of banks and banking markets. Here, I merely ex amine whether these characteristics have been associated with differences in banking sector performance. For example, small, localized banks may be more vulnerable to local eco nomic distress, while larger banks are able to diversify over regional or even national markets. Thus, the state is not defined as the relevant “market” for banks of all types. Nevertheless, performance differentials across the various types of banking institutions within a state may provide evidence as to how bank characteris tics can affect local banking conditions. ■ 5 The scale of a bank’s activities is usually defined in terms of bal ance sheet stocks, such as the volume of lending. The scope of a bank’s activities refers to the composition of financial services it provides (for ex ample, making loans versus funding securities). See Clark (1988) for a survey of these studies. 5 I TA8L E 1 Banks and Banking Assets Panel A By Size of Largest Banking Organization By Individual Bank Size 1992 1984 1992 1984 Number of Banks Total 14,451 11,419 14,451 11,419 Very small Small Medium Large Very large 11,769 2,171 210 274 27 8,823 2,037 229 293 37 9,830 1,985 412 1,445 779 7,399 2,196 344 855 625 Percentage of Banking Assets Total Very small Small Medium Large Very large Panel B 100.0 100.0 100.0 100.0 15.2 11.6 13.6 5.5 32.1 37.2 12.0 9.2 10.1 15.1 5.3 28.7 35.7 In Multibank Holding Company 1984 9.9 3.8 27.2 47.1 In Single Bank Holding Company 1992 1984 1992 3.3 18.8 58.6 Not in a Bank Holding Company 1984 1992 Number of Banks Total 3,748 3,295 4,967 4,891 5,736 3,233 Very small Small Medium Large Very large 2,426 989 132 176 25 2,030 871 141 220 33 4,088 728 58 91 2 3,932 828 64 63 4 5,255 454 20 7 0 2,861 338 24 10 0 Percentage of Banking Assets Total 67.0 72.7 Very small Small Medium Large Very large 3.7 7.4 3.3 18.7 33.9 3.0 6.2 3.4 26.2 33.9 23.2 • 20.7 5.5 4.8 1.5 9.6 1.8 5.3 5.3 1.5 5.3 3.3 9.8 6.6 6.0 2.9 0.5 0.4 0.0 3.3 2.1 0.6 0.6 0.0 o f single bank holding companies. Following convention, I characterize a bank’s size in terms of the dollar value of its assets. Banks are placed in five size categories, which are ad justed for inflation so that a bank’s classification will change only if its asset size has changed in real terms.6 The inflation-adjusted (constant dollar) ranges for the five size cohorts are pre sented in box 1. Between 1984 and 1992, the unadjusted (current dollar) ranges of these size classes rose by approximately one-third. Panel A of table 1 shows the distributions of banks and bank assets across the five categories. It also presents the distributions of banks and banking assets when each bank is classified by the size of its largest related organization. For example, in classifying a $100 million bank that is a subsidiary of a holding company with as sets of $5 billion, I include that bank’s data in the larger size cohort. This classification illus trates the distribution of banking assets by firm size when multibank holding companies are treated as branches of the holding company.7 Panel A shows the trend toward fewer, larger banks in the industry: The number of very small banks has declined markedly. It also indi cates that at the holding company level, the past decade has witnessed very large banking organizations growing into even larger ones. Panel B of table 1 presents the distribution of banks and of banking assets classified by both size and holding company affiliation as defined in box 1. It indicates that the decline in the number of small and very small banks reflects decreases in both holding company affiliates and unaffiliated (independent) institu tions. These data also underscore the emer gence of the bank holding company as a prom inent organizational entity. However, inde pendent banks continue to be well represented, especially among smaller institutions. Panel B of table 1 is also useful for under standing the measures employed in assessing banking conditions at both the national and SOURCE: Author’s calculations. ■ II. Trends in Industry Structure At the end of 1992, 11,419 domestic commer cial banks filed call reports. O f these institu tions, 71.7 percent were affiliates of a bank holding company. O f holding company affili ates, 40.3 percent were part of multibank hold ing companies and 59.7 percent were affiliates 6 Both the FDIC Quarterly Report and the Federal Reserve Bulletin publish data on performance trends for banks classified by nonindexed size cohorts. Empirical studies that use cross-sectional data in a given year do not need to index nominal asset size classifications. However, studies that pool data on banks across time should deflate asset values into real terms to evaluate differences associated with bank size. For example, Avery and Ber ger (1991) index their classification of large and small banks in assessing the implications of risk-based capital on these segments of the industry. ■ 7 The data on the number of banks require more clarification, as this category refers to the number of banks affiliated with holding compa nies of a given size, not the number of holding companies of that size. This indicates the potential misclassification associated with ignoring holding company affiliations. TABLE 2 III. Trends in Bank Lending and Capitalization Bank Balance Sheets: Lending and Capitalization (percentage of bank assets) All Banks Very Small Small Medium Large Very Large 52.9 52.9 51.6 56.5 60.0 56.7 58.5 65.2 60.0 60.5 65.4 60.1 65.0 63.4 58.3 and industrial loans 22.6 16.8 13.1 10.7 18.8 13.9 8.8 15.3 10.5 18.9 16.9 12.2 21.0 18.8 14.8 30.9. 23.5 20.0 Bank lending 1984 60.2 1989 62.4 1992 58.0 Commercial 1984 1989 1992 Commercial real estate loans 7.3 11.3 10.4 6.9 9.2 10.2 10.1 13.5 14.1 9.4 14.8 14.5 8.3 13.2 11.4 5.2 8.8 7.5 Capitalization 1984 6.1 1989 6.2 1992 7.5 8.6 8.9 9.3 7.2 7.6 8.3 6.9 6.7 7.7 5.8 6.1 7.6 4.8 4.8 6.6 1984 1989 1992 SOURCE: Author’s calculations. the state level. The focus here is on how per formance differs across banks of various sizes and holding company affiliations. I construct national-level fourth-quarter performance meas ures for each of the five size classes (and each size class disaggregated by the three types of holding company relationships) by first aggre gating the call report data for U.S. commercial banks in the size cohorts illustrated in panel B. The aggregated balance sheet and income statement of each cohort is then used to con struct measures of capitalization, lending, bank profitability, and asset quality. These measures are defined in box 1. The state-level measures analyzed in the study are also constructed in this manner— albeit with the call report data on the individual banks for a given state. As a point of reference, I first examine the recent trends in banking conditions evident at the national level. Table 2 illustrates bank lending and capitaliza tion (as percentages of bank assets) for the five size cohorts of commercial banks in se lected years. Despite distinct differences in both loan/asset and capital/asset ratios across the size classes, these balance sheet measures have moved somewhat in concert during the past decade. Larger banks appear to have invested a greater percentage of assets in loans than did smaller banks. While loan/asset ratios moved procyclically in medium and large banks, very small and very large banks did not exhibit this portfolio shift.8 However, banks of all sizes changed the types of loans they funded over the period. The percentage of assets invested in commercial and industrial (C&I) loans declined and the percentage held as commercial real estate (CRE) loans rose in all segments of the industry. Large and very large banks moved most aggressively into CRE lending in the mid1980s and have subsequently retrenched. For smaller banks, the monotonic shift to funding CRE loans is more indicative of a secular trend than of a cyclical real estate boom (and subse quent bust). Larger banks also appear to have been less capitalized than smaller banks. However, the greatest disparities across the size classes oc curred early in the 1980s, when industry capi talization was below 6 percent. Until the late 1970s, smaller banks faced higher capital re quirements: They were viewed as riskier be cause they could not diversify as much as larger institutions. Subsequently, regulatory changes have eliminated differences in capital ratios based on size in favor of requirements associated with portfolio risk. The result has been that capitalization has increased in the in dustry as a whole, but by relatively more in larger banks. There also appear to be differences in bank lending and capitalization across similar-sized banks that vary in holding company affiliation. The panels in figure 1 depict capitalization and loan/asset ratios for very small banks sorted by their holding company relationships. Multibank holding company affiliates were less capital ized than otherwise affiliated institutions and made more loans than other very small banks, ■ 8 See Boyd and Gertler (1993) for documentation of trends over the postwar period. FI GURE although the differences in loan/asset ratios di minished over the decade. Thus, in terms of lending and capitalization, industry-level data suggest that multibank holding company affili ates behaved more like “larger” institutions than did other very small banks. This also ap pears to have been the case for small banks. 1 Capitalization and Loan/Asset Ratios for Very Small Banks: by Holding Company Affiliation Percentage 11.0 A. Capitalization 10.5 Independent b a n k s ____ _ 10.0 IV. Trends in Bank Profitability and Asset Quality 9.5 9.0 SBHC affiliates 8.5 8.0 MBHC affiliates 7.5 7.0 1984 1985 1986 1987 1988 1989 1990 1991 1992 Percentage 57 B. Total Lending \ MBHC affiliates ^ SBHC affiliates"'“ '''■ 50 *- *** ^ x * Independent banks 49 ______ i______ i----- 1----- 1----- 1----- 1----- 1----- 1-1984 1985 1986 1987 1988 1989 1990 1991 1992 Percentage Percentage 13 12 D. CRE Lending 11 MBHC affiliates 10 9 SBHC affiliates 8 7 6 Independent banks -------1-------------- 1--------------L. 1984 1985 1986 1987 SOURCE: Author’s calculations. -I_________I_________ I_________L. 1988 1989 1990 1991 1992 Differences in loan/asset ratios and bank capi talization are important factors in assessing the relative profitability and risk of banks. How ever, these variations do not inevitably trans late into differences in risk or profitability. Although loans are a relatively risky class of in vestments (compared to securities), banks that have higher loan/asset ratios do not necessar ily have riskier portfolios. Larger banks (or those affiliated with multibank holding compa nies) may be able to diversify the risks in their loan portfolios more successfully than smaller, more localized institutions. Moreover, larger banks may be profitable in spite of lower profit margins because their higher leverage allows them to pay a greater return to stockholders for a given return on their assets. As evidenced by capital requirements in the past, smaller banks were viewed as riskier be cause of their limited ability to diversify. In the 1980s, government policies may have changed the relationship between bank size and bank risk by reducing the incentives for banks to man age losses prudently (Boyd and Graham [1991]). Deregulation, in tandem with changes in the treatment of problem institutions, may have in creased the risks that uninsured investors allow banks — especially larger ones — to assume. The policy that banks can be too big to fail and the usual method of resolving bank failures (via purchase and assumption by healthy banks) shift the cost of bank failure from these investors to the Federal Deposit Insurance Corporation (and ultimately to taxpayers). Unfortunately, ex ante portfolio risk and ex pected risk-adjusted yields are unobservable, so I examine data on ex post performance to infer indirectly how the risk-return relationship may vary across types of banks. I employ stan dard industry ratios used to measure bank profit ability and asset quality (see box 1). Bank profitability is measured by the return on assets (ROA) for each class of banks. Problems with asset quality are measured by nonperforming FIGURE 2 Industry Performance Measures by Bank Size Percentage A. Return on Assets Ì » ^ m• mè ^ —______ -1 • • ; 1985 1986 „ --- Very small — Small ---Medium — — Large ...... Very large . . *.. -...... J..... ..... 1_--------........ 1987 1988 1989 1990 1991 1992 \I* 1984 ---- ---- assets as a percentage of total assets, and net charge-offs for loan losses as a percentage of total loans. The panels in figure 2 illustrate how bank profitability and problems with asset quality var ied across the five size classes of banks between 1984 and 1992. These performance measures re veal less of a discernible relationship to bank size and holding company affiliation than do bank lending and capitalization. Panel A indicates that the ROA for very large banks was more volatile than for smaller institutions. However, the differ ences in ROA do not indict size per se as an indi cator of profitability. This is especially true from the perspective of bank stockholders; since larger banks are more leveraged, stockholders can earn a higher return on equity for a given ROA. Panels B and C of figure 2 illustrate that smaller (primarily agricultural) banks experienced problems with asset quality in the mid-1980s. These problems have been widely attributed to the impact of local economic conditions. The dra matic rise in both nonperforming assets and loan charge-offs by larger institutions from 1988 to 1991 is commonly viewed as stemming from the troubled commercial real estate markets on the East and West coasts. These disparate economic conditions make it difficult to identify a consis tent relationship between bank size and asset quality in the national-level data. V. A Regional Perspective on Banking Sector Performance SOURCE: Author’s calculations. Both banking sector performance and broader economic conditions varied widely across states during the past decade. At the same time, restric tions on branching and on bank holding com pany acquisitions were being eased in many states. In spite of these regulatory changes, a great deal of stmctural diversity remains both within and across state banking sectors. This di versity reflects the interaction of current regula tory environments with inherent market factors (such as size or population density). Table 3 summarizes the differences in bank ing sector conditions across states in 1984 and 1992 in terms of the maximum, median, and minimum values of each measure as well as their means and standard deviations. The data mirror the trends evident at the national level, yet the variation across states is striking. TABLE 3 State-Level Commercial Banking Industry Ratios (percentage of total assets) Capitalization Maximum Median Minimum Mean Standard deviation 1984 1992 9.7 6.7 5.2 7.0 1.0 12.8 CRE Lending 1984 1992 C&I Lending 1984 7.9 6.5 8.2 22.0 8.2 1.6 8.4 22.5 11.0 2.4 11.0 30.3 17.5 7.2 18.0 1.3 3.5 3.9 4.5 Return on Assets 1992 Net Loan Charge-offs 1992 1984 1992 1984 1992 1.6 0.8 -0.1 0.8 3.0 1.1 0.0 1.1 7.0 1.4 0.6 2.0 5.1 0.8 0.3 0.5 3.6 1.6 0.4 1.8 0.8 1984 24.4 12.0 5.7 12.8 4.1 Nonperforming Assets 1.9 1.3 0.5 0.2 0.7 0.3 1.2 0.5 1.0 NOTE: Net charge-offs are expressed as a percentage of total loans. SOURCE: Author’s calculations. Table 4 presents the distribution of banks within and across states by their holding com pany affiliation in 1984 and 1992. Each state is ranked according to its total number of banks in 1984. Industry consolidation has been the rule rather than the exception in state banking sectors. The number of banks fell in 42 states during this period, and for 20 of these states the numbers dropped by more than 20 percent, although 38 states still had more than 50 banks at the end of 1992. These declines were accom panied by a decrease in the number of inde pendent banks in 42 states. Overall, the percentage of smaller holding company affili ates also fell. This, however, is due to signifi cant decreases in some states (most notably, Texas), which outweigh the increases in these affiliates in other states. A trend toward one organizational type is not evident at the state level. Savage (1993) ar gues that the coexistence of holding company affiliates and independent banks within states indicates that there is not yet a dominant form of banking organization. The distribution of banks by size varies more substantially across states. In states with a tradition of unit banking, the industry tends to be populated by a large number of smaller institutions. As of year-end 1992, only 10 states had banks in the largest size cohort. Seven states had no banks with more than $1 billion (1987 dollars) in assets. However, except for very large banks, each size class is fairly well represented within and across states. VI. Assessing Disparities in Industry Conditions The diversity in the types of banks within and across states suggests a simple way of assess ing the extent to which variance in bank per formance can be attributed to differences in bank characteristics versus local economic con ditions. In the following analysis, each state is treated as an individual sector composed of banks that vary in size and holding company affiliation. I then test for differences in industry profitability and asset quality that can be attrib uted to these structural characteristics, control ling for local economic conditions and other state-specific fixed effects. The analysis features state-level data over the nine-year sample period of 1984 to 1992. As with the evaluation of national-level trends, I disaggre gate state-level measures of industry conditions into cohort-level measures for the five bank size categories, crossed with the three types of hold ing company relationships. Hence, the annual data yield 15 potential observations in each year on industry conditions within a state. Not all states have banks in each class.9 The nine years of data for 51 states (including the District of Co lumbia) yielded a data set of 4,062 observations on a given measure of bank performance. The rationale for analyzing the performance of a cohort of banks rather than that of each ■ 9 For example, almost all very large banks are multibank holding company affiliates (two are affiliates of single bank holding companies). The distribution of the data is discussed in the final section. KO TABLE 4 Distribution of Banks by Holding Company Affiliation In Multibank Holding C o m p a n y In Single Bank Holding C o m p a n y Not in a Bank Holding C o m p a n y Percent Percent Percent Total N u m b e r 1984 1992 1984 1992 1984 1992 1984 1992 13 15 16 19 21 25 26 27 32 46 50 59 60 12 8 18 22 17 20 22 21 40 38 48 28 54 7.7 20.0 18.8 5.3 23.8 20.0 34.6 22.2 46.9 2.2 16.0 35.6 18.3 16.7 37.5 38.9 40.9 23.5 45.0 22.7 28.6 50.0 42.1 18.8 25.0 13.0 69.2 33.3 18.8 52.6 4.8 32.0 15.4 33.3 15.6 32.6 20.0 18.6 38.3 41.7 37.5 27.8 31.8 11.8 15.0 50.0 42.9 25.0 31.6 29.2 50.0 25.9 23.1 46.7 62.5 42.1 71.4 48.0 50.0 44.4 37.5 65.2 64.0 45.8 43.3 41.7 25.0 33.3 27.3 64.7 40.0 27.3 28.6 25.0 26.3 52.1 25.0 61.1 North Carolina Oregon South Carolina Maryland New Mexico Washington Wyoming Massachusetts New Jersey South Dakota Mississippi Montana Virginia 63 72 73 88 95 102 116 124 125 140 153 167 176 78 48 81 96 84 94 63 63 106 121 121 120 170 1.6 13.9 1.4 30.7 41.1 16.7 56.0 48.4 38.4 23.6 0.0 44.9 26.1 21.8 20.8 18.5 38.5 44.0 14.9 36.5 19.0 23.6 28.9 5.0 35.8 27.1 28.6 20.8 28.8 11.4 32.6 21.6 24.1 23.4 20.8 32.9 46.4 29.3 10.2 25.6 27.1 38.3 26.0 38.1 21.3 41.3 46.0 37.7 47.1 53.7 40.8 22.4 69.8 65.3 69.9 58.0 26.3 61.8 19.8 28.2 40.8 43.6 53.6 25.7 63.6 52.6 52.1 43.2 35.4 17.9 63.8 22.2 34.9 38.7 24.0 41.3 23.3 50.6 North Dakota New York West Virginia Arkansas Alabama Tennessee Louisiana Ohio Pennsylvania Kentucky Michigan Indiana Georgia 177 190 227 258 269 293 302 320 326 336 365 378 383 143 177 164 259 215 248 221 271 281 311 215 270 397 24.9 30.0 18.5 16.7 26.0 17.1 0.0 31.3 15.3 4.5 54.5 2.1 25.6 21.7 19.8 59.8 34.0 25.1 27.4 7.2 32.1 33.1 37.3 42.3 43.3 30.0 46.3 21.1 16.3 39.9 25.3 38.6 52.0 12.2 22.7 49.7 8.5 46.3 26.9 64.3 46.9 18.9 44.4 49.3 54.0 60.2 26.6 40.6 46.9 38.1 39.3 39.5 28.8 48.9 65.2 43.4 48.7 44.4 48.0 56.6 62.0 45.8 37.0 51.6 47.5 14.0 33.3 21.3 21.6 25.6 18.5 32.6 41.3 26.3 15.8 19.5 17.4 30.5 427 446 449 472 538 590 628 629 713 738 1,241 1,853 394 349 451 374 393 445 508 542 510 593 1,006 1,089 44.0 49-6 6.9 10.0 8.7 32.0 3.3 21.8 45.4 24.4 25.9 40.9 29.9 47.6 8.9 26.5 15.0 40.2 17.7 29.7 36.9 - 27.2 37.0 20.7 23.4 30.3 34.7 61.4 57.6 28.3 66.6 49.4 28.5 47.8 33.1 23.0 32.0 29.5 31.9 52.9 59.3 35.7 61.8 54.8 42.7 54.3 40.6 40.8 32.6 20.2 58.4 28.6 33.6 39.7 30.1 28.8 26.1 27.8 41.0 36.0 38.1 22.9 59.2 20.6 25.7 24.0 20.5 15.5 20.4 18.5 22.5 38.6 Rhode Island Alaska Nevada Washington, D.C. Hawaii Idaho Maine Vermont Delaware Arizona Connecticut New Hampshire Utah Florida Colorado California Nebraska Oklahoma Wisconsin Kansas Iowa Missouri Minnesota Illinois Texas SOURCE: Federal Financial Institutions Examination Council, Reports o f Condition and Income, 1984 fourth quarter and 1992 fourth quarter. individual bank is to mitigate the effects of out liers and bank mergers. The cohort measures are averages of the individual banking data, where each bank is effectively weighted by its share of the cohort.10 However, the findings should be similar to those obtained using indi vidual bank data to assess differences across these classes of banks. Moreover, in examining the performance of classes of institutions, I can construct estimates of the performance differen tials associated with bank size and holding company affiliation. These adjusted measures are directly comparable to the national-level data presented in figures 1 and 2. Data on banks sorted by size and holding company affiliation within each state are used as cross-sectional observations on banking condi tions in each year of the sample period. I then pool the data for each year to estimate reducedform regressions for six measures of bank per formance. To identify variance in performance that may be attributed to differences in bank size and holding company affiliation, I control for other factors that affect banking conditions both within each state and over time. In each regres sion, the following control variables are included: D a dummy variable identifying the state of an ob servation to control for state-specific differences in banking conditions during the sample period; 2) a dummy variable indicating the year of an observa tion to control for economywide variation in bank ing conditions over time; 3) the contemporaneous and lagged values of both the growth rate of state personal income and the volume of per capita failed business liabilities to control for the effect of local business conditions on banking sector per formance; and 4) the ratio of state banking assets to state personal income in each year to control for variation in banking sector activity relative to that of the broader state economy. Finally, to test whether bank size and hold ing company affiliation can explain differences in bank performance, I include dummy vari ables in each regression that measure intercept shifts for all classes of banks. Two different specifications for each measure of industry con ditions are estimated; these vary the ways in which the dummy variables are interacted with ■ 10 This method mitigates the effects of outliers within a class of banks. Outliers, in terms of a performance measure, will affect the measure only to the extent of their relative importance in the cohort. For example, to measure the ROA of small independent banks in each state (and each year), I take the ratio of their aggregated net income to their aggregated assets. A $50 million bank will, on average, contribute less to each term in the ratio than a $100 million bank. Of course, this is also the case for the data gener ally used in industry analyses. http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis time to examine whether performance differ entials changed over the sample period. Both regression specifications include the same control variables and differ only in their treat ment of the dummy variables. VII. Evidence on Banking Sector Performance Specification 1 includes an intercept shift for each size class (Size) and type of bank holding company affiliation (HCA) as well as the set of control variables. This specification takes the form +X W Ec°n,nj+ ZnXsA i= 1 where Ration t, h is an observation of a bank performance measure, in the nth state and i th year, of banks in Size class 5 and HCA class h. Econi „ t( i= 1, 5) is the set of state-level economic variables that includes current and lagged personal income growth, current and lagged per capita failed business liabilities, and the ratio of bank assets to personal income. Here, the intercept shifts associated with Size and HCA will measure how the average per formance of banks with these characteristics varies. It is possible to estimate performance differentials only relative to a base group in each class. Very small banks are the Size base group and multibank holding company affili ates are the HCA base group. Table 5 presents selected results of the re gressions on six measures of banking condi tions (summarized in table 3) obtained using specification 1. For brevity, the individual coef ficient estimates of the intercept shifts for each state are not reported. However, they suggest that significant differences in banking condi tions across states can be attributed to statespecific factors (other than current economic conditions) during the sample period. These may reflect average differences in local indus try structure, including the structure of the banking sector. Similarly, the estimated inter cept shifts for each year of the sample (relative to the base year, 1992) indicate that in evaluat ing bank performance over time, it is impor tant to control for economywide trends that affect banks in all states. These coefficient esti mates mirror the trends in banking conditions evident in the national-level data shown in Regressions Explaining Cohort-Level Performance Capitalization CRE Lending C&I Lending Return on Assets Nonperforming Assets Net Loan Charge-offs 0.146 0.407 0.397 0.164 0.285 0.188 State Dummies (a) (a) (a) (a) (a) (a) Year Dummies (a) (a) (a) (a) (a) (a) -0.0155 (-0.37) 0.0038 (0.09) -0.0010 (-1.06) -0.0018 (- 1.99)b -0.0027 (-1.10) -0.1816 (—3.4l)a -0.1464 (-2.72)a 0.0787 (1.49) -0.0003 (-1.16) -0.2622 (—15.69)a -0.1780 (-8.42)a -0.0931 (-7.21)3 0.1043 (1.99)b -0.0025 (- 2.09)b -0.0027 (-2.34)a -0.0081 (-2.66)a 0.07447 (6.65)a 0.0661 (5.99)a -0.0017 (-0.66) -0.0009 (-2.81)a -0.0056 (-1.81) -0.0009 (-2.85)a 0.0053 (8.25)a 0.0029 (7.59)a 0.001 (1.00) 0.0070 (3.33)a -0.0062 (—3.33)a — -0.0258 (-9.78)a -0.0084 (-3.69)a -0.0272 (- 10.20)a 0.0026 (1.13) 0.0031 (5.65)a 0.0012 (2.53)a 0.0022 (2.51)a 0.0016 (2,l6)b 0.0013 (1.99)b 0.0005 (0.89) — — — — — — Small -0.0173 (-10.37)3 0.0222 (10.54)a 0.0150 (7.04)a -0.0229 (- 10.71)3 -0.0304 (- 14.42)3 0.0188 (6.99)a 0.0087 (3.28)a -0.0276 (-5.60)3 0.0340 (12.47)3 0.0432 (16.10)3 0.0856 (17.20)3 0.0003 (0.50) 0.0016 (1.80) 0.0021 (2.38)a 0.0066 (3.99)a -0.0009 (-1.77) Medium 0.0020 (4.49)a 0.0008 (1.34) 0.0011 (1.98)b -0.0008 (-0.77) Dependent Variable R2 Explanatory Variables Economic Controls Personal income growth Lagged personal income growth Failed business liabilities Lagged failed business liabilities Bank assets to personal income 0.0009 (2.24)b -0.1071 (-8.41)3 0.0010 (3.33)a 0.0014 (4.97)a 0.003 (4. I I ) 3 Holding Company Affiliation Dummies Not in a bank holding company In single bank holding company In multibank holding company Size Dummies Very small Large Very large -0.0389 (-9.93)a 0.0009 (1.34) 0.0012 (1.83) 0.0039 (3.23)a a. Significant at the 1 percent level. b. Significant at the 5 percent level. NOTE: T-statistics are in parentheses. Coefficients o f dum m y variables indicate the intercept shift relative to the omitted category, as indicated by dashed lines. SOURCE: Author’s calculations. 13 FIGURE 3 Adjusted and Unadjusted Performance Differentials in Return on Assets by Bank Size (Relative to Very Small Banks) Percentage Percentage Percentage Percentage SOURCE: Author’s calculations. table 2 and figure 3 (again, for brevity, the indi vidual coefficients are not reported). Bank performance does appear to reflect lo cal economic conditions, particularly in regard to bank profitability and asset quality. Both in come growth and failed business liabilities help explain ROA, nonperforming assets, and loan charge-offs in the expected ways. Profit ability as measured by bank ROA is positively related to income growth and negatively re lated to failed business liabilities. Symmetrically, asset problems measured in terms of both nonperforming assets and loan charge-offs are negatively related to income growth and posi tively related to failed business liabilities. C&I lending— and to a lesser extent CRE lending— is negatively related to failed business liabilities; banks appear to fund fewer loans when the credit quality of the local business sector dete riorates. However, the coefficients on state in come growth suggest that end-of-year lending as a share of assets is also lower when recent income growth has been higher. The coeffi cients on the ratio of banking assets to state personal income are positive in the regressions explaining both ROA and loan charge-offs. Hence, when banking activity is high relative to economic activity, both bank profitability and problems with asset quality are higher as well. Finally, bank capitalization is relatively unrelated to the economic control variables. These findings, then, reveal that the profit ability and asset quality of different segments of the industry to a large degree reflect the economic conditions impacting these institu tions: W hen the local economy has been far ing poorly, it is likely that the banking sector will follow suit. This analysis is consistent with most interpretations of banking trends. How ever, the results also hint that differences in bank performance can be attributed to differ ences in bank characteristics. The results for specification 1 yield signifi cant variations in banking conditions among institutions having different holding company relationships. Both single bank holding com pany affiliates and independent banks had a smaller share of assets invested in CRE loans than did multibank holding companies, while only independent banks held a significantly smaller share of C&I loans. Controlling for size, independent banks were, on average, more capitalized than multibank holding company affiliates, while single bank holding company affiliates were less capitalized. All else equal, both independent banks and single bank hold ing company affiliates earned a higher ROA 14 FIGURE 4 Adjusted and Unadjusted Performance Differentials in Nonperforming Assets by Bank Size (Relative to Very Small Banks) Percentage 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1984 1985 1986 1987 1988 1989 1990 1991 1992 1986 1987 1988 1989 1990 1991 1992 1986 1987 1988 1989 1990 1991 1992 1986 1987 1988 1989 1990 1991 1992 Percentage 2.0 1.5 1.0 0.5 0.0 -0.5 - 1.0 1984 1985 Percentage 1.5 1.2 0.9 0.6 0.3 0.0 -0.3 - 0.6 1984 1985 Percentage 0.6 0.4 0.2 0.0 - 0.2 -0.4 - 0.6 1984 1985 SOURCE: Author’s calculations. than banks in multibank holding companies. However, independent banks and single bank holding company affiliates also had more prob lems on average with asset quality, as reflected in nonperforming assets and loan charge-offs. Thus, controlling for size, the performance of banks in multibank holding companies dif fered from otherwise affiliated institutions; the former earned a lower average ROA, but also had fewer problems with asset quality. These affiliates invested a larger share of their portfo lios in loans, but they appear to have been better risks than both single bank holding com pany affiliates and independent banks in terms of their performance during the past decade. Bank performance also varied significantly across the five size classes. As illustrated in table 5, the differences in capitalization and in lending mirror those evident in the nationallevel data. The coefficients measuring the aver age differences in capitalization for each size class indicate that controlling for other factors, capitalization is inversely related to bank size. These coefficients imply that on average, the capital/asset ratio of very large banks was 3.9 percentage points lower than that of very small banks. C&I lending as a share of assets is posi tively related to bank size. Alternatively, while the middle three size classes of banks invested a higher percentage of assets in CRE loans than did very small banks, very large banks held a significantly smaller share. Perhaps not surprisingly, I find a less consis tent relationship between bank size and bank performance in terms of profitability and asset quality. All else equal, small and large banks earned significantly higher ROAs than did either very small or very large banks. Large and very large banks had higher loan chargeoffs and lower nonperforming asset ratios than did very small banks. Relatively small banks ($100 million to $500 million in assets, 1987 dollars) seem to have turned in the best per formance in terms of profitability and asset quality. Interestingly, these banks are about the size that some studies have shown to maxi mize economies of scale (Berger, Hanweck, and Humphrey [1987]). In summary, specification 1 estimates the average differences in bank performance that can be attributed to bank size and holding company relationships, controlling for local economic factors and aggregate trends that affect banking conditions. Thus, the results measure the extent to which the variation in bank performance w ithin states is related to these structural characteristics. I find systematic 19 F I G U R E differences in capitalization and lending across banks that vary in their size and holding com pany relationships. Indeed, the disparities in capitalization and lending observed at the na tional level appear to largely reflect these struc tural differences. The results also indicate some variation in bank profitability and asset quality across the different types of banks during the past decade. 5 Adjusted and U nadjusted Perform ance D ifferentials in N et Lo a n C harge-offs by Bank S ize (R e la tive to V e ry S m a ll Banks) V III. W ill Performance Differentials Continue? The observed differences between large and small banks may have changed during the past decade. Estimating how the average perform ance of one type of bank (over a number of years) compares to that of another may ob scure emerging differences in performance. For example, because of greater asset diversifi cation, the asset quality of larger banks could have been significantly better than that of smaller banks early in the 1980s. If, indeed, these institutions have increased the relative risk of their investments (because they are too big to fail), they may still be on par with small banks in terms of average performance al though their asset quality has been declining. To examine the possibility that structural dis parities in bank performance have changed in the past decade, I estimate a second set of regressions: Percentage (2) Ratio n tsh = a + $ *ate + *Year + *Year 5 +X 1984 1985 1986 1987 SOURCE: Author’s calculations. 1988 1989 1990 1991 1992 i= W ECOnt,rU+E>Us,h- 1 Specification 2 includes an intercept shift for each Size and HCA class for each year in the sample period, as well as the control variables included in specification 1. Again, it is possible to estimate differences only relative to a base group in each class; hence the choice of very small banks as the Size base group and multi bank holding company affiliates as the HCA base group. In these regressions, each inter cept shift associated with a class of banks measures the estimated difference between the performance of that class and its respective base group in a p articular year. Therefore, the estimated performance differentials associated with bank size and holding company relation ships are allowed to vary over time. 16 Here, I present the evidence of performance differentials in ROA, nonperforming assets, and loan charge-offs obtained using specification 2. The results for the control variables in these re gressions are similar to those presented in table 5. Thus, I focus on patterns in the time-varying in tercept shifts associated with bank size and hold ing company affiliation. The intercept shifts can be interpreted as annual bank performance differ entials that have been adjusted for state-specific factors, local economic conditions, and economywide trends. These adjusted performance differ entials are therefore estimates of the within-state variations in bank performance attributable to structural characteristics. In figures 3, 4, and 5, the green lines illustrate the adjusted performance differentials in ROA, nonperforming assets, and loan charge-offs for each Size class of banks. It is instructive to com pare these estimates to measured differences in performance that have not been adjusted for other economic factors. The national-level per formance measures presented in table 2 are used to construct unadjusted performance differentials of this sort for each Size class of banks. For exam ple, the unadjusted differential in the ROA of very large banks in each year is simply the ROA for this cohort of banks minus the ROA for very small banks (as illustrated in figure 2). The unad justed performance differentials in ROA, nonper forming assets, and loan charge-offs are depicted by the blue lines in figures 3, 4, and 5. The panels in figure 3 illustrate both the adjusted and unadjusted differences in ROA for each Size class of banks (relative to very small banks). A comparison of these series indicates that controlling for other economic factors miti gates the relatively poor performance of larger banks in recent years. The adjusted differentials in ROA for small banks suggest that, all else equal, they were more profitable than very small banks during the entire sample period. However, the adjusted differentials for the three largest classes reveal no emerging trends in profitability differen tials that can be attributed to bank size per se. The panels in figure 4 depict the unadjusted and adjusted differentials in nonperforming as sets by Size class. The adjusted series do not ex hibit the increasing disparities between sizes that are evident in the unadjusted data. Panels A and B do indicate that, all else equal, larger banks had significantly higher nonperforming asset ra tios than did smaller banks in the late 1980s. Sub sequently, however, the differences attributable to size decreased. Finally, the panels in figure 5 illustrate both the adjusted and unadjusted differences in loan charge-offs for various-sized banks. The ad justed differentials are measurably smaller than the unadjusted ones. As in the case of nonper forming assets, controlling for other economic factors mitigates the emerging relationship be tween size and asset quality problems sug gested by the unadjusted measures. However, again, it does appear that larger banks showed more asset quality problems than did smaller banks during the late 1980s. The key result yielded by these series is that, controlling for economic factors, there is no evidence of a trend toward increasing dispari ties in bank performance that can be attributed to bank size as a structural characteristic. Simi larly, the estimated performance differentials for the HCA classes do not suggest emerging dis parities in bank profitability and asset quality associated with holding company relationships. IX. Banking Sector Performance: Assessing the Trends One interpretation of observed bank perform ance in the past decade is that the disparities be tween larger and smaller banks may indicate that increasing risks are being borne by the largest players (Boyd and Gertler [19931)■Accordingly, the trend toward larger banks has been viewed with some concern. This paper investigates the merits of this perception by descriptively assess ing the extent to which differences in banking conditions can be attributed to variations in bank size and holding company affiliation. Taking local economic factors into considera tion reduces the disparities in bank performance attributable to these structural characteristics. There is some evidence that, after controlling for state-specific fixed effects, local economic condi tions, and national-level trends, larger banks perfonned worse during the 1980s than did smaller institutions. However, the trend appears to have since reversed. In addition, the results indicate that, all else equal, banks that are associated with larger organizations through multibank holding company affiliations tended to perform better than otherwise-affiliated institutions.11 ■ 11 In interpreting these findings, it is important to note that I am evaluating cohort-level banking conditions. Smaller banks that grow large because they are profitable are allowed to be reclassified into larger size cohorts. On the other hand, banks that are poor performers also may leave their cohorts as they are either closed or merged into larger institu tions. Although beyond the scope of this study, it would be interesting to examine the extent to which trends in banking sector performance have been associated with changes in the population of banks across size classes and holding company affiliations. One potential explanation for the finding that large banks performed relatively poorly is that these institutions may be more likely to make loans outside their locality. To the extent that this is true, I do not control for economic conditions where they made loans. An obvious example is the case of the huge write-offs asso ciated with loans to developing countries. The rationale for controlling for local economic con ditions is that certain institutions are more sub ject to these conditions. Thus, evidence that less constrained firms are riskier may suggest that they could and did take on more risk dur ing the past decade. In assessing the performance of large banks, it is also important to note that most of these in stitutions are part of multibank holding compa nies. Thus, a study of the behavior of large banks is effectively a study of the joint effects of both large size and this form of banking organization. O n the other hand, the evidence indicates that, all else equal, multibank holding company affili ations appear to benefit banks. This suggests that smaller affiliates have not experienced the same problems with asset quality as have larger institu tions. Thus, it seems that the performance of larger banks reflects the effects of size rather than holding company status per se. Indeed, risk ier loans may have been channeled to larger banks in the holding companies. The results of this analysis indicate that the U.S. banking industry during the 1980s may have been characterized by a duality related to bank size. Although reregulation in the past several years has attempted to address this possibility, the evolving role of banks indicates that the link between bank size and bank performance mer its further study from a regulatory perspective. References Amel, Dean F., and Michael J.Jacowski. “Trends in Banking Structure since the Mid-1970s,” Federal Reserve Bulletin, vol. 75, no. 3 (March 1989), pp. 120-33. Avery, Robert B., and Allen N. Berger. “Riskbased Capital and Deposit Insurance Re form,” Jo u rn a l o f Banking a n d Finance, vol. 15, nos. 4/5 (September 1991), pp. 847-74. Berger, Allen N., Gerald A. Hanweck, and David B. Humphrey. “Competitive Viability in Bank ing: Scale, Scope, and Product Mix Econo mies,” Journal o f Monetary Economics, vol. 20, no. 3 (December 1987), pp. 501 —20. Berger, Allen N., and Timothy H. Hannan. “The Price-Concentration Relationship in Bank ing,” Review o f Economics a n d Statistics, vol. 71, no. 2 (May 1989), pp. 291-99. Boyd, John H., and Mark Gertler. “U.S. Commer cial Banking: Trends, Cycles, and Policy,” National Bureau of Economic Research Working Paper No. 4404, July 1993Boyd, John H., and Stanley L. Graham. “Investi gating the Banking Consolidation Trend,” Federal Reserve Bank of Minneapolis, Q uar terly Review, vol. 15, no. 2 (Spring 1991), pp. 3-15. Clark, Jeffrey A. “Economies of Scale and Scope at Depository Financial Institutions: A Re view of the Literature,” Federal Reserve Bank of Kansas City, Economic Review, vol. 73, no. 8 (September/October 1988), pp. 16-33. Samolyk, Katherine A. “Bank Performance and Regional Economic Growth: Evidence of a Regional Credit Channel,” Federal Reserve Bank of Cleveland, Working Paper 9204, February 1992. Savage, Donald T. “Developments in Banking Structure, 1970-81,” Federal Reserve Bulletin, vol. 68, no. 2 (February 1982), pp. 77-85. ________ . “Interstate Banking: A Status Report,” Federal Reserve Bulletin, vol. 79, no. 12 (De cember 1993), pp. 1075-89. Wheelock, David C. “Is the Banking Industry in Decline? Recent Trends and Future Prospects from a Historical Perspective,” Federal Re serve Bank of St. Louis, Review, vol. 75, no. 5 (September/October 1993), pp. 3-22. Competition for Scarce Inputs The Case of Airport Takeoff and Landing Slots by Ian Gale Ian Gale is an economic advisor at the Federal Reserve Bank of Cleveland. The author thanks Mark Stegeman for helpful discussions. Introduction The process of deregulating airlines in the United States began in earnest with the passage of the Airline Deregulation Act of 1978. The Act set in place a timetable for removing govern ment controls on fares and entry into routes. However, airline access to airports has not been fully deregulated. In 1968, the Federal Aviation Administration (FAA) adopted the socalled “high-density rule” to combat congestion at four airports. Specifically, limits were placed on the number of operations per hour at Ken nedy, La Guardia, O'Hare, and Washington National. More and more airports are becoming crowded, given the difficulties in securing per mission from local authorities to expand exist ing facilities or to build new ones, so this prob lem will persist.1 For many years, the right to take off and land at the four crowded airports was deter mined by a committee system. Multilateral ne gotiations took place among the incumbent carriers and prospective entrants, with the FAA stepping in if an impasse occurred. Since 1986, ■ 1 Only one major airport has been built in the United States since http://fraser.stlouisfed.org/1974, that being the new Denver facility. Federal Reserve Bank of St. Louis incumbents have had property rights called “takeoff and landing slots” or simply “slots.” A slot permits the owner to make one operation (a takeoff or landing) during a specified time period. Carriers now have the right to buy and sell slots.2 Questions remain about the optimal way to allocate existing or new slots. The first question asks why the incumbents should be given valuable property rights.3 The second asks whether allowing carriers to buy and sell slots will lead to monopolization of airports, with resulting higher fares. Repossessing slots and then selling them back to carriers would raise revenue and would ensure that those car riers willing to pay the most would acquire the slots. This has positive welfare implications in a perfectly competitive environment, but the impact is less clear if there is the potential for wielding market power. This Economic Review analyzes competition for scarce inputs such as airport takeoff and landing slots. It describes the outcome of an auction of slots between two carriers, w ho may ■ ■ 2 For further discussion, see Grether, Isaac, and Plott (1989). 3 A slot at O’Hare recently rented for $66,000 per month. See “United Wins TWA Lease," New York Times, March 20,1992. have existing slots. This allows us to evaluate the welfare implications of selling slots. It like wise describes the outcome of a merger or takeover battle in which two large incumbents seek to buy up the slots of small competitors, if not the competitors themselves."4 The objects for sale (the slots) are sold in a sequence of auctions. The aggregate value of the slots, and each bidder’s valuation of the marginal slot, increase as slots become more concentrated in the hands of one bidder. These conditions arise naturally in market games, since control of slots can confer market power by limiting the competitiveness of one’s rivals. I focus primarily on the case of two bidders who start with no slots. The final allocation is typically unequal, and many different interme diate allocations lead to the same final alloca tion. In addition, if bidders start with nonzero holdings of slots, a bidder who holds the en tire initial allocation of slots often does not block new entry. I then discuss the extension to more than two bidders, starting with three bidders, and then with more bidders than slots. While the focus here is on the allocation of scarce inputs, the analysis contributes to the general theory of auctions as well.'’ Most theo retical studies of auctions have considered “one-shot” auctions in which a single object is sold or several objects are sold simultaneously. In reality, related objects are often sold sequen tially, either because sellers enter the market one at a time, or because it is practical for an individual seller to sell related objects as an on going process. (Goods auctioned sequentially include art, wine, procurement contracts, tim ber rights, and mineral rights.) McAfee and McMillan (1987) note that all levels of govern ments of western nations procure goods and services equal to 10 percent of gross national product annually, much of it by competitive bidding. Given the economic importance of these markets, further study is warranted. A common feature of many of these settings is that the value of the object for sale, be it a contract to produce or a scarce input, depends on the other objects obtained. If the objects are identical, then this observation implies that the total value of the objects obtained is a non linear function of the quantity obtained, which is the case here. ■ 4 For an example, see “American Deal W ill End Britt O’Hare Opera tion,” Chicago Tribune, December 20,1988. ■ 5 See McAfee and McMillan (1987) tor a survey of the literature on sequential auctions. I. The Model Suppose that N identical objects called slots are offered for sale by one or more sellers.6 Assume that N > 2 and N is an even number. The sellers could be a government agency wishing to allocate some or all of the takeoff and landing slots at an airport, or they could be carriers with small holdings of slots. (Large carriers may wish to take over the small carriers outright or they may simply wish to purchase their slots.) Two bidders compete for the slots. I denote a typical allocation of slots by the ordered pair (x, y), where x denotes the num ber of slots currently held by bidder X, y de notes the number of slots currently held by bidder Y, and 0 < x + y < N. If x +y = N, then (x, y) is a final allocation. At each nonfinal allocation, an auction allocates the next slot. Af ter all slots are allocated, each bidder receives a “final payoff’ that depends on the final allo cation. This payoff can be thought of as the profit from production. The specific auction format used is the second-price sealed-bid auction. In this auction, the high bid wins, and the winner pays the amount bid by the losing bidder. This format is used for ease of exposition because its outcome mimics that of the standard oral ascending-bid auction. Suppose, for example, that bidder X was willing to pay up to $100 for a slot, where as bidder Y was willing to pay up to $80. In an oral ascending-bid auction, where the auc tioneer raises the price, bidder Y should not drop out until the price hits $80 exactly. Thus, bidder X will win the auction and pay a price of (approximately) $80. If, instead, the bidders were asked to write down how much they were willing to bid, with the understanding that the bidder who submitted the higher fig ure would receive the slot at a price equal to the loser’s figure, then the outcome should be the same. The latter scheme is the secondprice sealed-bid auction. A minimal requirement for an equilibrium of this game is that each bidder’s strategy be optimal, given the strategy of the other bidder. This is the requirement that the strategies form a Nash equilibrium. A Nash equilibrium is selfenforcing in the sense that neither bidder has an incentive to change her strategy, given the strategy of the other bidder. Unfortunately, there can be many Nash equilibria, so we must be more discerning. ■ 6 The model and the analysis borrow extensively from Gale and Stegeman (1993). Suppose, for example, that bidder X values a slot at $10 while bidder Y values it at $8. In a second-price auction, it is a Nash equilibrium for X to bid $10 and for Y to bid any amount strictly below $10. Neither bidder has an incen tive to change her bid. If bidder X submits a bid above Y ’s bid, then she receives a surplus equal to the difference between her valuation ($10) and Y ’s bid. This holds regardless of X ’s exact bid, as long as she wins. Bidder Y has no incentive to change since he loses and re ceives zero surplus currently, whereas if he were to outbid X, then he would lose $2, since he would pay $10 for a slot that he values at $8. Thus, any pair of bids ($10, bY) with bY < $10 constitutes a Nash equilibrium, as does any pair of bids ( bx , $8) with bx > $8. While there are many Nash equilibria, only one is “robust” in a certain sense. Once we consider the possibility that bidders might make a mistake and submit the wrong bid, then bidder X should bid $10 and bidder Y should bid $8. If, for example, Y bids $9.50, ex pecting that X will bid $10, but X makes a mis take and bids only $9, then Y wins and pays $9, yielding a net surplus of -$1. Likewise, if X bids $9, thinking that Y will bid $8, but Y makes a mistake and bids $9.50, then bidder X has lost $.50 in surplus that would have ac crued had she bid $10. The bids bx = $10 and bY- $8 are referred to as “dominant strategies,” since they are optimal regardless of the other bidder’s bid. I restrict attention to dominant strategies, which leaves each bidder with a unique bidding strategy.7 I study the equilibrium sequences of winners and the associated prices. (Tie bids produce mul tiple sequences of winners.) The bidders know the initial allocation of slots, the auction rules, and the final payoffs (the profits from produc tion, given the final allocation of slots). I focus on the case in which both bidders start the auction with no slots, but consider other initial allocations as the analysis progresses. A numerical example illustrates the work ings of the sequential auction. Suppose that there are two slots for sale. The two firms re ceive a profit from production of $2 apiece if each of them owns one slot. A firm receives a profit of $5 if it controls both slots, in which case the other bidder receives zero. The slots are sold in a sequence of two second-price auctions. Suppose that bidder X wins the first auction. She will be willing to pay up to $3 for ■ 7 Strictly speaking, I eliminate, iteratively, all strategies that are http://fraser.stlouisfed.org/dominated by some other strategy. Federal Reserve Bank of St. Louis the second slot, which is her marginal valua tion. Bidder Y is willing to pay up to $2, which is his marginal valuation. Bidder X will win the second auction and pay a price of $2. Given this behavior in the second auction, both bid ders are willing to bid exactly $3 in the first auction. Therefore, the price of the first slot is $3 and the price of the second is $2. One bid der wins both, but both bidders receive net profits of zero. Now suppose that owning one slot is worth $3 instead, all else equal. After the first auc tion, the winner has a marginal valuation of $2, whereas the loser has a marginal valuation of $3. The loser of the first auction will win the second and pay a price of $2 for that slot. Thus, neither bidder will be willing to pay more than $2 to win the first slot, since by los ing he or she will win the second slot. Simi larly, each bidder will be willing to pay up to $2 for the first slot. Since winning the first auc tion at any price strictly less than $2 is better than winning the second auction at a price of $2, it follows that both bidders will bid $2 for the first slot. This process generalizes to N slots. The equilibrium bids are determined by backwards recursion, as outlined below. At any point during the sequence of auctions, the current allocation of slots is an ordered pair (x; y), where bidder X owns x slots and bidder Y owns y slots. The allocation (x, y) is followed by (x + 1, y) or (x, y + 1), depending on who wins at (x; y). For each number of slots, x = 0, 1, ..., N and each bidder i = X, Y, f t (x) denotes the final payoff given to bidder i if the final allocation is (x, N - x). Since the slots are assets,^ is increasing in x and f Y is decreasing. Each bidder i seeks to maximize the difference between her final payoff (x) and her total expenditure for slots purchased in auctions. Let Vi (x, y), i = X, Y, denote bidder i ’s valu ation of allocation (x, y). In other words, Vj (x, y) is bidder i’s equilibrium payoff in the game commencing at (x, y). The valuation of a final allocation is the final payoff: (1) Vi (x ,N - x ') = f {(x ), x =0, 1, 2, ...,N; i= X, Y. The valuation of an intermediate allocation (x, y) depends on the valuations of its sequels (x + 1, y) and (x, y + 1). There is a unique val uation of (x, y ) if each bidder adopts the strat egy in auction (x, y ) and in all subsequent auctions of bidding the increment to value re sulting from obtaining the next slot. (As noted above, any auction in which the resulting bids are unequal has many Nash equilibria, but the possibility of mistakes justifies the assumption that bidders will play dominant strategies.) Consider first the penultimate allocations (x, y), where 0 < x, 0 < y, and x +y = N - 1. The value of that next slot to bidder X, say, is the difference between her final payoff when she wins and when she loses: Vx (x + 1, y) Vx (x, y + 1). It is a dominant strategy for bid der i to bid Bi (x, y) in auction (x, y), where slots increases, while assumption (3d) means that aggregate final payoffs increase with the final concentration of slots. (Note that 13c] says that bidder Y ’s final payoff is increasing and con cave in his own number of slots.) Finally, assump tions (3e) and (30 are invoked for concreteness. The example discussed below satisfies these conditions, but the results of Theorem 1 do not require them. (2a) Bx (x ,y ) = Vx (x + l,y ) - Vx(x, y+ 1) ; (3b) fx (x) is strictly increasing and convex, (2b) By(x, y) = VY(x, y+ 1) - VY (x + l,y ). (3a) / x (x ) = f Y (N - x ) for all x. for x > N / 2. If B x(x, y) > B Y(x, y), then bidder X wins the next slot and pays bidder Y ’s bid. If B x(x ,y) < B y ( x , y), then bidder Y wins the next slot. If Bx (x, y) = B Y(x, y), then the tie-breaking rule determines the winner. The valuations at each penultimate allocation can now be calculated, given these equilibrium bids. This allows bids to be determined one auc tion earlier, at the antepenultimate allocations, and so on. Bids at any earlier allocation also take the form of (2). The bidders’ valuations of each allocation can be calculated recursively. The price paid in auction (x, y) is min [Bx (x, y), BY(x, y)]. The tie-breaking rule, which I leave unspecified for now, cannot affect bids or valuations in any given auction. The equilibrium of the auction game com prises the bidding functions Bx (x, y) and B v (x, y ), for all (x, y). Henceforth, I assume that the bidders are symmetric with respect to their final payoffs. The outcomes will typically not be symmetric, however. II. Competition for Scarce Inputs I now put some additional structure on the model. First, symmetry of bidders is imposed in (3a). Second, aggregate final payoffs, and each bidder’s valuation of the marginal slot, are assumed to rise as the final concentration of slots increases (that is, as the allocation of slots between bidders becomes more un equal). These conditions arise naturally for scarce inputs because higher concentration typically leads to higher output prices, increas ing both aggregate profits and the value of marginal output. Assumptions (3b) and (3c) mean that each bidder’s valuation of the mar ginal slot rises as the final concentration of (3c) f Y (x) is strictly decreasing and concave, for x > N /2. (3d) f ( x ) = fx(x ) +fY (x) is strictly increasing, for x > N /2. (3e) fx (x) = a x + (3x x+ d x x 2/2 , and (30 fy (x ) = a y + P Kx - 8}.x 2/2, where d Y > 5X >0. Assumptions (3a) - (3d) place additional restric tions on the parameters, which I leave unspeci fied for now. W inning a particular slot affects a bidder di rectly by increasing her current holdings of slots, but it also affects the competition for the remaining slots. Two countervailing effects de termine the number of slots won by, say, bid der X. The first effect argues for one bidder to acquire all slots, since that outcome maximizes the sum of the final payoffs. This effect is coun tered by the fact that the more slots bidder X wins, the higher is the marginal valuation of another slot to bidder Y (presuming that y < x), and the fiercer is the competition for the mar ginal slot. If bidder Y ends up with very few slots, then bidder X is a near-monopolist. Bid der Y benefits from X ’s relative passivity in the production game, so marginal capacity is most valuable to bidder Y when he has very little. (Empirically, airlines that dominate airports tend to have the highest fares. See Borenstein [1992].) This trade-off leads to an equilibrium alloca tion of slots that is typically neither symmetric nor monopolistic. A heuristic argument provides some intuition for the determinants of the equilibrium alloca tion of slots. I proceed by conjecturing the form of the equilibrium. I then show that it is optimal in each auction for bidders to bid in the way prescribed. The heuristic argument for deter mining the equilibrium allocation is now given. Consider an allocation (x, y) such that, in equilibrium, bidder X wins any auction (x ', y ') > (x, y) such that y ' < y + 2. In other words, if bidder X has at least x slots and bidder Y has y, y + 1, or y + 2 slots, then bidder X will win all remaining slots in equilibrium. If a deviation at (x, y) causes bidder Y to win one more slot, then any additional slot that bidder Y wins will, in equilibrium, be his last slot. Therefore, fol lowing the deviation, bidder Y bids the value of that last slot, B y (N - y - 2, y + 1), in every remaining auction. If the deviation did not oc cur, bidder Y would bid B Y(N - y - I, y) in every remaining auction. Since bidder X wins all remaining slots, even if a deviation causes her to lose one of those slots, the marginal con tribution to her final payoff of winning auction (x, y) is equal to the marginal value of winning the last auction, BX (N - y - 1, y). By winning auction (x, y ), bidder X in creases the price that she must pay upon win ning the remaining N - x - y - 1 auctions by ABY(N - y - 2, y+ 1) = B Y(N - y - 1, y )- B Y (N - y - 2, y + 1). Therefore, bidder X wins auction (x, y) if and only if BX(N - y - 1, y) > (N - x - y - 1 )ABY(N - y - 2 , y+ 1) + BY(N - y - 1, y). Recalling that /(x ) = f x (x) + / y-(x), the condition can be rewritten: (4 ) f ( N - y) - fC N - y -1 ) > (N - x - y - 1) AB Y CN - y - 2, y+ 1). In other words, bidder X wins auction (x, y) only if the increment to aggregate final payoffs (the left-hand side) exceeds the increment to total prices paid (the right-hand side). Define p to be the smallest integer such that (y + 2, y) satisfies (4) for y = p, p + 1, N/2 - 2. The equilibrium allocation will be (N - p, p), which I call the “modal allocation,” or its sym metric counterpart. Theorem 1. When (3) holds, either bidder X wins N - p slots and bidder Y wins p, or the re verse holds. The equilibria of the original game and the game starting from any intermediate allocation have a regular form. From here on, I assume that ties are won by bidder X if the allocation is even. This tie-breaking rule ensures that bidder X always has at least as many slots as bidder Y, so the equilibrium allocation in this case is (TV- p, p). If x = y, then there will be a tie in the next auction, which buyer X wins by the tie- breaking rule. Buyer X also wins the next auc tion to maintain her lead. Thereafter, different patterns are possible. If I assume that bidder Y wins all ties when he is behind, then the bidders alternate victories until buyer X commences her final string of victories, and the modal allo cation is reached. If x > y, then buyer Y wins a string of auctions first, but not enough to catch up. Either buyer Y catches up to within one slot, and then a pattern of alternation persists until the modal allocation is reached, or else he never catches up to within one, in which case buyer X wins all auctions after Y ’s string of vic tories. The point at which buyer X commences her string of victories is determined by (4). The theorem is illustrated by the following numerical example. Suppose that each slot repre sents one unit of capacity in a subsequent pro duction game, where the inverse demand is P = 24 - Q and average cost is zero, up to capacity. If bidder X has x slots, then she is able to pro duce up to x units of output. For quantities strictly less than x, marginal and average cost both equal zero, but for quantities strictly above x, marginal and average cost are infinite.8 Suppose that the firms behave like Cournot duopolists in the production game. In a Cour not duopoly, each firm chooses the optimal quantity to produce, given the level produced by its competitor. In the absence of capacity constraints, each duopolist produces eight units of output in the Cournot equilibrium. To see this, note that if bidder X produces q x units of output, then bidder Y ’s profit from producing q units is (24 - qx - q) q, a concave function that takes its maximum at q = (24 - qx )/2. Like wise, if Y produces qY, bidder X ’s optimal strategy is to produce q = (24 - qy)/2. If both bidders are unconstrained, then the unique Cournot equilibrium has qx = qY = 8. Assume, however, that exactly N = 16 slots are available, and x > 8. Concavity of the profit function ensures that the best responses for the two bidders are (5a) qx = min{(24 - qY)/2, x I (5b) qY = min {(24 - qx ) / 2 ,16 - x }. There are now four possibilities, depending on whether the bidders are constrained. The bid ders cannot both be unconstrained, since that would require that bidder Y produce eight units, which exceeds his capacity. It is also immediate ■ 8 In reality, the capacity constraint is not absolute, since more seats can be put on an airplane, but the qualitative properties need not change when this possibility is permitted. FIGURE 1 Possible Sequences of Equilibrium Victories wins and (3,0) is reached or whether Y wins and (2,1) is reached. The modal allocation is (10,6) here. Straight forward calculations confirm that (4) amounts to (7) x SOURCE: Author’s calculations. that the bidders cannot both be constrained. If Y produced 16 - x units, X would wish to produce only [24 - (16 - x W 2 = 4 + (x/2), whereas he has available x units of capacity. Because x > 8, 4 + (x /2 ) < x. Now suppose that X is constrained and Y is unconstrained. If qx = x, qY = min {(24 - x)/2, 16 - x } = 16 - x, which is a contradiction. The only remaining possibility is that bidder X is unconstrained and bidder Y is constrained. It is straightforward to confirm that qx = 4 + (x /2) and qY = 16 - x are the unique equilibrium outputs. Total output is 20 - (x/2). The price of output is (8 + x)/2, so the final payoffs (that is, the profits in the Cournot production game) are (6a) f x (x) = (8 + x)2/4, (6b) f Y(x) = (8 + x)(l6 - x)/2. Note that these final payoffs are consistent with (3a) - (30. I now return to the auction. The different possible sequences of equilibrium victories are noted in figure 1. Arrows indicate which alloca tions can follow a given allocation. For exam ple, starting from the allocation (2,0), where bidder X currently has two slots and bidder Y currently has none, the allocations (3,0) and (2,1) can both be reached. This means that the two bidders submit the same bid at (2,0), and the tie-breaking rule determines whether X (2y+ 1)/4 > (13 - 2y), or y >5.1. Thus, jj. = 6. Price drops occur be tween auctions separated by double cross marks. O n any path from the origin, the price is 10.5 in the first auction, 6.25 in the next 12 auctions, and 5.5 in the last three auctions. The equilibrium paths in the example have several other properties. 1. While the leader may ultimately win con siderably more slots than the follower, the leader strictly outbids the follower only when his lead is reduced to one or after the follower has already won all of her slots. A consequence is that there is an equilibrium path in which bidder X wins the first two slots, then bidder Y wins one, and then they alternate victories un til the allocation (fi +1, |i) is reached. At that point, bidder X wins all remaining slots. 2. The trade-off that determines the num ber of slots that the leader wants to win, along with the preceding observation, ensures that many intermediate allocations lead to the same final allocation. Any initial allocation (x, y) at which 0 < jy < x < | i + 1 and (x, y) # (|i + 1, |a + 1) leads to the same final allocation. In the example, the modal allocation is (10,6), and any initial allocation (x, y ) with 0 < y < 7, except ing only (7, 7), leads to the modal allocation. Thus, many different histories lead to the same equilibrium allocation of slots. This implies that a sizable initial advantage need not be maintained. 3. If the initial allocation is (0,0), then the price drops either once or twice. It drops im mediately after the leader is established (which equalizes net profits). It also declines immedi ately after the follower wins his last slot if a de viation causing him to lose that slot would have caused him to lose all subsequent slots as well. Since the follower loses some surplus in that case, he bids more than he would bid if a loss would be compensated by his winning another auction subsequently. The existence of at most three distinct prices depends on the precise functional form assumed. 4. An incumbent monopolist typically does not block entry if new slots become available. For instance, suppose that the initial allocation of slots is (12,0), which effectively makes firm X an unconstrained monopolist. Calculations show that if four new slots become available, then the prospective entrant, firm Y, wins the first two slots before firm X wins the last two. The reason for this is simply that it is often too costly to preempt an entrant completely, since the incumbent monopolist would leave the ca pacity idle whereas the entrant would find the capacity very valuable because the incumbent would still be a near-monopolist after entry. III. More Bidders In many markets, there are exactly two com petitors of consequence. This is particularly true at many airports and in many airline mar kets, so the assumption of two bidders is realis tic in such contexts. I now consider the impact of having more than two bidders. The exten sion is qualitatively different because the value of winning a slot typically depends on which rival would otherwise win it.9 Hence, a bidder often has no dominant strategy, so I focus on the equilibria in which each bidder bids the value of the next slot to her, assuming that the rival making the highest bid would otherwise win the auction. Multiple equilibria are common when produc tive capacity is sold, because keeping capacity away from constrained bidders is a public good that the unconstrained bidders may individually prefer not to provide. Thus, at an airport with two dominant incumbents, if new capacity be comes available, each incumbent may prefer that the other buy the new capacity and keep the en trant out. The following example illustrates the public good aspect in a case where all bidders start with no capacity. Suppose that three firms bid for three slots. Each slot represents one unit of capacity, where inverse demand is P = 8 (4 - Q ) and production is again costless up to the capacity constraint. In the post-auction production game, the firms again act as Cournot competitors. The final al location (3,0,0) returns final payoffs of 32, zero, and zero, respectively. The allocation (2,1,0) returns payments of 18, 12, and zero. The allo cation (1,1,1) returns a payment of 8 to each bidder. The payments from the other final allo cations are determined symmetrically. Equilibria are computed through backwards recursion. (The price paid in the second-price sealed-bid auction is the larger of the two losing bids here.) Consider the allocation (2,0,0). The last slot is worth 14 to the leader and bidder X, ■ 9 This is reminiscent of the situation in baseball pennant races, where contending teams attempt to trade for a player whose team is out of the race. How much a team is willing to pay depends on whether the other http://fraser.stlouisfed.org/ team vying for that player’s services is in the same division. Federal Reserve Bank of St. Louis and 12 to the others. Therefore, the leader pays 12 for the last slot. The payoff to entering this “subgame” (the game that takes place when the initial allocation is (2,0,0)) is 32 - 12 = 20 for the leader and zero for the followers. Consider next the allocation (1,1,0). The last slot is worth 10 to a leader if it would other wise go to the follower, bidder Z, but it is worth only 6 if it would otherwise go to the other leader. It is worth 8 to the follower. If tie bids are broken fairly, then there are three equilibria of interest. The follower bids 8 in all three equilibria. In one equilibrium, one leader bids 10 and the other 6; in another, their roles are reversed; in the last, they both bid 8. In the first two equilibria, the payoff to entering this subgame is 10 for the leader who wins, 12 for the other leader, and zero for the follower. Both leaders are willing to bid enough to shut out the follower, but each prefers that the other do it. If we assume that the leaders randomly coordinate on one of the two equilibria, the average payoff to entering this subgame is thus 11 for each leader and zero for the follower. At the allocation (1,0,0), the next slot is worth (32 - 1 2 )- 1 1 = 9 to the leader and 11 to each follower. (If the leader wins, he goes on to win the third slot at a price of 12.) One of the followers wins, and the payoffs to enter ing this subgame are 11 for the leader and zero for the followers. At the initial allocation (0,0,0), each firm bids 11. Summarizing, the assumption that firms ran domly coordinate among equilibria at the time of the auction generates an equilibrium allocation that is a permutation of (2,1,0). The final payoffs to the firms are 18, 12, and zero, and the equi librium prices are p x= 11, p~, = 11, and p 5 = 8. It is possible for three firms to earn higher profits if they coordinate on an equilibrium be fore the auctions begin. In particular, suppose that they coordinate on the equilibrium in which firm X wins auctions (1,1,0) and (1,0,1) and firm Y wins auction (0,1,1). Then the leader’s payoffs in these subgames are either 10 or 12, instead of 11. In auction (1,0,0) the equilibrium price is 12, and in auctions (0,1,0) and (0,0,1) the equilibrium price is 10. The pay offs in these three subgames are now 10 for the leader and zero for the followers, except that firm Z’s payoff in subgame (0,1,0) is 2 in stead of zero, and its payoff in subgame (0,0,1) is 12 instead of 10. If tie bids are broken fairly, then firm Z wins the initial auction (0,0,0) with a bid of 11 and pays a price of 10. Since the payoff to winning is 12, firm Z earns profits of 2, unlike the zero profits earned without advance coordination. The price sequence is p x= 10, p 2 = 10, p 3 = 8. Profits appear because advance coordination creates an asymmetry that gives firm Z an advantage from the start. This holds because firm Z is never put in the position of purchasing capacity that will not be fully util ized, since the other leader is assumed to win at (1,0,1) and at (0,1,1). In those cases, the other leader produces only 1.5 units even though it purchases 2. Finally, consider the possibility of there be ing more bidders than slots available. In Krishna (1993), an incumbent monopolist and potential entrants bid for new capacity. The monopolist wins only the last unit of new ca pacity, assuming that marginal costs are con stant, market demand is concave, and entrants always produce to capacity. The presence of potential entrants, each of whom will produce up to capacity, means that the bid from an en trant is always the equilibrium price of output should the entrant win. With two or three bid ders, all bidders may have some slots, in which case they all internalize the impact of increased (aggregate) production on the value of their current holdings. This makes the potential en trants less competitive in the small numbers case, so an incumbent monopolist will win more slots. IV. Concluding Remarks This paper has provided a framework for ana lyzing competition for scarce inputs such as air port takeoff and landing slots. The analysis describes the outcome of an auction of slots between two carriers, who may have existing slots, and it also depicts the outcome of a merger or takeover wave. The equilibrium allo cation of slots is typically asymmetric, even though firms are ex ante identical. It is not typi cally monopolistic, however, since the more concentrated the allocation of slots becomes, the higher is the price that the leader must pay for the marginal slot. Many different histories, and many different al locations of slots, lead to the same equilibrium allocation of slots, implying that an initial advan tage need not be maintained. Thus, the concern with monopolization may be misplaced. Future work will consider risk aversion, capital con straints, and matching of slots between airports. References Borenstein, Severin. “The Evolution of U.S. Airline Competition,” Jo u rn a l o f Economic Perspectives, vol. 6, no. 2 (Spring 1992), pp. 45-73. Gale, Ian, and Mark Stegeman. “Sequential Auc tions of Endogenously Valued Objects,” University of Wisconsin, Social Systems Re search Institute working paper, 1993Grether, David M., R. Mark Isaac, and Charles R. Plott. The Allocation o f Scarce Resources-. Experimental Economics a n d the Problem o f Allocating Airport Slots. Boulder, Colo.: Westview Press, 1989. Krishna, Kala. “Auctions with Endogenous Valuations: The Persistence of Monopoly Re visited,” Am erican Economic Review, vol. 83, no. 1 (March 1993), pp. 147-160. McAfee, R. Preston, and John McMillan. “Auc tions and Bidding,” Jo u rn a l o f Economic Lit erature, vol. 25, no. 2 (June 1987), pp. 699-738. Regional Wage Convergence and Divergence: Adjusting Wages for Cost-of-Living Differences by Randal IW . Eberts and Mark E. Schweitzer Introduction One of the basic tenets of economics is that the mobility of labor and capital tends to equal ize prices across markets. This tendency to ward price convergence is particularly notable across regional markets in the United States. For as long as regional income data have been collected, per capita income and wage rates have generally become more alike.1 In light of this long-run trend, a surprising reversal has occurred in several regional price measures. Since the early 1980s, the regional dispersion of wages, housing prices, and the general cost-of-living indexes has been on the rise. Browne (1989) provides evidence that re gional disparities in per capita income have been widening, while Eberts (1989) finds an in crease in regional wage dispersion. In addition, we demonstrate below that housing costs and regional price indexes have been following a similar pattern. Curiously, however, wages ad justed for regional cost-of-living differences ■ 1 Eberts (1989) demonstrates this trend. Unfortunately, the regional wage series is relatively short, beginning only in the 1950s. However, the same general pattern is found in regional per capita income, which is http://fraser.stlouisfed.org/ largely composed of wages and which extends well into the 1800s. Federal Reserve Bank of St. Louis Randan w . Eberts is the director of the W.E. Upjohn Institute for Em ployment Research in Kalamazoo, Michigan, and Mark E. Schweitzer is an economist at the Federal Reserve Bank of Cleveland. (which for brevity we refer to as locally a d justed wages) have continued to converge. Temporary deviations from the tendency toward convergence are not unexpected, as localized shocks can result in significant adjust ments to regional prices. Eberts and Stone (1992) and Blanchard and Katz (1992) show that nega tive localized employment shocks to a metro politan area can depress wages there by as much as 40 percent of their original level for up to six years before equilibrium returns. Even so, a significant period of increasing dispersion, as observed in the 1980s, is rare. In the last century, regional per capita income diverged only one other time, between 1920 and 1940. This paper focuses on the details of regional convergence or divergence in goods prices, na tionally adjusted wages (wages deflated accord ing to the national price level), and locally adjusted wages. Our goal is to identify and de scribe these obviously related phenomena. The characterization of this relationship follows Roback’s (1982) model of equilibrium in local la bor and land markets in the presence of local quality-of-life and production differences. The dispersion in locally adjusted wages depends on the dispersion of its components: nationally adjusted wages and local prices. We demonstrate the linkage between wages and prices by showing how the comovements of nationally adjusted wages and regional prices affect locally adjusted wages. Given that the two components of locally adjusted wage variation have followed similar paths, it is the growing covariance of these measures that results in con tinued convergence of locally adjusted wages between census regions. We also show that trends in the two wage dispersion series primarily reflect regional differ ences in market valuations of worker characteris tics rather than shifts in the levels of workforce characteristics. We modify the decomposition used by Eberts (1989) in examining the U-turn in nominal wage dispersion. He identifies two factors: 1) regional differences in the return on various worker attributes and in wage differen tials among industries and occupations, and 2) regional differences in the level of worker at tributes and in the distribution of workers among industries and occupations. Basically, these two factors distinguish between wage dispersion caused by regional markets placing different values on identical attributes, and dispersion caused by regions having different composi tions of attributes, even though regional mar kets value these attributes similarly. The analysis supports previous studies show ing that changes in regional wage differentials over time result from varying valuations of work er attributes, not from shifts in the regional com position of the workforce. The additional insight offered by this paper is that market forces pro duce different patterns of regional dispersion of nationally and locally adjusted wages. While not directly explained here, these differences are con sistent with the view that workers and businesses pursue separate objectives or place unequal weights on local prices and amenities. I. Explaining Regional Wage and Price Differentials The key to understanding potentially perma nent regional wage differentials is to recognize that not all factors are mobile across regions. Workers and firms interact in regional labor markets, determining wages and prices. Al though firms and their employees may respond quickly to changes in local market conditions, some factors that are unique to a region, such as geographic and climatic characteristics, re main the same. Even for those areas that share common features, the quality and quantity of site-specific characteristics may differ. There fore, firms and households may be willing to pay or accept different levels of compensation depending on the value they place on those at tributes. These immobile, site-specific features are referred to here as amenities: consumptive amenities apply to households and productive amenities apply to firms. A few examples of potential sources of con sumptive and productive amenities indicate their conceptual breadth and complexity. The prototypical consumptive amenity is a weather advantage. California and Florida attract people who prefer a warm climate and who are will ing to accept the higher costs of living there. Other potential consumptive amenities include familial or historical ties to an area, regionspecific recreational activities (skiing or surfing, for example), community spirit, and the quality and age of the housing stock. Despite the posi tive connotation of the term amenities, in our usage it also encompasses the negative features of an area, such as high crime rates or a combi nation of high local taxes and poor local gov ernment services.2 Port facilities are an excellent example of a productive amenity, since they can lower transportation costs for firms located nearby. Productive amenities also include low-cost dis tribution channels, informational advantages provided by firms’ proximity to other similar producers or suppliers, and state or local gov ernment protections or restrictions pertaining to local businesses. Interpreting Wage and Price Dispersion Interpreting regional wage and price conver gence in this framework is difficult. House holds and businesses can and will move to locations where they can better prosper. If both labor and capital are mobile, factor prices could converge or diverge in response to shifts in either firms’ and workers’ valuation of local amenities or changes in the availability of amenities in various locales. Another source of apparent convergence or divergence in regional wages and prices is the economy’s constant adjustment to a stream of shocks. The demand for and supply of labor in ■ 2 Local taxes are potential negative amenities to the extent that they are not included in prices. However, property taxes are essentially incor porated into the Consumer Price Index under the “rental equivalence measure" of housing costs. an area may be radically altered by technological changes or shifts in consumption preferences. Although households and businesses are mobile, adjustment delays may result in temporary peri ods of divergence. Studies by Eberts and Stone (1992) and Blanchard and Katz (1992) suggest that the adjustment period to a local labor-market shock may be as long as 10 years. Since housing and locally produced goods and services represent a major portion of a household’s budget, these prices become an important component of household utility and thus of household decisions. If local goods ac counted for the entire household budget, then consumption would equal household wages de flated by local prices. On this basis, we assume that given a stable value for local amenities, lo cally adjusted wages represent the primary moti vator of household mobility. By contrast, the price of local goods and services, including land, plays a smaller role in business decisions. Wages are generally a larger fraction of most firms’ costs than are local goods. Furthermore, for producers of local goods, an increase in local prices would affect both reve nues and costs. The marginal firm most likely to relocate would be a producer of national goods for whom any rise in local wages (or prices) relative to those faced by its competitors would immediately lower profits. For these firms, local wages (nationally adjusted), with little regard for local prices, should be the determining factor in their location once amenities and previous capi tal investments have been accounted for. It is important to compare observationally equivalent workers if we are to measure regional wage differentials accurately. Firms consider the skill level of their workforce as well as the size of their payrolls when making location decisions. Similarly, workers must evaluate the marketabil ity of their skills in various regions when compar ing locally adjusted wages. Therefore, regional shifts in factors associated with worker productiv ity, such as average educational attainment or workforce experience, should be controlled for in any analysis of factor-price adjustments. Shift ing patterns of employment by industry or occu pation, which may be related to compensating differentials associated with features of those jobs, should also be considered. II. Wage and Price Trends Wages Wages of individual workers are obtained from the March Current Population Survey’s (CPS) wage supplements for the years 1973 through 1991. The March survey reports annual wage and salary data and weeks worked from the previous year. Dividing annual earnings by weeks worked yields average weekly earnings for the years 1972 and 1990. For purposes of the respondents’ confidentiality, these data are coded by the Bureau of Labor Statistics (BLS) with a maximum salary for individuals whose pay exceeds the top-code value (for example, $199,998 after 1989). Average weekly earnings are computed after correcting for top-coding by assigning these individuals the mean of the appropriate Pareto distribution.3 The sample is limited to full-time workers who were em ployed all year or who, if unemployed for part of the year, spent that time actively seeking work. Because only full-time workers are in cluded, average weekly wages closely approxi mate average hourly wages. Two definitions of regions are used in this paper: Metropolitan Statistical Areas (MSAs) and census regions. Since cost-of-living indexes are available only for metropolitan areas, the basic unit of analysis is the MSA. The CPS iden tifies 44 MSAs, but the limited availability of price data for some of them reduces the usable number to 21. The small number of respon dents in most MSAs lowers the efficiency of estimation for that section of the analysis deal ing with the sources of wage convergence and divergence. To increase the number of indi viduals sampled in a given period, we pool to gether three years of individual responses for each MSA, resulting in a much broader cover age of worker characteristics and wages. Each of our six periods is identified by the middle year of the pooled three-year sample.4 For example, the first period, which consists of earnings in 1972, 1973, and 1974, is referenced as 1973 in the figures and tables. To provide another means of increasing the sample size for geographic comparisons, as ■ 3 See Shryock and Siegel (1971) for details on how a Pareto distri bution may be applied to truncated wage data. The Pareto distribution as sumes an exponential decline in the number of individuals with incomes above a certain amount, which is a reasonable characterization of higher income levels. ■ 4 The final period covers 1987 to 1990. F I G U R E 1 Variance of Regional Log Wage Differentials: Single vs. Grouped Years Variance of logs 0.012 0.010 0.008 Single years _ /r j/ - Grouped years 0.006 0.004 0.002 L 1972 J— 1974 _i_ 1976 „ i_ 1978 _ L „ i . 1980 1982 FIGURE I , , L ...... 1 1984 1986 1988 2 Variance of Regional Log Wage Differentials: Regions vs. Weighted MSAs Variance of logs Variance of logs 0.0035 0.012 / 0.010 'v. Regions 0.008 _ Jr ~ 0.0020 0.006 0.0015 0.004 Weighted MSAs 0.000 > 1973 I 1976 1 1979 1982 FIGURE Variance of Local Price Indexes Variance of logs SOURCE: Authors’ calculations. 1985 1988 0.0010 _ _ 0.0005 0.0000 1990 well as to be consistent with earlier work by Eberts (1989) and Browne (1989), MSAs are aggregated by proportional population weight ing to represent the nine census regions. Each of these regions contains at least one of the 21 MSAs, except for the East South Central states (Kentucky, Tennessee, Alabama, and Mississip pi). As shown below, the patterns of wage and price dispersion for MSAs and the constructed census regions are quite similar. To adjust for the effects of inflation, wages are deflated to 1982 levels by the GDP implicit price deflator. Wage variance across regions exhibits a marked U-shaped pattern between 1972 and 1990, with wages converging during the first half of the period and then diverging thereafter (figure 1). From 1972 to the trough, the vari ance of wages is cut roughly in half. By the end of the period, the variance surpasses the level at which it started in the early 1970s. This convergence and subsequent divergence is ap parent for single and grouped years. The same basic pattern of wage dispersion is found in the MSAs aggregated to simulate the census regions (figure 2). The level is generally lower for the weighted MSA results because metropolitan wages are more alike across re gions — even though major MSA wages gener ally reflect their region’s differential. The differ ences between the two variances, shown in figure 2, reflect the degree to which regional wage differentials are altered by including smaller MSAs and mral areas. These patterns are generally consistent with the convergence/ divergence phenomenon reported by Eberts (1989) using the May CPS, and by Browne (1989) using per capita income (of which wages ac count for a large portion). 3 Prices Indexes that measure regional cost-of-living dif ferences as well as price changes over an ex tended period are not readily available. The BLS releases a Consumer Price Index for selected MSAs that records price changes for each area over time. However, the index is constructed to ignore price differences across these MSAs by benchmarking the series to 100 on the basis of 1982 to 1984 prices within each area. In order to include this component in a regional price in dex, we rebenchmarked these indexes using the relative metropolitan cost-of-living index from the 1981 BLS Report on Family Budgets. (The report has not been updated because funding for the project was eliminated.) The F I G U R E gated to the nine census regions exhibits a similar pattern. The largest component of the regional price index — and the one that accounts for most of the difference in prices across MSAs — is the cost of housing services. This measure is closely linked to the general price index, as indicated by correlations between the MSAs’ relative prices and housing costs of greater than 0.95 in each period. Thus, locally adjusted wages can be viewed as wages adjusted for local housing prices. 4 Locally Adjusted W age Variation Variance of logs 0.012 0.010 0.008 0.006 0.004 Locally Adjusted Wages 0.002 0.000 1972 1974 1976 1978 1980 F I G U R E 1982 1984 1986 1988 1982 1984 1986 1988 5 Factors in the Convergence of Locally Adjusted W ages Logs 0.016 0.014 0.012 0.010 0.008 0.006 0.004 0.002 0.000 - 0.002 1972 1974 1976 1978 1980 Locally adjusted wages refer to wages divided by local prices (including cost-of-living differ ences between localities). For comparisons between census regions, these wages are ag gregated in the same fashion as regional wages and prices Locally adjusted wages do not con form to the marked pattern of regional conver gence/divergence found in nationally adjusted wages and prices. Rather, the measure gener ally converges throughout the entire period. This is most pronounced for MSAs aggregated to the nine census regions, as shown in figure 4. From peak to trough, the variance of locally adjusted wages declines by almost 50 percent. This tendency toward convergence is confirmed at the metropolitan level for locally adjusted wages, except for a slight increase in the last pe riod. In order to be consistent with the previous literature, we focus on census regions below. a. Covariance o f nationally adjusted wages, local prices. SOURCE: Authors' calculations. metropolitan cost-of-living differences are based on a consumption basket appropriate for a fourperson family with an intermediate income. The index that we construct identifies metropolitan price differences over time, which we use as our deflator instead of relying on a national price deflator.5 Figure 3 tracks the variance in the log of the metropolitan price index over the same period as wages. We use the log form to be consistent with the use of log wages to measure wage dis persion. Note that the dispersion of local prices follows a similar U-shaped path, declining dur ing the first half of the period, reaching its na dir in the early 1980s, and then returning to previous levels. The dispersion of prices aggre ■ 5 As already noted, wages are deflated by the GDP implicit price http://fraser.stlouisfed.org/deflator to eliminate the effects of inflation. Federal Reserve Bank of St. Louis III. The Relationship between Wages, Prices, and Locally Adjusted Wages Figure 5 offers a complete picture of locally adjusted wages and its two components. As dis cussed earlier, the variance of the log of nation ally adjusted wages is considerably smaller than both the variance of log prices and the variance of the log of locally adjusted wages. The covari ance of the log of nationally adjusted wages and prices is positive, but smaller than the individual variances. This positive but weak covariance sug gests that MSAs with above-average rents also pay above-average wages, which is consistent with Gabriel, Shack-Marquez, and Wascher’s (1988) finding that higher rents are only weakly associated with higher wages. The relationship of the dispersion of wages, prices, and locally adjusted wages can be seen by decomposing the variance of the log of locally adjusted wages. (1) var[\n(wr/pr)] = var[\n(wr)] + var 1ln( p r ) ] sex. The dispersion of regional wage differen tials over time is decomposed into two compo nents: changes in worker characteristics and changes in labor market implicit valuations of worker characteristics (as measured by regres sion coefficients). Because we are not the first to attempt to account for workforce differences, we start by reviewing the existing literature. -2 cov[ In(wr), ln(p r) ], where r refers to the region, wr is the average wage in region r, and p r is the relative price level in region r. The variances are calculated independently for each year. Thus, the change in the variance of price-adjusted wages between two time periods (0 and 1) can be decomposed for each region as follows, dropping the redun dant r subscripts: ( 2) var[\n(wx/px)]~ var[\n(wQ/pQ)] = [var[\r\(u\ )] - var[ ln (^ 0) l } ,+ {var[ ln(/?j )] - var[ ln(/>0) l } - 2 {cov[ l n ( ^ ), ln(pj ) ] - cov[\n(ivQ), ln (p 0) l }. Note that price-adjusted wages can converge, while nominal wages and prices diverge only if the difference in the covariances is at least half the magnitude of the two differences in vari ances. Local amenities, whether the benefits are accmed by firms or workers, provide a link be tween wages and prices. Thus, there are substan tial grounds for the existence of non-negligible covariances between relative wages and prices. Figure 5 does identify a significant covariance, the rise of which is coincident with the increase in wage variation among regions. IV. Does Regional Wage Convergence/ Divergence Represent Shifting Incentives? In order to explore regional wage differences, observationally equivalent workers must be compared. The role of regional workforce dif ferences in the relative wages of regions should be isolated from pay differentials that compara ble workers would receive in other regions. We account for most sources of wage disparity by evaluating the typical differences in returns associated with worker characteristics, including education levels, experience, industry, race, and Previous Studies Previous studies examining the relative size of the two components of wage differentials have focused primarily on explaining differences be tween the South and other regions of the United States. Sahling and Smith (1983), for example, compare the southern states with four other re gions of the country: the Northeast, the West, the North Central states, and the New York metropolitan area. They estimate separate priceadjusted and nominal wage equations using a sample of residents from 29 of the largest MSAs in these five regions. Worker-attribute variables include measures of schooling, experience, race, occupation, sex, industry, job status, and union membership. Using two cross sections of data from the May 1973 and May 1978 CPS, the authors conclude that cost-of-living adjust ments dramatically increased the wages of southern workers relative to their counterparts across the United States. Farber and Newman (1987) extend Sahling and Smith’s analysis to look explicitly at changes in characteristic prices over time. In addition to analyzing regional wage differentials in two separate years (1973 and 1979), they estimate the changes in differentials between the two years for various pairs of regions. Their results show that more than half of the predicted shifts in South/non-South wage ratios can be accounted for by changing relative returns to worker characteristics. Using the same framework adopted in the current paper, Eberts (1989) examines the sources of nominal regional wage convergence and divergence on a full sample of workers from the May CPS. He finds that differences in the returns to worker characteristics account for both the convergence in regional wages from 1973 until 1982 and the divergence thereafter. Other studies, using similar techniques but more detailed data, do not necessarily agree with the conclusion that characteristic prices explain regional wage differentials. Bellante (1979) and Gerking and Weirick (1983), for example, find that regional wage differences result primarily from variations in the level of worker character istics. These findings leave open the possibility that both characteristic prices and levels are likely sources of regional wage differentials.6 Defining Sources of Wage Differentials Following the human capital specification of Hanoch (1967) and Mincer (1974), we specify the logarithm of individual wages — expressed in either nominal or price-adjusted terms — as a function of various worker attributes, includ ing education level (entered as dummy variables for the completion of four levels of schooling, from high school to graduate studies), and po tential experience (age, minus years of educa tion, minus six, entered as a quadratic). Dummy variables indicating race, gender, occupation, and industry are also included as recognized factors in individual earnings. Time dummies are incorporated to account for aggregate fluc tuations, including the business cycle, within each of the pooled three-year periods.7 We estimate hedonic wage equations sepa rately for each period and for each of the 21 MSAs. Prior to the estimation, individual wages are deflated by either the national or local price index, as described previously. We weight re gional wages and estimated wage components by their respective population shares in order to construct a regional measure. The East South Central region is excluded from the analysis be cause no metropolitan area price data were available for cities in these states. We then compare the regional wage estimates to na tional estimates based on the same regression and the sample of workers from all 21 MSAs. The technique used to account for the two sources of wage differentials follows the ap proach of Oaxaca (1973), with modifications by Sahling and Smith (1983). The decomposition assumes that y, the logarithm of wages, can be appropriately described as a function of the worker and industry characteristics discussed earlier ( X and the hedonic labor market val uation of each characteristic ( b ): (3) y = bXi + ui . ■ 6 Dickie and Gerking (1988) provide a comprehensive and insight ful critique of the literature. ■ 7 If business cycle fluctuations alter general earning levels, then ignor ing that variation would result in inconsistent estimates. Dummy variables http://fraser.stlouisfed.org/ account for the mean aggregate differences between the two periods. Federal Reserve Bank of St. Louis Estimating a well-specified earnings equation for each region accounts for the value associated with regional concentrations of particular work force traits by identifying the^average valuation of these traits in the region ( bSt for region S at time 0- Using y for ln (^ ), we can decompose the percentage difference in wages between the regions during one time period as follows: (4 ) ( yst ~ y N t) = ( ^st ~ bN t) XNt + ( XSt - X Nt) bNt+ ( bSt - bNt) (x st - XNt). The first term on the right side accounts for the difference in labor market valuations of worker attributes between a region and the base. The second term denotes the difference in levels of worker and industry characteristics. The third term, a remainder, is generally assumed to be small and in fact proved trivial in our analysis. Below, we examine the relative contribution of the first two right-side components of equation (4) over time for both wage series. Decomposing the Variance of Regional Wages Having decomposed the regional wage differen tials into separate factors, a number of variance decompositions are possible. The traditional decomposition focuses on the variance of the first two terms of equation (4), neglecting the third term (the interaction term): (5) v a r{ y st - yNt ) = v a r[ (b st- b Nt) XNt ] + var[ ( X SI- X NI) bNt] + 2 cov [ ( bSt- bNt) X Nt, ( x 5i - XNt) bNt ] + interaction term. This approach generally yields the correct inter pretation of the sources, although it is incom plete due to exclusion of the interaction term. If the covariance is significant but is not re ported, then the decomposition is even less complete. We report the results of this decom position for purposes of comparison with the existing literature. The focus of our paper, however, is on re gional wage differences when local prices are factored in, so it is valuable to consider how 33 price-level corrections affect the variance de compositions. The adjustment for local prices is applied to individual earnings as follows: 6 ( ) y*it = yit / Pst’ where p St is constant within the locality at time t and y*t and yit are the vectors of individuallevel wage observations for region S at time t. Given that each MSA’s wage equation is esti mated independently for every three-year block of time, we can expect that the transformation of the dependent variable will adjust the bSt estimates proportionally. This follows from the normal equations for annual estimates: (7) b l = ( X ' X ) - ' X ' j - = j - t sr With pooling over three years of data, the adjust ment would be a weighted average of the rele vant p St’s. Thus, in terms of the decomposition of regional wage differentials shown in equa tion (4), only the price term [(bSt- bNt) XNt\ and the interaction term reflect the adjustment of wages for the local cost of living. Consequently, only the valuation compo nent of the variance decomposition (equation 151) would be altered, perhaps indicating that cost-of-living adjustments affect the variance of regional wages through the valuation of skills alone. These variance terms, however, are not a complete decomposition of the sources of re gional wage differentials unless the covariance between the valuation and workforce charac teristic terms is zero. The covariance term rep resents the correlations between regional concentrations of labor skills (or other charac teristics) and differentials paid to those skills. A simple supply and demand model without per fectly elastic or inelastic demand or supply would predict non-zero correlations. The co variance would then be reduced to the extent that labor or firm mobility eliminated either re gional skill concentrations or the wage differen tials paid to specific skills. However, regional production and consumption amenities should ensure that this covariance is non-zero. The problems with the commonly applied decomposition in this context suggest the need for an alternative decomposition that accounts for the covariance term in a meaningful way. A more complete decomposition that satisfies this requirement is provided by the covariances between the dependent variable and the addi tive factors.8 To simplify the notation of the price and quantity equation (4), let y be regional differentials, b be the valuation term, 5c be the workforce characteristics term, and i be the in teraction term. The interaction term, which we still expect to be small, is included so that the definition of the decomposition is complete. In place of equation (5), applying a covariance decomposition to the factors shown in equa tion (4) results in (8) v a r (y ) = co v (y ,b ) + co v (y ,x ) + c o v { y ,i). The three decomposition terms in equation (8) are easily interpreted as the effect of a factor on the dependent variable after covariances with all other factors have been accounted for. The first term represents the effect of labor market valuations, the second represents the effect of labor force differences, and the third is the ef fect of the small interaction term. Factors can be either positive or negative, depending on whether they augment or offset the sum of the other factors contributing to the variation. If the factors are fully independent, then the decom position simplifies to the basic variance decom position for the independent variables shown in equation (5), with a covariance equal to zero. Splitting the parenthetical terms in equation (8) distinguishes the components of the three terms. Equation (9) shows that each term of this decomposition includes an expression for the relationship between valuations and work force characteristics. (9) var( y ) = [var(b) + cov(b,x) + cov(b, i ) 1 + 1v a r(x ) + cov(b,x) + cov(x ,i) ] + [var(i) + cov(b ,i) + co v (x ,i)]. Adjusting wages by a local deflator alters these correlations. Beyond this simple statistical rela tionship, factors are adjusted for the degree to which higher locally adjusted wages for skills correspond to concentrations of those skills. Regional skill concentrations are fundamentally linked to the mobility decisions of workers and firms. Locally and nationally adjusted wages should result in different decomposi tions due to the reactions of firms and workers to wage differentials. ■ 8 A similar decomposition is applied in Schweitzer (1993) to iden tify sources of earnings inequality. F I G U R E Sources of Nationally Adjusted Wage Differentials 6 Decomposition of Nationally Adjusted Wage Variation Variance of logs SOURCE: Authors’ calculations. TABLE 1 Decomposition of Nationally Adjusted Regional Wage Variation Total Variation Characteristics Valuations Covariance Decomposition 1972 1975 1978 1981 1984 1987 to to to to to to 1974 1977 1980 1983 1986 1990 0.0022 0.0017 0.0014 0.0009 0.0011 0.0034 0.0002 0.0003 0.0007 0.0004 0.0008 0.0015 0.0020 0.0016 0.0008 0.0004 0.0003 0.0020 Variances of Factors 1972 to 1974 0.0022 1975 1978 1981 1984 1987 0.0017 0.0014 to to to to to 1977 1980 1983 1986 1990 0.0009 0.0011 0.0034 0.0006 0.0002 0.0006 0.0005 0.0008 0.0014 0.0026 0.0016 0.0007 0.0005 0.0004 0.0020 SOURCE: Authors’ calculations based on data from the U.S. Department of Labor, Bureau of Labor Statistics. Figure 6 addresses the question of whether the convergence/divergence pattern of regional wage differences results more from variations in labor market valuations or from variations in attribute levels (for example, the decomposition of regional wage differentials in equation [7]). The shaded area under the curve represents the portion of the variance of log wages ac counted for by differences in labor market valu ations. The remainder of the area under the curve is the portion of the variance explained by differences in attribute levels. In some years, the covariance decomposition terms for valuations and attributes do not add up to the total variance because of the interaction term, which is not reported. It is evident from the fig ure that differences in valuations follow the same U-shaped pattern as total wage variance. O n the other hand, differences in workforce at tributes follow a generally upward trend. This suggests that the pattern of convergence and then divergence of nationally adjusted wages results more from regional labor markets’ valu ing attributes differently than from an increas ing dissimilarity of workers within regions. Nonetheless, regional differences associated with workforce attributes have been playing a growing role in regional wage differentials. Table 1 compares the covariance decompo sition results with the variances of the two significant components. In this case, the covari ance between the quantity and characteristic price component is small; thus, the variances sum to approximately the total variation and are similar to the covariance decomposition terms. This confirms Eberts’ (1989) results for nationally adjusted wages in a sample of the full-time metropolitan workforce. The results in table 1 and figure 6 can be in terpreted in two ways: Either incentives for firms to move toward lower-wage areas are growing, or local productive amenities are on the rise. Both conclusions hinge on our having captured the majority of worker productivity differences between regions with the worker attributes included in the wage equations. If significant productivity differences are not cap tured by the wage equations, and if the unob served productivity factors have been growing nationally in value, then we could mistakenly identify productivity differentials between re gions as price differences. Along these lines, Juhn, Murphy, and Pierce (1993) argue that 35 FI GURE 7 higher wage payments to unobserved skills ex plain the rise in total earnings inequality dur ing the 1980s. A final caveat to our results is that the analysis does not account for fringe benefit costs. Differences in these costs be tween regions would of course result in a dif ferent distribution of total compensation. Decomposition of Locally Adjusted Wage Variation Variance of logs Sources of Locally Adjusted Wage Differentials SOURCE: Authors’ calculations. TABLE 2 Decomposition of Locally Adjusted Regional Wage Variation Total Variation Characteristics Valuations Covariance Decomposition 1972 1975 1978 1981 1984 1987 to to to to to to 1974 1977 1980 1983 1986 1990 0.0119 0.0107 0.0072 0.0062 0.0047 0.0040 0.0005 - 0.0002 - 0.0001 - 0.0002 0.0000 0.0002 0.0120 0.0110 0.0075 0.0065 0.0048 0.0039 Variances of Factors 1972 1975 1978 1981 1984 1987 to to to to to to 1974 1977 1980 1983 1986 1990 0.0119 0.0107 0.0072 0.0062 0.0047 0.0040 0.0006 0.0002 0.0006 0.0005 0.0008 0.0014 0.0129 0.0116 0.0083 0.0075 0.0058 0.0052 SOURCE: Authors’ calculations based on data from the U.S. Department of Labor, Bureau o f Labor Statistics. While firms m ight be adjusting to these wage differentials, households should react to wages that reflect their cost of living. The pattern in the variance of locally adjusted wages is quite different from that of nationally adjusted wages. Instead of exhibiting a U-shaped pattern, locally adjusted wages steadily converge over the sample period (figure 7). Moreover, the dispersion of lo cally adjusted wages is roughly five times greater than the dispersion of nationally adjusted wages. Significantly, it is differences in labor market valu ations that explain most of the total wage vari ance. While the dispersion in labor costs relevant to firms (nationally adjusted wages) has increased in recent years, the dispersion of regional differ ences in workers’ returns to labor has declined. Comparing the covariance decomposition re sults with simple variances indicates, in this case, that accounting for covariance between factors alters our interpretation of the compo nents of the decline in locally adjusted wage dispersion between regions. Table 2 reveals that, unlike the nationally adjusted wage case, a significant negative covariance exists between the characteristic price component of regional wage differences and the regional distribution of attributes. This is evident both in the fre quently negative quantities component and in the fact that the simple variances of the compo nents substantially overshoot the total variances of locally adjusted wages. Evaluating these results in terms of worker location decisions, we find that the declining differences in factor returns between regions is consistent with workers’ moving to equalize labor market differences. A larger impetus for mobility is indicated by the greater wage varia tion between regions when cost-of-living differ ences are factored in. The mobility of house holds responding to significant, but declining, consumptive amenities in the high-price MSAs could explain this reduction in locally adjusted wage differentials between MSAs. Furthermore, the differences between locally adjusted wage differentials appear to be almost purely the re sult of differences in valuations of labor rather than differences in labor force characteristics. These differentials could encourage significant worker movement, which could lead to rising na tionally adjusted wage differences as wages are driven up in high-price areas and down in lowprice areas. But it could just as well be that firms have moved to more costly areas, driving up wages, in pursuit of an amenity that has been ris ing in value. The unobservability of the full set of amenities, either consumptive or productive, pre cludes a direct test of these explanations. V. Conclusion The theoretically surprising fact that regional wages appeared to diverge in the 1980s does not hold up when cost-of-living differences are taken into account. Our decompositions con firm that wage differences are driven by vary ing returns to worker attributes rather than by regional differences in workforce charac teristics. Further, the possibility is raised that workers and firms are optimizing over differ ent value functions (nationally versus locally adjusted wages) or different local amenities. In particular, local prices, and therefore locally ad justed wages, may be more important for work ers. The difference in the patterns of nationally versus locally adjusted wage differentials is consistent with a story of competing adjust ments rather than of slowing adjustments. However, other explanations are not elimi nated by these results, because neither the ad justment processes nor the values of amenities have been explicitly incorporated. These short comings provide an obvious direction for fu ture research. Given the limited observability of amenities, a sensible strategy would be to estimate the adjustment processes of firms and workers. This would make more explicit the link between convergence rates and differen tials in the two wage series. Although our con clusion is largely descriptive, the diverse patterns in nationally versus locally adjusted wages clearly support analyzing regional wage differentials from the perspective of both em ployees and firms. References Bellante, Don. “The North-South Differential and the Migration of Heterogeneous Labor,” Am erican Economic Review, vol. 69 (March 1979), pp. 166-75. Blanchard, Olivier Jean, and Lawrence F. Katz. “Regional Evolutions,” Brookings Papers on Economic Activity, vol. 1 (1992), pp. 1 —61. Browne, Lynn E. “Shifting Regional Fortunes: The Wheel Turns,” Federal Reserve Bank of Boston, New E ngland Economic Review, May/June 1989, pp. 27 - 40. Dickie, Mark, and Shelby Gerking. “Interre gional Wage Differentials in the United States: A Survey,” in Alan Schlottmann et al., eds., M igration a n d Labor Market Adjust ment. Boston: Kluwer Academic Publishers, 1988. Eberts, Randall W. “Accounting for the Recent Divergence in Regional Wage Differentials,” Federal Reserve Bank of Cleveland, Eco nom ic Review, vol. 25, no. 3 (1989 Quarter 3), pp. 14-26. _________ , and Joe A. Stone. Wage a n d Employ ment Adjustment in Local Labor Markets. Kalamazoo, Mich.: W.E. Upjohn Institute for Employment Research, 1992. Farber, Stephen C., and Robert J. Newman. “Ac counting for South/Non-South Real Wage Differentials and for Changes in Those Differ entials over Time,” Review o f Economics an d Statistics, vol. 69 (May 1987), pp. 215-23. Gabriel, Stuart A., Janice Shack-Marquez, and William L. Wascher. “Regional Labor Mar kets, Cost-of-Living Differentials, and Migra tion,” Board of Governors of the Federal Reserve System, Economic Activity Section, Working Paper No. 91, December 1988. Gerking, Shelby, and William Weirick. “Compen sating Differences and Interregional Wage Dif ferentials,” Review o f Economics an d Statistics, vol. 65 (August 1983), pp. 483-87. Hanoch, Giora. “An Economic Analysis of Earn ings and Schooling,” Jo u rn a l o f H um an Re sources, vol. 2 (Summer 1967), pp. 310 - 29. Juhn, Chinhui, Kevin M. Murphy, and Brooks Pierce. “Wage Inequality and the Rise in Re turns to Skill,” Jo u rn a l o f Political Economy, vol. 101, no. 3 (June 1993), pp. 410-42. Mincer, Jacob. Schooling, Experience, a n d Earnings. New York: National Bureau of Economic Research, distributed by Colum bia University Press, 1974. Oaxaca, Ronald. “Male-Female Wage Differen tials in Urban Labor Markets,” International Economic Review, vol. 14 (October 1973), pp. 693-709. Roback, Jennifer. “Wages, Rents, and Quality of Life,” Jo u rn a l o f Political Economy, vol. 90 (December 1982), pp. 1257-78. Sahling, Leonard G., and Sharon P. Smith. “Re gional Wage Differentials: Has the South Risen Again?” Review o f Economics a n d Sta tistics, vol. 65 (February 1983), pp. 131 - 35. Schweitzer, Mark E. “Accounting for Earnings Inequality in a Diverse Workforce,” Federal Reserve Bank of Cleveland, Working Paper 9314, December 1993. Shryock, Henry S., and Jacob S. Siegel. The Methods a n d Materials o f Demography, vol. 1. Washington, D.C.: U.S. Department of Commerce, Bureau of the Census, 1971. Economic Review M 1993 Quarter 2 H Using Bracket Creep to Raise Revenue: A Bad Idea Whose Time Has Passed by David Altig and Charles T. Carlstrom Required Clearing Balances by EJ. Stevens The CPI as a Measure of Inflation by Michael F. Bryan and Stephen G. Cecchetti Cyclical Movements of the Labor Input and Its Implicit Real Wage by Finn E. Kydland and Edward C. Prescott Money and Interest Rates under a Reserves Operating Target by Robert B. Avery and Myron L. Kwast M 1993 Quarter 3 Capital Requirements and Shifts in Commercial Bank Portfolios by Joseph G. Haubrich and Paul Wachtel FDICIA’s Emergency Liquidity Provisions by Walker F. Todd Efficiency and Technical Progress in Check Processing by Paul W. Bauer 1993 Quarter 4 The Inaccuracy of Newspaper Reports of U.S. Foreign Exchange Intervention by William P. Osterberg and Rebecca Wetmore Humes H 1994 Quarter 1 Institutional Aspects of U.S. Intervention by Owen F. Humpage The 1995 Budget and Health Care Reform: A Generational Perspective by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff On Disinflation since 1982: An Application of Change-Point Tests by Edward Bryden and John B. Carlson Second Quarter Working Papers Current Working Papers of the Cleveland Federal Reserve Bank are listed in each quarterly issue of the Economic Review. Copies of specific papers may be re quested by completing and mailing the attached form below. Single copies of individual pa pers will be sent free of charge to those who request them. A mailing list service for personal subscribers, however, is not available. Institutional subscribers, such as libraries and other organiza tions, will be placed on a mail ing list upon request and will automatically receive Working Papers as they are published. ■ 9405 The Federal Reserve Board before Marriner Eccles ■ 9406 Optimal Fiscal Policy when Public Capital Is Productive: A Business Cycle Perspective ■ 9407 Anticipating Bailouts: The Incentive—Conflict Model and the Collapse of the Ohio Deposit Guarantee Fund by Walker F. Todd by Kevin J. Lansing by Ramon P. DeGennaro and James B. Thomson Please complete and detach the form below and mail to: Research Department Federal Reserve Bank of Cleveland P.O. Box 6387 Cleveland, Ohio 44101 Check item(s) requested Please send the following Working Paper(s): □ 9405 □ 9406 □ 9407 Send to: Please print Name Address City State Zip L ______________________________________________________________________________ J Economic Commentary Enterprise Liability: A Prescription for Health Care Reform? by Charles T. Carlstrom July 1, 1993 Making the SAIF Safe for Taxpayers by William P. Osterberg and James B. T h o m s o n N o v e m b e r 1, 1993 T h e Evolving Loan Sales Market by Joseph G. Haubrich and James B. T h o m s o n July 15, 1993 C o m m u n i t y Lending and Economic Development by Jerry L. Jordan N o v e m b e r 15, 1993 Free Markets and Price Stability: Opportunities for Mexico by Jerry L. Jordan August 1, 1993 Replacing Reserve Requirements by E.J. Stevens D e c e m b e r 1, 1993 Assessing Real Interest Rates by John B. Carlson August 15, 1993 Airline Deregulation: Is It Tim e to Finish the Job? by Paul W . Bauer and Ian Gale September 1, 1993 T h e Decline in U.S. Saving Rates: A Cause for Concern? by Jagadeesh Gokhale September 15, 1993 Credibility Begins with a Clear C o m m i t m e n t to Price Stability by Jerry L. Jordan October 1, 1993 Th e Budget Reconciliation Act of 1993: A S u m m a r y Report by David Altig and Jagadeesh Gokhale October 15, 1993 Long-Term Health Care: Is Social Insurance Desirable? by Jagadeesh Gokhale and Lydia K. Leovic D e c e m b e r 15, 1993 Monetary' Policy and Inflation: 1993 in Perspective by Gregory A. Bauer and John B. Carlson January 1, 1994 Report of the Fourth District Economists’Roundtable by Michael F. Bryan and John B. Martin January 15, 1994 Are Service-Sector Jobs Inferior? by M a x D u p u y and Mark E. Schweitzer February 1, 1994 Th e National Depositor Preference L a w by James B. T h o m s o n February 15, 1994 Issues in C R A Reform by Mar k S. Sniderman March 1, 1994 Back to the Future: A V i e w of Prospective Deficits through the Prism of the Past by David Altig and Jagadeesh Gokhale March 15, 1994 Central Bank Independence by O w e n F. H u m p a g e April 1, 1994 Health Care Reform from a Generational Perspective by David Altig and Jagadeesh Gokhale April 15, 1994 Lessons from the Collapse of Three State-Chartered Private Deposit Insurance Funds by Walker F. T o d d M a y 1, 1994 Assessing Progress toward Price Stability: Looking Forward and Looking Backward by John B. Carlson M a y 15, 1994 T he Go ve rnment’ s Role in the Health Care Industry: Past, Present, and Future by Charles T. Carlstrom June 1, 1994 Are the Japanese to Blame for O u r Trade Deficit? by O w e n F. H u m p a g e June 15, 1994 A Beginner’ s Guide to the U.S. Payments System by Paul W . Bauer July 1, 1994