The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of St. Louis REVIEW Federal Reserve Bank of St. Louis P.O. Box 442 St. Louis, MO 63166-0442 M ay / Ju ne 2 0 1 3 Vo l u m e 95 , Nu m b e r 3 REVIEW Financial Regulatory Reform: A Progress Report Glenn Hubbard Big Banks in Small Places: Are Community Banks Being Driven Out of Rural Markets? R. Alton Gilbert and David C. Wheelock Foreign Currency Loans and Systemic Risk in Europe May/June 2013 • Volume 95, Number 3 Pınar Yeşin Labor Mismatch in the Great Recession: A Review of Indexes Using Recent U.S. Data Maria E. Canon, Mingyu Chen, and Elise A. Marifian REVIEW Volume 95 • Number 3 Director of Research Christopher J. Waller Policy Adviser Cletus C. Coughlin Deputy Director of Research 181 Financial Regulatory Reform: A Progress Report Glenn Hubbard David C. Wheelock Review Editor-in-Chief William T. Gavin Research Economists Richard G. Anderson David Andolfatto Alejandro Badel Subhayu Bandyopadhyay Maria E. Canon YiLi Chien Silvio Contessi Riccardo DiCecio William Dupor Carlos Garriga Rubén Hernández-Murillo Luciana Juvenal Kevin L. Kliesen Fernando M. Martin Michael W. McCracken Alexander Monge-Naranjo Christopher J. Neely Michael T. Owyang Adrian Peralta-Alva B. Ravikumar Juan M. Sánchez Rajdeep Sengupta Yongseok Shin Daniel L. Thornton Yi Wen Christian M. Zimmermann 199 Big Banks in Small Places: Are Community Banks Being Driven Out of Rural Markets? R. Alton Gilbert and David C. Wheelock 219 Foreign Currency Loans and Systemic Risk in Europe Pınar Yeşin 237 Labor Mismatch in the Great Recession: A Review of Indexes Using Recent U.S. Data Maria E. Canon, Mingyu Chen, and Elise A. Marifian Managing Editor George E. Fortier Editors Judith A. Ahlers Lydia H. Johnson Graphic Designer Donna M. Stiller Federal Reserve Bank of St. Louis REVIEW May/June 2013 i Review is published six times per year by the Research Division of the Federal Reserve Bank of St. Louis and may be accessed through our website: research.stlouisfed.org/publications/review. All nonproprietary and nonconfidential data and programs for the articles written by Federal Reserve Bank of St. Louis staff and published in Review also are available to our readers on this website. These data and programs are also available through Inter-university Consortium for Political and Social Research (ICPSR) via their FTP site: www.icpsr.umich.edu/pra/index.html. Or contact the ICPSR at P.O. Box 1248, Ann Arbor, MI 48106-1248; 734-647-5000; netmail@icpsr.umich.edu. The views expressed are those of the individual authors and do not necessarily reflect official positions of the Federal Reserve Bank of St. Louis, the Federal Reserve System, or the Board of Governors. Single-copy subscriptions (U.S. addresses only) are available free of charge: https://research.stlouisfed.org/publications/review/subscribe/. General data can be obtained through FRED (Federal Reserve Economic Data), a database providing U.S. economic and financial data and regional data for the Eighth Federal Reserve District. You may access FRED through our website: research.stlouisfed.org/fred. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Please send a copy of any reprinted, published, or displayed materials to George Fortier, Research Division, Federal Reserve Bank of St. Louis, P.O. Box 442, St. Louis, MO 63166-0442; george.e.fortier@stls.frb.org. Please note: Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Please contact the Research Division at the above address to request permission. © 2013, Federal Reserve Bank of St. Louis. ISSN 0014-9187 ii May/June 2013 Federal Reserve Bank of St. Louis REVIEW Financial Regulatory Reform: A Progress Report Glenn Hubbard The 2007-09 financial and economic crisis was the result of a lack of effective regulation. The author addresses the problems with regulations in effect at the time of the crisis and offers proposals for regulation reform to address future crises. He notes that reforms should be based on solid principles, including reduction of system risk and contagion and increased transparency to promote investor protection. Any new financial regulatory structure must be able to achieve these goals, while acknowledging and managing trade-offs between enhancing accountability and mitigating systemic risk from contagion. (JEL G01, G18, G28) Federal Reserve Bank of St. Louis Review, May/June 2013, 95(3), pp. 181-97. he conventional assessment of the 2007-09 financial and economic crisis places blame on a dearth of regulation. That assessment is simplistic at best and entirely inaccurate at worst. The truth is that the financial crisis is the result of a lack of effective regulation. Several themes emerge from the crisis. First, we need more effective regulation. Although we need new regulation in some previously unregulated areas, the crisis has shown that the most precarious sectors of our financial system are those already subject to a great deal of regulation— regulation that has proved woefully ineffective. Any call for further reform means that new or revised regulations should be based on solid principles, chief among them the reduction of systemic risk and contagion. Second, we must increase transparency in the financial system to promote investor protection. More information enables the market to price assets, risk, and other relevant inputs more accurately. Much of the recent crisis can be attributed to a lack of critical information (and perhaps, in some cases, misinformation). Third, we must build a financial regulatory structure able to achieve these goals. Finally, we need to acknowledge and manage trade-offs between enhancing accountability for individual institutions and mitigating systemic risk from contagion. T Glenn Hubbard is the dean and the Russell L. Carson Professor of Finance and Economics in the Graduate School of Business and a professor of economics in the Graduate School of Arts and Sciences of Columbia University; he is a research associate at the National Bureau of Economic Research; he is also a director of a regulated financial services firm, MetLife, and has consulted for financial services firms. This article was based on the author’s Homer Jones Lecture, “A Roadmap for Financial Regulatory Reform,” at the Federal Reserve Bank of St. Louis, April 3, 2013. © 2013, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW May/June 2013 181 Hubbard THE CRISIS AND THE REGULATORY RESPONSE Severity of the Crisis The 2007-09 financial crisis was the most serious such event since the Great Depression. The crisis manifested itself in credit losses, write-downs, liquidity shocks, deflated property values, and a contraction of the real economy. The sharp contraction in U.S. gross domestic product in 2009 can be traced to the adverse effects of the crisis on household consumption and business investments. Costs directly attributable to the crisis include outlays by the federal government, including the Troubled Assets Relief Program (TARP) and the stimulus package passed in February 2009. In the housing sector, banks took advantage of low interest rates and securitization opportunities to institute relaxed lending standards that drove the boom in mortgage lending from 2001 to 2006. Although the number of households in the United States increased only marginally between 1990 and 2008, the aggregate mortgage debt outstanding more than quadrupled during that same period. Increased borrowing by U.S. households was partially offset by climbing asset prices. However, the period of rising property values came to a close after the second quarter of 2006, with home prices eventually falling by a third by the end of 2008. The burst of the housing bubble virtually eliminated construction and sales activity for a while. The percentage of delinquent mortgages is at an all-time high, and more than 20 percent of all mortgages are in a negative equity position. The financial crisis of 2007-09 and its continuing aftermath have complex origins, but those origins share a four-letter word: risk. The mispricing of risk—with inflationary consequences for asset prices in the boom, a downward spiral of collapsing asset prices and economic activity in the bust, and contagion in the unwinding—must be central to economic analysis of and policy responses to the crisis. Underlying factors include (i) global saving and investment imbalances that contributed to low real interest rates and risk premia in international capital markets for many years, (ii) excessively expansionary U.S. monetary policy in the years 2003-05, and (iii) significant gaps in regulation in theory and practice. For the purpose of this article, I concentrate on some key weaknesses in regulation and components of regulatory reform related to capital adequacy, resolution processes for insolvent institutions, and the organization of regulation and policies to manage contagion during a financial crisis.1 Regulatory Goals and Principles An obvious opening question in a debate over major regulatory reform is “What problem are we trying to solve?” Observing the discussion leading up to and following the passage of the Dodd-Frank Wall Street Reform and Consumer Protection Act in 2010, that question is tough to answer. For many economists and policymakers concerned about housing finance, the law was largely silent. The focus on lines of business in the so-called Volcker rule appears somewhat disconnected from the painful problems actually experienced during the financial crisis. And, as I argue below, we must decide whether we are more interested in policies to address the likelihood of failure of individual institutions or the likelihood of contagion across assets, markets, and institutions.2 182 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard Going a step further, effective regulatory reform can occur only when policymakers take account of fundamental regulatory principles. The most important of these principles is that regulation should reduce systemic risk. Systemic risk is the risk of collapse of an entire system or entire market, exacerbated by links and interdependencies; with systemic risk, the failure of a single entity or cluster of entities can cause a cascading failure. There are at least five externalities particular to financial markets that contribute to systemic risk. First, the spread of speculative information through the market can create the perception that economic difficulties affecting one financial institution will also affect similarly situated firms. Second, customers of failed institutions may subsequently find themselves in a less-friendly market when seeking to redirect their business. Third, considerable interconnectedness exists among the financial institutions participating in modern financial markets, so the failure of one firm can affect many others. Fourth, falling asset prices and resulting liquidity constrictions may create a negative spiral. Fifth, falling asset prices and liquidity crises may cause institutions to become reluctant to extend credit. Regulation may be legitimately imposed for a variety of other reasons. Disclosure is important for investors’ well-being, in view of the potential for an individual investor to undertake a less-than-adequate investigation before making an investment decision. Further, improving the quality of information and the ease with which it is disseminated is important in protecting consumers from instances of unfair, predatory, and fraudulent behavior. Regulation is also useful in mitigating the risk associated with an investor giving funds to an agent with only very limited control over how this investment is directed. Regulation is likewise important for opening up access to the financial markets, permitting new entrants to join established players, and thereby increasing competition. Finally, regulation can be used effectively to limit the influence of moral hazard that arises from state-provided safety nets and, in particular, to ensure that firms and capital suppliers are not permitted to take advantage of taxpayer support and engage in undue risk-taking.3 A final principle of regulation applies to all the other principles as well: the cost-benefit rule. That is, a regulation should be promulgated only when its benefits outweigh its costs. Furthermore, if different kinds of regulation can achieve the same benefit, the regulation with the least cost should be adopted (a point that may seem obvious, but given recent regulatory actions, bears repeating).4 REDUCING SYSTEMIC RISK Again, the most compelling justification for financial regulation is the need to reduce externalities—particularly systemic risk. I now consider measures to reduce systemic risk across important sectors of the financial system: capital adequacy requirements, the regulation of nonbank institutions, and the resolution process for insolvent financial institutions. Regulation of Capital Historically, capital regulation has been the dominant regulatory mechanism for constraining risk-taking by banks.5 By providing a cushion against losses, capital is supposed to act as a Federal Reserve Bank of St. Louis REVIEW May/June 2013 183 Hubbard first line of defense against bank failures and their secondary consequences for systemic risk. Yet the existing capital regime—effectively established by the Basel Capital Accords—failed to prevent several of the largest U.S. and European financial institutions from failing or becoming distressed to the point that they required government bailouts. Accordingly, it is important to consider major structural weaknesses in the capital-regulation framework, focusing on institutional coverage, calibration, timing effects, the risks of large institutions, framework design, and capital composition. Institutional coverage. Until the crisis, it was well understood that firms that were not regulated as banks (or thrifts) and not subject to capital regulation were excluded from the Federal Reserve’s safety net. The Fed’s emergency measures during the crisis have upended this understanding. These measures may have been justified by the exigencies of the crisis, but they have created structural moral hazards and impediments to a level playing field to the extent that institutions with access to the Fed’s safety net are not subject to capital regulation. Looking beyond the crisis, we need to realign the institutional costs and benefits of capital regulation. Institutions with the ability to borrow from the Federal Reserve in its role as the lender of last resort should be subject to some form of capital regulation. Calibration. Despite the critical role played by capital adequacy in the regulatory framework, existing capital requirements were set without an explicit link to a target standard of solvency for individual banks or for the system as a whole. Though an understandable reaction to the overleveraging of the system would be to raise capital requirements across the board, the lack of empirical research on capital calibration suggests that the costs and benefits of higher capital requirements for banks are uncertain. Recent arguments (see, for example, Admati et al., 2010) have emphasized that banks cannot simply argue that raising additional equity capital will reduce return on equity and thereby investment in banking. This conclusion arises from the observation that, all else equal, higher levels of bank equity reduce both risk and the costs of leverage. And that conclusion—that riskadjusted return on equity is the more relevant construct—is correct. From the perspective of economists, the Admati et al. (2010) argument begins with the claim that the differential cost of equity financing relates to the “riskiness” of equity relative to debt (with the attendant claim that if equity is mandated to rise, the cost of loans will rise). Of course, all else equal, if banks held more equity capital relative to risky assets, then the “risk” of equity would fall. Indeed, when the textbook conditions of Modigliani and Miller (1958) hold, banks’ overall cost of financing is invariant to leverage. This point is correct (and corroborated by empirical evidence; see, for example, Hanson, Kashyap, and Stein, 2011). It is also, however, a straw man and not a serious contention in the debate over bank equity capital requirements. The question for economists and regulators is whether the Modigliani-Miller (1958) conditions are a good approximation for the corporate financing environment banks face. An obvious departure from the basic Modigliani-Miller conditions is the presence of corporate taxation of equity returns—specifically, that payments of interest to debt holders are deductible from the corporate tax, while payments of dividends to equity holders are not. If, for example, equity replaced the firm’s long-term debt in a bank’s capital structure, the bank loses the value of tax deductions of the associated interest payments. 184 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard But, beyond taxes, the Modigliani-Miller (1958) framework faces challenges when applied either to arguments in banking theory or in current bank regulatory discussions about the costs of raising equity capital. Essentially, virtually all of contemporary banking theory—whether based on models of costly state verification, adverse selection, or agency costs—begins with this understanding. And, as I argue below, one cannot assume away equity financing costs related to asymmetric information simply because changes in equity capital arise from regulatory changes. That costs of asymmetric information raise equity financing costs in banking is uncontroversial to financial economists. The classic “lemons” model of Myers and Majluf (1984) demonstrates adverse selection costs of raising external equity (through underpricing of unobservably healthy banks) or applies more generally to junior securities (say equity as opposed to senior debt); see also Fazzari, Hubbard, and Petersen (1988). Such models calibrate costs of adverse selection in the pricing of an equity offering (or, put differently, in a higher incremental cost of new equity). Alternatively, equity financing can be modeled as a costly source of new funds because of ex post costs of verification, another type of asymmetric information. Classic models by Diamond (1984) and Gale and Hellwig (1985) demonstrate that banks’ use of debt contracts economizes on these costs.6 Further extensions to models of banking (see, for example, Calomiris and Kahn, 1991, and Diamond and Rajan, 2000) use agency costs to motivate demandable debt contracts for banks supported by bank equity capital. In such models, increases in the ratio of equity to risky assets would raise the cost of equity financing and reduce the volume of bank lending. It is important to note that costs of equity financing arising from asymmetric information are not eliminated simply because regulation mandates a higher ratio of equity to risky assets. Two points are important here. First, even if all banks were forced to raise equity simultaneously, banks in better condition have an incentive to spend more on underwriting (to give investors that information), raising the cost of new equity. Second, the regulatory change targets the ratio of equity to risky assets. Again, heterogeneity in bank health (known to bank managers, though ex ante not known to potential investors) would generate differences in the degree to which new equity would be raised (with the costs described above) versus shrinking risky assets. Banks in (unobservably) better condition would, all else equal, likely shrink assets to avoid the costly dilution of shareholders (see, for example, Aiyar, Calomiris, and Wieladek, 2012). That is, higher capital requirements applying to all banks do not eliminate costs of adverse selection and, by extension, costs of new equity financing. This second cost would be mitigated to the extent that regulators focused attention on a dollar target for new equity capital, as with the “stress tests” in the United States. Two categories of empirical research are useful for understanding these arguments. First, many studies have documented the relative costliness of external equity financing 7 and of more junior as opposed to more senior debt claims.8 Second, evidence that credit supply responds to a loss of equity capital or to a higher required ratio of equity to risky assets necessarily demonstrates that equity financing is costly.9 To the extent that bank equity financing is costly, a contraction in bank credit supply is socially costly when there is limited substitutability of bank credit and other funding to some or all borrowers (see the reviews in Hubbard, 1995; Hubbard and O’Brien, 2013; and Hubbard, Federal Reserve Bank of St. Louis REVIEW May/June 2013 185 Hubbard O’Brien, and Rafferty, 2013). Both theoretical and empirical research has highlighted the imperfect substitutability of bank and non-bank sources of funds (for example, securities). Bank lending involves more information-intensive screening and contracting, and many borrowers face high information-related costs in non-bank lending.10 The flip side of this point is, of course, that if available non-bank sources of funds are relatively close substitutes for bank credit, an increase in bank equity capital requirements will, all else equal, raise the volume of lending in those “close substitute” channels. It is to this link between changes in capital requirements and shadow banking to which I turn below. Before moving on, it is important to note that the argument that raising bank equity is costless is logically inconsistent with any policy argument for so-called macroprudential capital regulation. For such regulation to be effective, it is necessarily the case that (i) equity financing must be costly relative to other forms of financing and (ii) changes in bank equity capital affect lending (that is, capital requirements are binding and bank and non-bank sources of funds are imperfect substitutes). But, of course, these very factors relate to the social cost of higher capital requirements. Timing effects. Another feature of the current capital-regulation framework is that minimum capital levels are fixed, while bank losses (or adverse earnings events) vary considerably over the economic cycle. The implication is that solvency standards are not constant during an economic cycle but are dependent on the “state of the world.” The solvency level of a given capital requirement depends critically on the period over which it is calibrated and on assumptions of the state of the world going forward. In view of the cyclical nature of bank losses, the effect of a fixed capital requirement is to force banks to raise capital in the downturn as losses mount and capital levels are depleted. A key revision to the existing framework for the regulation of capital adequacy to be studied should be a shift to time-varying capital requirements. An alternative to letting capital requirements fall during a downturn would be to allow—or require—banks to hold some form of contingent capital that can be called upon as losses mount. Countercyclical capital ratios can be achieved in two ways. The first would be to encourage dynamic provisioning. This encouragement could be done without conflicting with existing securities regulation or accounting standards by providing that additional reserves over “known” losses did not run through the income statement but rather constituted a special appropriation of retained earnings. The second way would be to require contingent capital.11 One mechanism for reducing the cost of higher equity capital is to mandate the use of contingent capital that would increase a bank’s equity when some prespecified action occurs. One manifestation, contingent convertibles (see Flannery, 2005, or Calomiris and Herring, 2011), would convert into bank equity if regulatory capital (or in some forms, stock market value) declines below a prespecified value. An alternative manifestation is capital insurance (see Kashyap, Rajan, and Stein, 2008), in which insurance provided by investors would pay off to a bank when a prespecified threshold event occurs. In contrast to the claims in Admati et al. (2010), contingent capital offers important advantages in a reform to increase bank equity capital. Focusing on higher equity alone cannot fully solve the difficulty of maintaining a desired level of capital. This difficulty arises because the concept of shareholders’ equity used in the Basel Accord lags its time value (relying on account186 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard ing principles that combine book value, fair value, and market value constructs). As a consequence, regulators must keep pace with financial innovation in security design. It is, of course, hard for regulators to assess the desired amount of bank equity capital both at a point in time and over time. Well-designed contingent capital requirements can foster incentives for banks to maintain the right amount of capital (in the combination of equity plus contingent capital). Importantly, relative to our earlier discussion of costly equity financing, contingent capital offers a key benefit: Issuing contingent capital need not reduce shareholder value, while mandated equity issues necessarily do so (through both the loss of interest deductions for corporate tax purposes and costs arising from asymmetric information in raising equity). Calomiris and Herring (2011) suggest market-based measures as a trigger for the conversion of contingent capital (as contingent convertible debt) to equity stock prices. Specifically, they suggest using a 90-day moving average of the ratio of the market value of equity to the sum of the market value of equity and the face value of debt. Given the advantages of contingent capital in mitigating costs of additional equity capital, serious consideration should be given to its design. Systemically important institutions. The crisis, so far, has disproportionately affected the largest U.S. financial institutions. At the same time, the initial capital injections from the TARP were also concentrated in the largest U.S. banks. Very large banks may pose unique risks to the government because of their systemic consequences. As a result, unless policymakers comprehensively address the problem of too big to fail, one might argue that a large or important bank should be required to hold a larger capital buffer. The flip side of the point is also worth making: To the extent we are willing to let very large banks fail, the high costs of extra capital are less necessary. And, as I argue below, they are very imperfect protection against contagion. Having said that, certain types of financial institutions generally do not pose systemic risk to the financial system, including asset managers (mutual fund managers, as well as managers of private funds, such as hedge funds and private equity funds) and traditional insurers, since their bankruptcy would not set off a chain reaction of financial institution failures.12 Capital composition. While most of the debate about the Basel framework has focused on the risk assessment of individual banks (which is reflected in the denominator of the Basel capital ratio), the crisis has also raised new concerns about what “counts” as capital in the numerator of the ratio. At present, the regulatory definition of Tier 1 capital is not consistent with tangible common equity, the principal accounting measure of shareholders’ exposure to losses. It is also different from Tier 1 common capital, a new definition of capital used in the “stress test.” We need a new and consistent definition of capital. From costly bank equity capital to “shadow banking.” The costs of new equity financing I have identified raise concerns about the effects of large increases in equity capital relative to assets on bank lending. For very large banks lending to very large firms with access to non-bank sources of credit, the higher bank cost of financing puts banks at a competitive disadvantage, leading to a search for ways to increase bank leverage. As Hanson, Kashyap, and Stein (2011) point out, there is a pronounced inverse relationship between the ratio of equity to assets and bank size. This search manifests itself, inter alia, in the quest for lending activity outside the regulated banking sector—“shadow banking.” And, as I argue below, arguments for macroprudential regulation of bank equity capital requirements extend to more than just deposit-taking banks. Federal Reserve Bank of St. Louis REVIEW May/June 2013 187 Hubbard The experience of the 2007-09 financial crisis offers lessons of the costs of “runs” and financial fragility in the shadow banking system.13 As many economists have emphasized (see particularly Gorton and Metrick, 2012), the collapse of the asset-backed securities market had in common with a textbook bank run the inability of purchasers of asset-backed securities to renew short-term financing. Mechanically, the bank-run-like removal of short-term lending in shadow banking (the asset-backed securities market) can be understood in the context of “haircuts” in repurchase agreements used to finance conduits holding asset-backed securities in the shadow banking system. A haircut is akin to a down payment constraint, an amount to be posted when an investor borrows in the repo market. Prior to the financial crisis, haircuts on top-rated asset-backed securities were at most 2 percent. In the thick of the crisis, haircuts on both asset-backed securities linked to subprime mortgages and consumer asset-backed securities increased substantially, in some cases to more than 50 percent. This change led to liquidations, further reducing asset prices. Any change in bank equity capital requirements that does not incorporate the shadow banking system ignores important systemic risk; indeed, it is possible that a substantial increase in capital ratios could increase systemic risk by expanding the shadow banking system. To be effective, such regulation would need to extend to the shadow banking system through regulation of haircuts. To be specific, consistent capital regulation reform in a macroprudential setting would require that a bank loan or the same holding within an asset-backed security be subject effectively to the same “capital requirement.” The case of the loan is straightforward: The asset-backed security holding would have a minimum haircut depending on collateral quality and tranche seniority (see, for example, Geanakoplos, 2010, and Stein, 2012). Such a regulatory approach reduces implicit taxes on bank lending as opposed to shadow bank lending and diminishes the severity of the fire sales in shadow banking. The alternative to this more-expansive notion of bank capital regulation is to combat runs by extending the federal safety net, as, for example, in the financial crisis. But this policy extends the costly moral hazard built into the current banking safety net, requiring further thought about the pricing of such “insurance.” Back to goals and principles. The most basic arguments for government-mandated capital requirements (as opposed to levels determined by purely market focus) reflect ex ante moral hazard concerns about the regulatory safety net. As I observed above, this ex ante reassurance is not costless. And, more important for reform in the aftermath of the financial crisis, capital requirements cannot be the complete answer in response to the ex post runs by short-term creditors whose investments are exposed to potential fire-sale losses. Regulation of Non-Bank Financial Institutions Hedge funds. Although hedge funds have been around for some 60 years, it was not until the 1990s that these private pools of capital became major players in the global financial markets. Since 1990, assets in the hedge fund industry grew by a factor of more than 12. By the summer of 2008, there were about 10,000 hedge funds, with approximately $2 trillion under management. 188 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard The key to considering hedge fund regulation is a non-superficial understanding of the role hedge funds play in the global financial markets—an understanding not only of the risks they pose, but also of the risks they mitigate. The special features of hedge funds have enabled them to both take on risks otherwise borne by traditional financial institutions and bring greater efficiency to the capital markets. On that account, hedge funds have, in fact, contributed to the overall stability of the financial system. Because hedge funds often bet against the market by shorting financial instruments and using other contrarian strategies, they play a major role in reducing the emergence of financial bubbles that may culminate in market instability. Likewise, their active participation in the market for credit derivatives enables them to reduce the risks borne by institutions closer to the center of the financial system. Finally, the arbitrage strategies used by hedge funds and the sheer volume of their trading activity promote greater efficiency in the capital markets. Nonetheless, some hedge funds may pose a systemic risk to the financial system. A very large, unsustainably leveraged fund exposes a number of large financial institutions to increased counterparty risk. Any effective regulatory regime should aim to curb this systemic risk while enabling the hedge fund industry to continue to provide liquidity, absorb financial risks, and increase the efficiency of the capital markets. Private equity. Private equity (PE) firms are partnerships that acquire ownership stakes in cash-generative commercial businesses, such as retailers, industrial companies, computer firms, and health care concerns. Because PE firms do not normally borrow, extend credit, serve as derivatives counterparties, or perform other functions normally associated with depository institutions, they could hardly be considered part of the shadow banking system. To be sure, PE sponsors made greater use of debt to finance deals during the years 2005-07 than they had in previous years, a fact that has caused some analysts to express concern that defaults at PE-sponsored companies will increase in the coming years. Though some defaults have occurred in the difficult macroeconomic environment, these companies have some important advantages relative to their public competitors. Recent research found that during periods of acute financial stress, productivity growth at PE-sponsored companies was significantly higher than productivity growth at comparable non-PE businesses (see Bernstein et al., 2010). In addition, it is important to recognize that the failure of a portfolio company is not likely to have secondary effects on the larger financial system. Portfolio companies are broadly diversified across industries, and neither PE funds nor portfolio companies are cross-collateralized. These factors, taken as a whole, demonstrate that PE firms pose little systemic risk. Money market mutual funds. Since they began operations in the 1970s, money market mutual funds (MMMFs) have come to play an increasingly important role in the U.S. money markets. Offering a very low-risk, stable investment mechanism for retail investors and large sophisticated investors, MMMFs are also a source of short-term liquidity for the secondary markets. A distinguishing feature of MMMFs is their historically stable share price, usually $1.00 per share, which has facilitated their use as cash management devices as an alternative to banks. By law, MMMFs are limited to investing in high-quality, low-risk assets with very short maturities to limit the risks and thereby maintain the stable share price. Despite their low-risk profile, the financial crisis, as it escalated in the wake of the collapse of Lehman Brothers, created Federal Reserve Bank of St. Louis REVIEW May/June 2013 189 Hubbard tremendous instability for money market funds, drying up the flow of short-term liquidity they provided to the market. The Primary Reserve Fund was shown to have been exposed to increasingly risky Lehman Brothers commercial paper that, while giving it the competitive advantage gained from offering higher yields to its investors, nevertheless set the stage for large losses once Lehman Brothers fell. Significantly, as a result of the losses and the rush by investors to redeem their investments, the Primary Reserve Fund “broke the buck,” prompting further runs on other MMMFs. As a result of this increasing spread of systemic risk, the Department of the Treasury decided to guarantee the accounts of shareholders in MMMFs existing on the date the guarantee was issued. The crisis has highlighted the need for reform of the regulatory structure underpinning MMMFs. In particular, MMMFs should adopt better crisis management and more-robust mechanisms for risk monitoring, transparency, and analysis. MMMFs must ultimately be required to compensate taxpayers for the government’s guarantees of protection for money market accounts. Returning to the central problem of contagion I raised at the outset, concerns over liquidity remain important. The Securities and Exchange Commission is considering alternatives for restructuring redemptions, an approach opposed strongly by the industry. Back to goals and principles. From the perspective of systemic risk and contagion, additional regulatory intervention in non-bank financial institutions is difficult to justify. However, assessing liquidity in MMMFs requires further analysis. Resolution Process for Failed Financial Institutions Recent market events have revealed both the strengths and the weaknesses of current insolvency regimes for complex financial institutions. Certain insolvencies have had a far greater systemic effect than others, in part because the law that governs the insolvency of a financial company depends on the company’s form of organization. Specifically, the insolvency of banks insured by the Federal Deposit Insurance Corporation (FDIC) is governed by the Federal Deposit Insurance Act (FDIA); the insolvency of registered broker-dealers is governed by the Securities Investor Protection Act; and the insolvency of most other financial companies is governed by the U.S. Bankruptcy Code. The FDIA enables regulators to more effectively combat systemic risk. Notably, it creates a flexible insolvency regime that provides for pre-resolution action, receivership and conservatorship, and many methods of resolution, including liquidation, open bank assistance, purchase and assumption transactions, and the establishment of bridge banks. This regime has been very successful in promoting stability in the banking system by reducing uncertainty for depositors and counterparties while successfully mitigating losses for banks, counterparties, and the deposit insurance fund. However, the FDIA resolution regime is available only to banks, excluding from coverage many systemically significant financial companies, including bank holding companies. One significant aspect of the FDIA, compared with the Bankruptcy Code, is that it permits the transfer of certain derivatives and other qualified financial contracts to third parties, thus eliminating the downward spiral of prices that can result from a rush to liquidate collateral. A better alternative is to implement a comprehensive Financial Company Resolution Act, applicable to all financial institutions and based on the FDIA, that is applicable to all financial companies. 190 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard The path taken by the Dodd-Frank Act of 2010—the Orderly Liquidation Authority—still leaves many questions unanswered in this regard. Offering a general resolution antidote to too big to fail is particularly important as the legislation’s regulatory burdens give an advantage to very large banks relative to smaller banks. And estimates of the value of the funding advantage through an implicit guarantee for very large institutions are in the tens of billions of dollars per year (and, in some estimates, hundreds of billions per year). Similar national efforts to reform insolvency regimes for complex financial institutions are underway abroad. However, attention also should be paid to the resolution of cross-border financial companies, particularly banks. International working groups at the World Bank, the International Monetary Fund, the Bank for International Settlements, and elsewhere are currently considering various approaches. An effective international framework for resolving cross-border banks would reduce the pressure on national banking regulators to “ring-fence” the assets of a branch or subsidiary of a foreign bank in the event of its insolvency. This work is important and should be supported. Back to goals and principles. Resolution mechanisms described in this essay, ranging from contingent capital and creditor bail-ins to good-bank/bad-bank structures to Dodd-Frank’s Orderly Liquidation Authority, focus on individual institutions. There is still an open question about the extent to which to protect short-term creditors to reduce the likelihood of contagious runs. That is, while ex post resolution mechanisms are specifically designed to force losses on equity and debt holders of failing institutions, those mechanisms may do little to address contagion by exposing short-term creditors to greater risk. I return to this point later. MINDING THE STORE: REGULATORY STRUCTURE IN THE FUTURE An Alternative to the Dodd-Frank Structure Effective financial regulation requires a reorganization of the current U.S. regulatory structure. Any decision regarding that structure must be uniquely tailored to the needs of the United States. However, some other major industrial countries have moved toward more consolidated financial oversight. A rapidly dwindling share of the world’s financial markets is supervised under the fragmented, sectoral model still used by the United States. The January 2009 Report of the Committee on Capital Markets Regulation, which I co-chair, summarized appropriate relative responsibilities for the regulatory bodies in a system of consolidated oversight. The Dodd-Frank Act pursued a different path. The Federal Reserve would retain its exclusive control of monetary policy and its lender-of-last-resort function as part of its role in ensuring financial stability. One regulator should have the authority and accountability to regulate matters pertaining to systemic risk. The Committee on Capital Markets Regulation proposed a new U.S. Financial Services Authority (USFSA) that would regulate all aspects of the financial system, including market structure and safety and soundness for all financial institutions (and possibly consumer and investor protection with respect to financial products if this responsibility were lodged with the USFSA). Federal Reserve Bank of St. Louis REVIEW May/June 2013 191 Hubbard With this approach, the Department of the Treasury would coordinate the work of the regulatory bodies. The Treasury should also be responsible for the expenditure of public funds to provide support to the financial sector. In addition, any existing Fed loans to the private sector that are uncollateralized or insufficiently collateralized should be transferred in an orderly fashion to the balance sheet of the federal government (through asset purchases by the Treasury from the Federal Reserve). The United States should draw on the experiences of leading jurisdictions in devising a stepby-step regulatory consolidation process. Three options for supervising financial institutions merit serious consideration. The Fed could be placed in charge of supervising financial institutions determined to be “systemically important,” and the USFSA could supervise all other institutions. Alternatively, the Fed could be placed in charge of supervising all financial institutions. Finally, the USFSA could be placed in charge of supervising all financial institutions. A vigorous agency for consumer and investor protection could exist as a division within the USFSA. The Dodd-Frank Act of 2010 authorized the creation of a separate agency to be housed within the Federal Reserve. Most of the issues addressed above reach beyond the borders of any one country. Indeed, the international dimensions of the current financial crisis are so important that it is difficult to characterize this crisis as anything but global. In such an interconnected world, there is a particular need for an effective system of international financial oversight. Such a system should perform three distinct tasks. First, it should build the capacity to harmonize basic global rules. Second, it should serve as an early warning system capable of coordinating swift responses to brewing crises with systemic implications. And third, it should provide some sort of process for efficient dispute resolution when conflicts among regulatory regimes arise. Though it would be theoretically possible to harmonize financial regulation across borders through international treaties, regulators have instead turned to so-called regulatory networks to deal with the increasing globalization of finance. But these industry-specific networks failed to perform effectively during the recent crisis. Accordingly, the Obama administration and the G-20 have suggested entrusting international regulatory oversight to the Financial Stability Board. A strengthened Financial Stability Board is a good idea, so long as it is flexible and expert enough to harmonize basic rules for the regulation of international finance while still taking a broad view of all the markets in which modern financial conglomerates participate. Regardless of the multilateral networks and institutions in place, problems are bound to arise when countries pursue different approaches to financial regulation, as evidenced by the war of words in the aftermath of the crisis between the United Kingdom and Iceland over who should take responsibility for failed Icelandic banks doing business in the United Kingdom. Even if some harmonization were successful, issues would still arise when countries pursued different regulations for the same activity. When such conflicts occur, there must be some system for resolving them. In preparation for these expected disagreements, we believe governments ought to strengthen their “regulatory dialogues,” if only to maintain open lines of communication among their high-level officials. 192 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard Back to the Problem of Contagion in a Financial Crisis In some important ways, recent regulatory reforms have exacerbated problems of systemic risks through contagion. One way costly runs by short-term creditors can be reduced is by lender-of-last-resort support from the central bank. In the teeth of the financial crisis, the Federal Reserve expanded its provision of liquidity through new borrowing facilities. The DoddFrank Act, however, reduced the Fed’s lending authority and raised collateral requirements for emergency lending.14 These limits, particularly given the new power granted to the Secretary of the Treasury, are unlikely to reassure short-term creditors, an essential problem given the fragile maturity transformation inherent in banking. While the law sought to reduce potential taxpayer losses from intervention (in the crisis lending, the Fed actually profited), the trade-off in costs of contagion cannot be ignored. A second regulatory intervention that makes contagion-fighting more difficult is restrictions on public capital injections. Bailouts were unpopular with much of the public and many policymakers because of ex post taxpayer expense and ex ante moral hazard. That bailouts of automobile companies occurred as well only enhanced public and policymaker concern that such interventions were politically motivated. And there are legitimate fears that such interventions allow politicians to prop up insolvent institutions for an extended period of time. Without guarantee or insurance systems for short-term creditors, however, public capital injections are a potentially valuable tool for addressing contagion and systemic risk during a financial crisis. What, then, should the United States do in the event of another significant financial crisis? Even if the government lacks standing authority to inject capital, Congress should be prepared to act early during the crisis in the context of a comprehensive plan of action. And the government should increase accountability for equity holders and management. Analyzing ways to resolve competing goals of institutional accountability and preventing contagion should rise in importance in both economic research and policy design—both for commercial banking and shadow banking. AND THEN THERE WERE NONE In Agatha Christie’s novel, And Then There Were None, a murder in a grand mansion startles the remaining guests, who then calm down until another murder occurs. And another, and another…As I write this essay, in early 2013, both Washington and Wall Street have slipped into complacency. The case for financial regulatory reform I have outlined here remains strong, and the basic building blocks of regulatory reform appear in many thoughtful proposals for change around the world. The time for rethinking financial regulatory reform is now. But now is also the time for serious thinking and analysis. What is needed is a clear plan for reducing systemic risk, enhancing transparency, and modernizing regulatory institutions. Despite its 2,300 pages, the DoddFrank Act of 2010 offers as many questions as answers, guaranteeing—not ending—a vigorous debate over the proper scope of financial regulation in the years to come. Federal Reserve Bank of St. Louis REVIEW May/June 2013 193 Hubbard That debate requires us to confront systemic risk and contagion. It also requires us to evaluate trade-offs between mitigating failure of individual institutions versus reducing systemic risk. Finance and the financial system are valuable for savers and borrowers and for the provision of risk-sharing, liquidity, and information services. Proper regulation can preserve and enhance this value. NOTES 1 In another essay (Hubbard, 2011), I tackled policy questions related to credit default swaps, securitization, and regulatory structure. I also draw on that essay’s points in this lecture. 2 Much of the early policy discussion in the aftermath of the Lehman failure and the AIG rescue concerned interconnectedness of assets and/or liabilities. While such interconnectedness has not been conclusively demonstrated, the problem of contagion—particularly arising from reliance on short-term borrowings by both bank and non-bank institutions—was very important. 3 The Report by the Committee on Capital Markets Regulation enumerated 57 recommendations for reform. The broad principles I outline are similar to those proposed by the Group of 30 in Financial Reform (January 2009; http://fic. wharton.upenn.edu/fic/policy%20page/G30Report.pdf) and by the Group of 20 in Enhancing Sound Regulation and Strengthening Transparency (March 2009; http://rbidocs.rbi.org.in/rdocs/PublicationReport/Pdfs/20_010409.pdf). The appendix to the Report summarizes similarities and differences in leading proposals for financial regulatory reform. 4 Cost-benefit analysis also is recognized in the Congressional Oversight Panel’s Special Report on Regulatory Reform (January 2009; https://www.un.org/ga/president/63/commission/regulatoryreform.pdf) and in the Counterparty Risk Management Policy Group III’s report, Containing Systemic Risk: The Road to Reform (August 2008; http://www.crmpolicygroup.org/docs/CRMPG-III.pdf). 5 See, for example, chapters 10, 11, and 12 of Hubbard and O’Brien (2013). 6 In much of this research (see especially Diamond, 1984, and Boyd and Prescott, 1986), intermediaries offer low-cost means of monitoring some classes of borrowers. Because of informational frictions, nonmonitored finance entails deadweight spending resources on monitoring. A free-rider problem emerges, however, in public markets with a large number of creditors. The problem is mitigated by having a financial intermediary hold the loans and act as a delegated monitor. Potential agency problems at the intermediary level are reduced by having the intermediary hold a diversified loan portfolio financed principally by publicly issued debt. This line of research argues rigorously that borrowers for whom monitoring costs are significant will be dependent on intermediaries for external finance. 7 Costly external financing is an important element in contemporary macroeconomic models identifying financial factors in propagating relatively small shocks, factors that correspond to “accelerator” models that explain investment data relatively well. Three common empirical implications have merged from these models (see, for example, Townsend, 1979; Blinder and Stiglitz, 1983; Farmer, 1985; Williamson, 1987; Bernanke and Gertler, 1989, 1990; Calomiris and Hubbard, 1990; Hubbard and Kashyap, 1992; Kiyotaki and Moore, 1997; Greenwald and Stiglitz, 1993; Fazzari, Hubbard, and Petersen, 1988; and Bernanke, Gertler, and Gilchrist, 1996). The first is that uncollateralized external finance is more expensive than internal finance. Second, the spread between the cost of external and internal finance varies inversely with the borrower’s net worth relative to the amount of funds required. Third, an adverse shock to a borrower’s net worth increases the cost of external finance and decreases the ability of the borrower to implement investment and production plans. 8 For debt financing, see, for example, James (1987). Ediz, Michael, and Perraudin (1998) find that U.K. banks face lower costs of Tier 2 capital than equity capital. 9 A substantial body of empirical work has studied bank capital crunches to examine links between adverse shocks to bank equity capital and bank lending. See, for example, Bernanke (1983), Bernanke and Lown (1991), Kashyap and Stein (1995, 2000), Peek and Rosengren (1997, 2000), Calomiris and Mason (2003), and Hubbard, Kuttner, and Palia (2002). Other analysis has focused on links between higher bank equity capital requirements and bank lending contraction. See, for example, Furlong (1992), Peek and Rosengren (1995a,b), and Lown and Peristiani (1996). Using panel data on U.K. banks over the 1996-2007 period, Francis and Osborne (2009) estimate a long-run internal target for the ratio of capital to risky assets. They find that, all else equal, deficits of capital relative to this target have 194 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard lower growth in credit. Using their estimated effects, imposing three 1-percentage-point increases in capital requirements in 1997, 2001, and 2003 would have reduced the stock of lending by more than 5 percent by 2007. 10 A substantial body of empirical evidence supports the idea that banks offer special services in the lending process. For example, James (1987) and Lummer and McConnell (1989) find that the announcement of a bank loan, all else equal, raises the share price of the borrowing firm, likely reflecting the information content of the bank’s assessment. In a similar spirit, Fama (1985) and James (1987) find that banks’ borrowers, rather than banks’ depositors, bear the incidence of reserve requirements (indicating that borrowers must not have easy access to other sources of funds). Petersen and Rajan (1994) show that small businesses tend to rely on local banks for external funds. 11 This recommendation is broadly similar to reforms proposed by the Congressional Oversight Panel in its Special Report on Regulatory Reform (January 2009; see endnote 4), the United Kingdom’s Financial Services Authority in The Turner Review: A Regulatory Response to the Global Banking Crisis (March 2009; http://www.fsa.gov.uk/pubs/other/ turner_review.pdf), the Group of Thirty in Financial Reform (January 2009; see endnote 3), and the Financial Stability Forum in its Report on Enhancing Market and Institutional Resilience (October 2008; http://www.financialstabilityboard.org/publications/r_0810.pdf). The recommendation that large institutions be held to a higher solvency standard is broadly similar to reforms proposed by the Council on Foreign Relations in the Special Report on Reforming Capital Requirements for Financial Institutions (April 2009; http://www.cfr.org/economics/reforming-capital-requirements-financial-institutions/p19001) and by the Congressional Oversight Panel in Regulatory Reform (January 2009). 12 The Dodd-Frank Act sets forth a $50 billion threshold for subjecting bank holding companies to enhanced prudential standards. In the spirit of treating systemically important banks and non-banks similarly, the decision by the Financial Stability Oversight Council (FSOC) to use a bright-line asset threshold for non-bank systemically important financial institution (SIFI) determinations is also a good one. However, for non-banks that meet the Stage 1 $50 billion threshold, the FSOC should reconsider excluding the categories of institutions I mentioned earlier from SIFI designation. 13 As noted earlier, asset contraction shrinks financing for investment and working capital, leading to a decline in out- put. Asset contraction across banks can lead to fire sales of assets as banks scramble to raise capital. Such fire sales, as Stein (2012) observes, create an externality: A given bank does not take into account the effect of its own leverage decisions on collateral values of other banks’ assets and their ability to raise financing. 14 In addition, the Department of the Treasury cannot guarantee MMMF investors against runs, nor can the FDIC insure senior bank debt without specific congressional approval. REFERENCES Admati, Anat R.; DeMarzo, Peter M.; Hellwig, Martin F. and Pfleiderer, Paul. “Fallacies, Irrelevant Facts, and Myths in the Discussion of Capital Regulation: Why Bank Equity Is Not Expensive.” Working paper, Rock Center for Corporate Governance, Stanford University, 2010; updated March 23, 2011; https://gsbapps.stanford.edu/researchpapers/library/RP2065R1&86.pdf. Aiyar, Shekhar; Calomiris, Charles W. and Wieladek, Tomas Z. “Does Macro-Pru Leak? Evidence from a U.K. Policy Experiment.” NBER Working Paper No. 17822, National Bureau of Economic Research, February 2012; http://www.nber.org/papers/w17822.pdf.?new_window=1. Bernanke, Ben S. “Non-Monetary Effects of the Financial Crisis in the Propagation of the Great Depression.” American Economic Review, June 1983, 73(3), pp. 257-76; http://fraser.stlouisfed.org/docs/meltzer/bernon83.pdf. Bernanke, Ben S. and Gertler, Mark L. “Agency Cost, Net Worth, and Business Fluctuations,” American Economic Review, March 1989, 79(1), pp. 14-31. Bernanke, Ben S. and Gertler, Mark L. “Financial Fragility and Economic Performance.” Quarterly Journal of Economics, February 1990, 105(1), pp. 87-114. Bernanke, Ben S.; Gertler, Mark L. and Gilchrist, Simon. “The Financial Accelerator and the Flight to Quality.” Review of Economics and Statistics, February 1996, 78(1), pp. 1-14. Bernanke, Ben S. and Lown, Cara S. “The Credit Crunch.” Brookings Papers on Economic Activity, 1991, 2, pp. 205-39. Bernstein, Shai; Lerner, Josh; Sorensen, Morten and Strömberg, Per. “Private Equity and Industry Performance.” Harvard Business School Entrepreneurial Management Working Paper No. 10-045, March 2010. Federal Reserve Bank of St. Louis REVIEW May/June 2013 195 Hubbard Blinder, Alan S. and Stiglitz, Joseph E. “Money, Credit Constraints, and Economic Activity.” American Economic Review, May 1983, 73(2), pp. 297-302. Boyd, John H. and Prescott, Edward. “Financial Intermediary-Coalitions.” Journal of Economic Theory, April 1986, 38(2), pp. 211-32. Calomiris, Charles W. and Herring, Richard J. “Why and How to Design a Contingent Convertible Debt Requirement.” Working paper, Columbia Business School, April 2011, revised November 2011; http://www1.gsb.columbia.edu/mygsb/faculty/research/pubfiles/5631/contingent_convertible_debt.pdf. Calomiris, Charles W. and Hubbard, R. Glenn. “Firm Heterogeneity, Internal Finance, and ‘Credit Rationing.’” Economic Journal, March 1990, 100(399), pp. 90-104. Calomiris, Charles W. and Kahn, Charles M. “The Role of Demandable Debt in Structuring Optimal Banking Arrangements.” American Economic Review, June 1991, 81(3), pp. 497-513. Calomiris, Charles W. and Mason, Joseph R. “Consequences of Bank Distress During the Great Depression.” American Economic Review, June 2003, 93(3) pp. 937-47. Diamond, Douglas W. “Financial Intermediation and Delegated Monitoring.” Review of Economic Studies, July 1984, 51(3), pp. 393-414. Diamond, Douglas W. and Rajan, Raghuram G. “A Theory of Bank Capital.” Journal of Finance, December 2000, 55(6), pp. 2431-65. Ediz, Tolga; Michael, Ian and Perraudin, William. “The Impact of Capital Requirements on U.K. Bank Behaviour.” Federal Reserve Bank of New York Economic Policy Review, October 1998, 4(3), pp. 15-22; http://www.ny.frb.org/research/epr/98v04n3/9810ediz.pdf. Fama, Eugene F. “What’s Different About Banks?” Journal of Monetary Economics, January 1985, 15(1), pp. 29-39. Farmer, Roger E.A. “Implicit Contracts with Asymmetric Information and Bankruptcy: The Effect of Interest Rates on Layoffs.” Review of Economic Studies, July 1985, 52(3), pp. 427-42. Fazzari, Steven M.; Hubbard, R. Glenn and Petersen, Bruce C. “Financing Constraints and Corporate Investment.” Brookings Papers on Economic Activity, 1988, 1, pp. 141-95. Flannery, Mark J. “No Pain, No Gain? Effecting Market Discipline via ‘Reverse Convertible Debentures,’” in Hal S. Scott, ed., Capital Adequacy Beyond Basel: Banking, Securities, and Insurance. New York: Oxford University Press, 2005, pp. 171-96. Francis, William B. and Osborne, Matthew. “Bank Regulation, Capital and Credit Supply: Measuring the Impact of Prudential Standards.” Occasional Paper Series No. 36, U.K. Financial Services Authority, September 2009; http://www.fsa.gov.uk/pubs/occpapers/op36.pdf. Francis, William B. and Osborne, Matthew. “On the Behavior and Determinants of Risk-Based Capital Ratios: Revisiting the Evidence from U.K. Banking Institutions.” International Review of Finance, December 2010, 10(4), pp. 485-518. Furlong, Frederick T. “Capital Regulation and Bank Lending.” Federal Reserve Bank of San Francisco Economic Review, 1992, No. 3, pp. 23-33; http://www.frbsf.org/publications/economics/review/1992/92-3_23-33.pdf. Gale, Douglas and Hellwig, Martin. “Incentive-Compatible Debt Contracts: The One-Period Problem.” Review of Economic Studies, October 1985, 52(4), pp. 647-63. Geanakoplos, John. “The Leverage Cycle.” Cowles Foundation Discussion Paper No. 1715R, Cowles Foundation for Research in Economics, Yale University, July 2009, revised 2010; http://cowles.econ.yale.edu/P/cd/d17a/d1715-r.pdf. Gorton, Gary B. and Metrick, Andrew. “Securitized Banking and the Run on the Repo.” Journal of Financial Economics, June 2012, 104(3), pp. 425-51; http://www.sciencedirect.com/science/article/pii/S0304405X1100081X. Greenwald, Bruce C. and Stiglitz, Joseph E. “Financial Market Imperfections and Business Cycles.” Quarterly Journal of Economics, February 1993, 108(1), pp. 77-114. Hanson, Samuel; Kashyap, Anil K. and Stein, Jeremy C. “A Macroprudential Approach to Financial Regulation.” Journal of Economic Perspectives, Winter 2011, 25(1), pp. 3-28. Hubbard, R. Glenn. “Is There a ‘Credit Channel’ for Monetary Policy?” Federal Reserve Bank of St. Louis Review, May/ June 1995, 77(3), pp. 63-77; http://research.stlouisfed.org/publications/review/95/05/Credit_May_June1995.pdf. 196 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Hubbard Hubbard, R. Glenn. “The Morning After: A Road Map for Financial Regulatory Reform,” in Roger B. Porter, Robert R. Glauber, and Thomas J. Healey, eds., New Directions in Financial Services Regulation. Cambridge, MA: MIT Press, 2011, pp. 77-98. Hubbard, R. Glenn and Kashyap, Anil K. “Internal Net Worth and the Investment Process: An Application to U.S. Agriculture.” Journal of Political Economy, June 1992, 100(3), pp. 506-34. Hubbard, R. Glenn; Kuttner, Kenneth N. and Palia, Darius N. “Are There Bank Effects in Borrowers’ Costs of Funds? Evidence from a Matched Sample of Borrowers and Banks.” Journal of Business, October 2002, 75(5), pp. 559-81. Hubbard, R. Glenn and O’Brien, Anthony P. Money, Banking, and the Financial System. Second Edition. Upper Saddle River, NJ: Prentice Hall, 2013. Hubbard, R. Glenn; O’Brien, Anthony P. and Rafferty, Matthew. Macroeconomics. Second Edition. Upper Saddle River, NJ: Prentice-Hall, 2013. James, Christopher. “Some Evidence on the Uniqueness of Bank Loans.” Journal of Financial Economics, December 1987, 19(2), pp. 217-35. Kashyap, Anil K.; Rajan, Raghuram G. and. Stein, Jeremy C. “Rethinking Capital Regulation,” in Maintaining Stability in a Changing Financial System. Proceedings of the 2008 Jackson Hole Economic Policy Symposium, Jackson Hole, Wyoming, August 21-23, 2008. Kansas City, MO: Federal Reserve Bank of Kansas City, 2008, pp. 431-71; http://www.kc.frb.org/publicat/sympos/2008/KashyapRajanStein.08.08.08.pdf. Kashyap, Anil K. and Stein, Jeremy C. “The Impact of Monetary Policy on Bank Balance Sheets.” Carnegie-Rochester Conference Series on Public Policy, June 1997, 42(1), pp. 151-95. Kashyap, Anil K. and Stein, Jeremy C. “What Do a Million Observations on Banks Say about the Transmission Mechanism of Monetary Policy?” American Economic Review, June 2000, 90(3), pp. 407-28. Kiyotaki, Nobuhiro and Moore, John. “Credit Cycles.” Journal of Political Economy, April 1997, 105(2), pp. 211-48. Lown, Cara, and Peristiani, Stavros. “The Behavior of Consumer Loan Rates During the 1990 Credit Slowdown.” Journal of Banking and Finance, December 1996, 20(10), pp. 1673-94. Lummer, Scott and McConnell, John. “Further Evidence on the Bank Lending Process and Capital Market Response to Bank Loan Agreements.” Journal of Financial Economics, November 1989, 25(1), pp. 99-122. Modigliani, Franco and Miller, Merton H. “The Cost of Capital, Corporation Finance and the Theory of Investment.” American Economic Review, June 1958, 48(3), pp. 261-97. Myers, Stewart C. and Majluf, Nicholas S. “Corporate Financing and Investment Decisions When Firms Have Information That Investors Do Not Have.” Journal of Financial Economics, June 1984, 13(2), pp. 187-222. Peek, Joe and Rosengren, Eric S. “Bank Regulation and the Credit Crunch.” Journal of Banking and Finance, June 1995a, 19(3-4), pp. 679-92. Peek, Joe and Rosengren, Eric S. “The Capital Crunch: Neither a Borrower Nor a Lender Be.” Journal of Money, Credit, and Banking, August 1995b, 27(3), pp. 625-38. Peek, Joe and Rosengren, Eric S. “The International Transmission of Financial Shocks: The Case of Japan.” American Economic Review, September 1997, 87(4), pp. 495-505. Peek, Joe and Rosengren, Eric S. “Collateral Damage: Effects of the Japanese Bank Crisis on Real Activity in the United States.” American Economic Review, March 2000, 90(1), pp. 30-45. Petersen, Mitchell A. and Rajan, Raghuram G. “The Benefits of Lending Relationships: Evidence from Small Business Data.” Journal of Finance, March 1994, 49(1), pp. 3-37. Stein, Jeremy. “Monetary Policy as Financial-Stability Regulation.” Quarterly Journal of Economics, February 2012, 127(1), pp. 57-95. Townsend, Robert. “Optimal Contracts and Competitive Markets with Costly State Verification.” Journal of Economic Theory, October 1979, 21(2), pp. 265-93. Williamson, Stephen D. “Costly Monitoring, Optimal Contracts, and Equilibrium Credit Rationing.” Quarterly Journal of Economics, February 1987, 102(1), pp. 135-46. Federal Reserve Bank of St. Louis REVIEW May/June 2013 197 Hubbard 198 May/June 2013 Federal Reserve Bank of St. Louis REVIEW 199 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Big Banks in Small Places: Are Community Banks Being Driven Out of Rural Markets? R. Alton Gilbert and David C. Wheelock The shares of total U.S. banking assets and deposits held by the very largest banking organizations have increased markedly over the past 25 years, while the shares held by small “community” banks have declined. Advances in information technology may have reduced the advantages of small scale, close proximity, and local ties that traditionally have given small, community-focused banks a competitive advantage in lending to small businesses and other “informationally opaque” borrowers. This article examines trends in deposit shares of banks of different sizes in rural U.S. counties. If the community banking model is to remain viable, it is likely to be in rural markets with (i) a relatively high percentage of informationally opaque borrowers and (ii) relatively low costs of acquiring qualitative information about potential borrowers. The authors find that rural deposit shares of both the smallest and very largest banking organizations changed little between 2001 and 2012, despite the upheavals of the financial crisis and recession, and in contrast to the 1980s and 1990s, when the shares held by the smallest banks declined markedly. The evidence suggests that well-managed community banks remain competitive, at least in rural markets, where their niche is most likely stronger than in urban markets. (JEL G21, G28, L1) Federal Reserve Bank of St. Louis Review, May/June 2013, 95(3), pp. 199-218. he U.S. banking system has consolidated in recent years, and the shares of total banking system assets and deposits held by the very largest banking organizations have increased markedly. The number of commercial banks reached its post-World War II peak of 14,483 banks in 1984. By year-end 2012, the number of commercial banks had declined to 6,096. In 2001, the five largest commercial banks held 30 percent of total U.S. banking system assets; of these, Bank of America was the top asset holder with $552 billion in assets. Ten years later, in 2011, the five largest banks held 48 percent of total system assets and four banks had total assets in excess of $1 trillion. The largest commercial bank—JPMorgan Chase Bank—had $1.8 trillion in assets in 2011, equal to 14 percent of the total assets of all U.S. commercial banks. T R. Alton Gilbert is an economist emeritus and David C. Wheelock is a vice president and deputy director of research at the Federal Reserve Bank of St. Louis. The authors thank Dean Amel, Fernando Martin, Andrew Meyer, and Kenneth Spong for comments on a previous version of this article. David L. Lopez provided research assistance. © 2013, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW May/June 2013 199 Gilbert and Wheelock The relaxation of state and federal government restrictions on branching removed an important impediment to consolidation of the banking industry. In 1984, more than half of all U.S. banks were “unit banks”—that is, they had only one office; in 2011, only one-tenth of all banks were unit banks.1 The Riegle-Neal Interstate Banking and Branching Efficiency Act of 1994 allowed banks to branch across state lines, and several banks now operate branches in multiple states, including JPMorgan Chase with 5,488 domestic branches and Bank of America with 5,755 domestic branches as of December 31, 2011. As very large banks have seen their shares of total U.S. banking assets and deposits rise, small banks have seen their shares dwindle. Between 2001 and 2012, for example, the shares of total U.S. commercial bank assets and deposits held by banks with less than $1 billion in assets declined from 16 percent and 20 percent, respectively, to 9 percent and 10 percent.2 The continuing decline in shares held nationally by small banks has raised the question whether small, community-focused banks can remain economically viable. Analysts point to both public policies and technological changes as favoring larger banks over their smaller competitors. For example, the treatment of very large banks as too big to fail could allow large banks to attract depositors and issue other forms of debt less expensively than smaller banks. The Dodd-Frank Wall Street Reform and Consumer Protection Act of 2010 is intended to eliminate such too-big-to-fail protections, but some analysts are skeptical. If the public perceives that the government will protect depositors and other creditors of very large banks from loss, but not afford similar protection to creditors of smaller institutions, then smaller banks will be competitively disadvantaged.3 Technological changes, particularly advances in information-processing technology, may also favor larger banks. Such advances have lowered the costs of obtaining “hard” information about potential borrowers, such as audited financial statements and standardized credit reports. At the same time, these changes have also lowered the cost to banks of monitoring deposit and loan accounts and managing large branch networks. Thus, technological changes may have reduced the advantages of small scale, close proximity, and local ties that traditionally have given small, community-focused banks a competitive advantage in lending to small businesses and other borrowers for which hard data do not exist or are limited (Petersen and Rajan, 2002; Berger, 2003; Bernanke, 2006). Despite public policies and technological advances that appear to favor large banks with far-flung operations, many community banks continue to thrive, even in challenging times such as the recent financial crisis and recession (Gilbert, Meyer, and Fuchs, 2013). Further, some studies conclude that the community bank business model, with its focus on lending to small businesses and consumers on the basis of “soft” information derived from customer relationships, community ties, and so forth, remains viable for well-managed banks (see, e.g., DeYoung, Hunter, and Udell, 2004). The jury is still out, however, about whether the evolving structure of the U.S. banking system will retain a place for community banks over the long run. The Federal Deposit Insurance Corporation (FDIC, 2012) recently released a comprehensive study of community banks. The study compares the performance of community and noncommunity banks, community bank lending patterns, capital strategies, locations, and trends in market share in different-sized banking markets. The study reports several important findings. For example, it shows that the share of U.S. banking assets held by community banks declined 200 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock by more than 50 percent between 1984 and 2011. The study also finds that, in general, community banks have been less profitable than other banks. However, community banks remain the only banking presence in more than 600 counties (nearly 20 percent of all U.S. counties) and they continue to hold the majority of banking deposits in rural counties and small cities. The study finds that community bank market shares declined between 1984 and the early 1990s in both rural counties and small cities, as well as in urban markets, but their share of rural county markets changed little between 2000 and 2011. The FDIC study does not draw strong conclusions about the long-term viability of the community bank business model. However, the study notes that economic and population growth were slower during the past decade in rural counties and small cities than in urban markets; and, if those trends continue, the growth opportunities for community banks may be limited. This article examines trends in the deposit market shares of small, mid-sized, and very large banking organizations in rural U.S. banking markets between 2001 and 2012. Although community banks operate in both rural and urban markets, we focus on rural markets because the performance of community banks in rural markets is likely to be more informative about the viability of the community bank business model, with its focus on relationship lending and reliance on core deposits. The financial reports that banks file with regulators provide little information about bank customers. Data on small-denomination loans to businesses and farmers are available, but the borrowers of these small-denomination loans are not necessarily smallbusiness enterprises. Similarly, banks report information on deposits in accounts below the federal deposit insurance limit, but these data do not reveal the income or wealth of the depositors. We assume that bank offices in rural areas are more likely to conduct banking business with households and small businesses than banks in urban areas and therefore are more likely to engage in traditional banking activities, such as making loans on the basis of customer relationships and other soft information in which community banks are thought to specialize.4 Thus, if we were to find evidence that very large banks with far-flung operations are beginning to dominate rural markets, it would suggest that the community bank model may not be viable. That conclusion, in turn, would have implications for small businesses and other customers that traditionally have relied on community banks, as well as for the banks themselves. LITERATURE REVIEW Possible Advantages of Community Banks Lending to Small Businesses Many studies have addressed questions about the viability of the community bank business model. Some researchers argue that community banks have a comparative advantage in lending to “informationally opaque” borrowers, such as entrepreneurs, small businesses, and other borrowers for which audited accounting information, credit reports, and other hard information are limited. One reason community banks may have an advantage in lending to informationally opaque borrowers is that managers of community banks typically have more control over the allocation of their banks’ resources than do branch managers of large banks with far-flung operations. Federal Reserve Bank of St. Louis REVIEW May/June 2013 201 Gilbert and Wheelock Stein (2002) argues that this control gives managers of community banks more incentive to develop relationships with potential customers and to take other steps to acquire qualitative information about borrowers, which may lead to more-informed credit decisions about informationally opaque borrowers.5 Similarly, Brickley, Linck, and Smith (2003) argue that granting significant decisionmaking authority to local bank managers facilitates lending on the basis of qualitative information. However, to control agency problems, managers with a high degree of authority must have an ownership stake or be closely monitored by local stockholders, which generally is not possible in large banks with widely dispersed offices. Hence, Brickley, Linck, and Smith (2003) argue that small, locally owned banks are likely to dominate in small urban and rural markets where small businesses and other informationally opaque borrowers are the predominant loan customers. Empirical evidence has tended to support economic theory in finding that community banks lend disproportionately to small businesses and other informationally opaque borrowers. By contrast, large banks lend mainly to larger firms with good accounting information and households on the basis of credit ratings and other reliable quantitative information (Berger et al., 2005). Further, research finds that community banks (i) rely more heavily on developing customer relationships and investing in other forms of qualitative information and (ii) account for larger shares of banking business in smaller urban and rural markets than in large urban markets. Cole, Goldberg, and White (2004), for example, examine lending to small businesses, as reported in a 1993 Federal Reserve survey. The authors find that large banks are more likely to rely primarily on standard accounting records and other quantitative information in making credit decisions, whereas small banks use a more discretionary approach based heavily on preexisting customer relationships. Using the same survey, Berger et al. (2005) find that compared with large banks, small banks are more likely to (i) communicate with their small-business customers through personal means such as face-to-face office visits, (ii) develop more long-term relationships with their customers, and (iii) lend to local businesses. Petersen and Rajan (1994) examine the impact of customer relationships on the cost and availability of credit for small businesses. The study finds that firms pay higher interest rates when they have more than one lender, suggesting that a relationship with a single bank lowers a firm’s borrowing costs. Further, the study finds that the length and extent of a firm’s relationship with its bank lender have a significant effect on the supply of credit available to the firm. A firm will have a greater supply of credit from a bank the longer the firm’s relationship with the bank, the more distinct services it receives from the bank, and the more concentrated its borrowing with the bank. Community banks may have more information about the value of real estate in their communities than lenders with no local ties. Cortés (2012) presents evidence that residential mortgage lenders with a physical branch near the property being financed have better information about home-price fundamentals than nonlocal lenders. Home prices increased less from 2002 to 2006 in areas where lenders with local offices accounted for a higher percentage of residential mortgage loans. Moreover, home prices fell less from 2006 to 2009 in areas where more of the loans were made by local lenders. 202 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Effects of Credit Scoring on the Advantages of Community Banks in Lending to Small Businesses Advances in information-processing technology have increased the amount of quantitative and other hard information about potential borrowers, suggesting the importance of developing close relationships with potential borrowers may have decreased over time. Survey evidence finds that community banks are increasingly using credit scoring and other quantitative information to evaluate small-business loans (Berger, Cowan, and Frame, 2011). Petersen and Rajan (2002) investigate how the physical distance between small businesses and their lenders relates to credit quality and loan performance. Historically, only the highest-quality small-business borrowers could obtain loans from distant banks, and Petersen and Rajan (2002) find that informationally opaque firms are located closer to their lenders. However, the study also finds that the distance between small businesses and their lenders has tended to increase over time and that lender-borrower interactions have become less personal. Further, the study finds that the relationship between borrower-lender distance and credit quality has diminished over time: This observation is consistent with the increased availability of hard information on the creditworthiness of potential small-business borrowers brought about by changes in information technology.6 Further research has found somewhat varied effects of increased use of credit-scoring technology on small-business lending. DeYoung, Glennon, and Nigro (2008) use a large dataset consisting of loans partially guaranteed by the Small Business Administration to investigate the relationships among the use of credit scoring, borrower-lender distance, and loan performance. The study finds that loan default rates increase with borrower-lender distance when credit scoring is not used, but that distance has no impact on default when credit scoring is used. However, DeYoung, Glennon, and Nigro (2008) find that banks that use credit scoring tend to experience higher loan default rates. The study speculates that banks that use credit-scoring models are willing to accept higher default rates because of ancillary benefits associated with higher loan volume, such as scale economies and higher fee income. Alternatively, credit-scoring underwriting processes may simply rely on less information about borrowers than underwriting based on customer relationships and thus lead to more loan approval errors. In any event, the results are similar to those of Berger et al. (2005), who find that banks that use credit-scoring methods to evaluate small-business loans experience higher nonperforming loan rates. Increased use of credit scoring and other forms of information technology to obtain hard information about borrowers has likely somewhat diminished the comparative lending advantage of small, community-focused banks, particularly lending to small businesses. However, qualitative information obtained through personal contacts and customer relationships may remain important, especially for lending to borrowers lacking established credit histories, audited financial reports, and so forth. Performance of Community Banks in Urban and Rural Areas A few studies have examined differences in the types of markets where large and small banks locate offices and performance differences of loans to small businesses in different market Federal Reserve Bank of St. Louis REVIEW May/June 2013 203 Gilbert and Wheelock types. These studies find a greater tendency of community banks to locate in smaller urban and rural markets than in large urban markets. Studies also find that loans made by small banks in rural markets tend to outperform those made by small banks in urban markets. Brickley, Linck, and Smith (2003) examine the location of banks and bank branches across urban and rural markets in California and Texas in 1998. Whereas California had allowed statewide branching since the 1920s, Texas prohibited branching before 1986. The different regulatory histories suggest that the ownership patterns observed in one state might differ from those of the other state. However, Brickley, Linck, and Smith (2003) find that in both California and Texas, the probability that a given bank office is owned by a bank with total assets exceeding $1 billion increases with the population of the office’s market. Further, controlling for population, the probability of large bank ownership is higher in large cities but smaller in suburban areas, smaller metropolitan areas, and rural markets. DeYoung et al. (2012) compare the performance of small-business loans made by small banks in rural and urban markets. Using data on over 18,000 small-business loans made by community banks between 1984 and 2001, the study finds that loans made by rural banks have a significantly lower default rate than those made by small urban banks, with the lowest default rates found in the smallest rural markets. The authors attribute the performance advantage of community bank loans in rural markets to the “ruralness” of the borrower-lender relationship— that is, where information, institutions, and culture are most conducive to relationship or “character” lending. The performance advantage diminishes for loans made by a bank in one rural market to a borrower in a different rural market, as well as with increasing distance between borrower and lender. Effects of Large Banking Organizations on the Performance of Rural Community Banks Studies of small bank performance in rural markets generally find that, rather than large banks posing a threat to rural community banks, community banks in rural counties may actually benefit from the presence of large multimarket banks. Pilloff (1999), for example, finds that the presence of one or more large multimarket banks boosts return on assets (ROA) of small banks that operate only within a single rural market. Using a larger dataset covering the years 1996 through 2004, Hannan and Prager (2009) also find that the profitability of single-market banks in rural locations is directly affected by the presence of large, as well as small, multimarket banks. However, the direction of the impact on the ROA of a single-market bank located in a rural area appears to depend on the concentration of the market and the size of the single-market bank. The study finds that the presence of a large multimarket bank lowers the ROA of very small single-market rural banks, as well as those operating in highly concentrated markets. However, the presence of large (and small) multimarket banks boosts the ROA of a single-market bank of average asset size operating in a rural county with average market concentration. Finally, the study finds no strong relationship between the profit rates of small single-market banks located in urban areas and the presence of large multimarket banks. 204 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Like Pilloff (1999) and Hannan and Prager (2009), Cyree and Spurlin (2012) find that the presence of a large multimarket bank raises the ROA of small single-market banks in rural counties. However, when more than one large bank is present, the effect on small bank profit rates is reduced. Cyree and Spurlin (2012) also find that the presence of a single large bank lowers the profit efficiency of small banks in the market but the impact is lower when more than one large bank competes in the market. Taken together, the results for ROA and profit efficiency suggest that the presence of a single large multimarket bank generally reduces competition in a market, enabling all banks in the market to earn higher returns while operating less efficiently, whereas the presence of more than one large bank increases market competition and reduces profit rates for all banks in the market. Pilloff (1999), Hannan and Prager (2009), and Cyree and Spurlin (2012) argue that their evidence indicates that the presence of large banking organization offices tends to reduce competition in rural banking markets. An alternative explanation, which Cyree and Spurlin (2012) attempt to control for in their study, is that large banking organizations tend to locate offices in rural banking markets where local banks can earn relatively high profits. Using either explanation, however, the evidence is consistent with the conclusion that large banking organizations do not drive out local community banks when they open offices in rural market areas. BANK DEPOSIT SHARES IN RURAL MARKETS The United States has 3,142 counties, 2,043 of which are rural (i.e., located outside metropolitan statistical areas [MSAs]). Among the rural counties, 2,017 had at least one office of a commercial bank or savings institution in 2012.7 Banks of all sizes have offices in rural counties; but larger banks tend to have more of their offices and deposits in urban markets, whereas smaller banks have relatively more offices in rural counties. As of June 2012, 36 percent of the offices and 23 percent of the deposits of banking organizations with less than $1 billion in assets were in rural counties.8 By contrast, just 10 percent of the offices and 4 percent of the deposits of organizations with more than $50 billion in assets were in rural counties. The main focus of this article is on trends in the deposit shares of different-sized banking organizations in rural U.S. counties between 2001 and 2012. Community banks are often defined on the basis of their size, and studies commonly define community banks as banking organizations with total assets of either less than $1 billion or less than $10 billion.9 Here, we examine trends in the deposit shares of banking organizations in four size groups: (i) organizations with less than $1 billion in assets (identified in the figures as “Small”), (ii) organizations with $1 billion to 10 billion in assets (“Small-Mid”), (iii) organizations with $10 billion to $50 billion in assets (“Mid-Large”), and (iv) organizations with $50 billion or more in total assets (“Large”). Our study includes all bank holding companies and independent commercial banks and savings institutions except credit card banks, bankers’ banks, and other special purpose banks. We aggregate data for banks owned by holding companies to the holding company level and treat all offices of banks owned by a parent organization as belonging to that one organization.10 Further, except as noted, we adjust data on the total assets of each organization for inflation using the consumer price index (2006 = 100). Federal Reserve Bank of St. Louis REVIEW May/June 2013 205 Gilbert and Wheelock Our study of market shares among banks with offices located in rural counties is limited to data on total deposits, which are available at the branch office level from the annual Summary of Deposits reports filed by banking organizations.11 Data on lending and other services provided by banks are not available by bank branches. Some large financial institutions are successful in marketing financial services to residents of rural areas without operating physical offices in rural areas. Given the lack of comprehensive data on such activities, however, we focus exclusively on the shares of deposits held by banks of different sizes in rural areas. National Market Share Trends Previous research on trends in rural deposit shares of different sizes and types of banks found that community banks lost market share during the 1980s and 1990s, whereas large banking organizations saw an increase in average market shares (Gilbert, 2000, and FDIC, 2012). Our study encompasses the period 2001-12 and thereby includes the financial crisis and recession of 2007-09, as well as the surrounding years. Thus, we examine how trends in the market shares of banks of different sizes may have been affected by the crisis and recession, as well as whether the trends identified in other studies continued after 2001.12 We also investigate differences in market share trends in small rural counties (i.e., those with a population less than 10,000 according to the 2010 Census) compared with larger rural counties and counties located in urban areas. Further, we compare trends in different regions of the country. Previous research (Gilbert, 2000) found considerable regional differences in the presence and market shares of large banks in rural counties during the 1990s, which seemed to reflect the effects of state restrictions on branch banking and population density. Specifically, large banks had greater presence and market shares in (i) rural counties in states that permitted statewide branching in 1980 and (ii) more densely populated rural counties. Large banks were less often present or had smaller market shares in small rural counties in states that limited or prohibited branching in 1980. Many states relaxed their restrictions on branching during the 1980s, and the United States has permitted interstate branch banking since 1997. Hence, regional differences in the presence and market shares of large banks in rural areas might have been expected to dissipate over time if state branching restrictions were the principal reason for these differences before deregulation and within a few years thereafter. Figure 1 reports average deposit shares for banking organizations of different sizes in U.S. rural counties. The shares, especially of both the smallest and the very largest organizations, were remarkably stable over the 12-year period despite high numbers of mergers, a major financial crisis and recession, and a large number of bank failures. In 2001, banks with less than $1 billion in assets held on average 65 percent of rural county deposits, whereas in 2012, they held 64 percent. By contrast, very large banking organizations—those with assets in excess of $50 billion— held on average 15 percent of rural county deposits in 2001 and 14 percent in 2012. Over the same years, the average share of rural deposits held by organizations with $1 billion to $10 billion in assets rose from 13 percent to 17 percent, whereas the average share of those banks with $10 billion to $50 billion in assets fell from 7 percent to 4 percent. Next we consider differences between small rural counties—those with fewer than 10,000 inhabitants (in 2010)—and larger rural counties. Figures 2A and 2B present the average deposit 206 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Figure 1 Average Deposit Shares in All Rural Counties 0.9 Bank Size Small Small-Mid Mid-Large Large 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 shares of each size of banking organization in small and large rural counties. Not surprisingly, the smallest organizations had larger average deposit shares in small rural counties than in larger rural counties, whereas the very largest banks had smaller shares in smaller rural counties than in larger rural counties. The trends over time in average deposit shares, however, were similar for small and larger rural counties. In both small and large rural counties, the smallest banks and the very largest banks both experienced slight declines in average deposit shares of 1 percentage point or less. By contrast, banks with assets of $1 billion to $10 billion saw increases in average share in both small and large rural counties, whereas banks with assets of $10 billion to $50 billion had small declines. Thus, across all rural U.S. counties, the average deposit shares of both the smallest and very largest banks changed little between 2001 and 2012, with little difference in trends between small and large rural counties. Regional Market Share Trends In 1980, rural market deposit shares of large banking organizations differed substantially across geographic regions, which Gilbert (2000) linked to regional differences in state branch banking laws. Large banks had larger market shares in rural counties of states that had long permitted statewide branching and smaller shares in states that limited branching. Small banking markets invariably had only small banks where branching was prohibited or restricted. However, in states that permitted statewide branching, such as California, many small communities and rural counties had branch offices of large banks. Federal Reserve Bank of St. Louis REVIEW May/June 2013 207 Gilbert and Wheelock Figure 2 Average Deposit Shares in Rural Counties A. <10,000 Population 0.9 0.8 0.7 Bank Size 0.6 Small Small-Mid Mid-Large Large 0.5 0.4 0.3 0.2 0.1 0.0 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 B. >10,000 Population 0.9 Bank Size Small Small-Mid Mid-Large Large 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 2001 208 May/June 2013 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Figure 3 Presence of Large Banks in U.S. Rural Counties A. 50 Percent Threshold Urban (MSA) counties or counties with no banks (2001-12) Rural counties where large banks held less than 50 percent of total deposits Rural counties where large banks held 50 percent or more of total deposits B. 90 Percent Threshold Urban (MSA) counties or counties with no banks (2001-12) Rural counties where large banks held less than 90 percent of total deposits Rural counties where large banks held 90 percent or more of total deposits Federal Reserve Bank of St. Louis REVIEW May/June 2013 209 Gilbert and Wheelock Many states relaxed or removed barriers to within-state branching between 1980 and 1995. Many also began to permit entry by out-of-state bank holding companies. Interstate branching was permitted beginning in 1997. Gilbert (2000) found that the rural market shares of large banking organizations rose in most states during the 1980s and 1990s, but especially in those states that had limited or prohibited branching before 1980. He also found that large organizations generally had higher market shares in more densely populated counties and in low-populationdensity counties in states that permitted statewide branching in 1980. Gilbert (2000) also found that some regional differences in market shares of large banks were still present in 1999. Specifically, large banks continued to have higher average rural deposit shares in regions where most states permitted statewide branching in 1980. Figures 3A and 3B show that these patterns were still present in 2012. Figure 3A shows that rural counties in which very large banking organizations held at least 50 percent of county deposits were more prevalent in the western states and on the East Coast, specifically in Maryland and North and South Carolina. The western states and these East Coast states allowed statewide branching in 1980. Similarly, Figure 3B shows that the few rural counties in which the very largest organizations held at least 90 percent of county deposits were primarily in western states that permitted statewide branching in 1980. Figures 4A through 4I show the average rural county deposit shares of different-sized banks in 2001, 2006, and 2012 by U.S. Census division. The states in each Census division are listed in Table 1, which also identifies states that permitted statewide branching in 1980. Similar to Gilbert’s (2000) findings for the 1990s, Figures 4A through 4I show that the very largest banks had the largest rural deposit shares in regions where most states had long permitted branch banking, such as the South Atlantic, Mountain, and Pacific regions. The deposit shares of very large banks tend to be lower in regions where most states restricted branching in 1980, such as the West North Central, East South Central, and West South Central regions. An exception is New England, where most states permitted statewide branching in 1980 but rural deposit shares of the largest organizations were relatively low in the 2000s. Turning to small banks, Figures 4A through 4I show that small banking organizations had the largest average rural deposit shares in regions where states tended to restrict branching before the 1980s. Small bank shares were especially high in the West North Central and West South Central regions, where nearly all states restricted branching before the 1980s. For example, in 2012, on average, 81 percent of the deposits of rural counties in the West North Central region were held by organizations with less than $1 billion in assets. By contrast, on average, only 28 percent of the deposits of rural counties in the Pacific region were held by the smallest organizations. Data for 2001 through 2012 indicate little or no convergence across Census divisions in the rural deposit shares of community banks. Figures 5A and 5B show more clearly the differences in the rural deposit shares of differentsized banks between branching and non-branching states in 1980. Although most states relaxed their restrictions on branch banking in the 1980s and interstate branching has been permitted since 1997, large differences remained as recently as 2012 in the average rural county deposit shares of banks grouped by size between states that permitted branching in 1980 and states that did not. 210 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Figure 4 Average Deposit Shares in Rural Counties A. New England Census Division B. Middle Atlantic Census Division 0.9 0.9 2001 0.8 2001 0.8 2006 0.7 2006 0.7 2012 2012 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Small Small-Mid Mid-Large Small Large Small-Mid Mid-Large Large Bank Size Bank Size C. East North Central Census Division D. West North Central Census Division 0.9 0.9 2001 0.8 2001 0.8 2006 0.7 2006 0.7 2012 2012 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Small Small-Mid Mid-Large Bank Size Large Small Small-Mid Mid-Large Large Bank Size Cont’d on p. 212. The persistence of regional differences in rural county deposit shares more than two decades after most states began to permit statewide branching and interstate banking is somewhat puzzling. However, subsequent to the enactment of the Riegel-Neal Interstate Banking and Branching Efficiency Act in 1994, many states imposed various limits on both interstate branching, including prohibitions on establishing branches de novo (rather than through acquisition of existing banks) and acquiring recently chartered banks. Several states also imposed statewide deposit caps. According to Rice and Strahan (2010), such restrictions continued to affect local banking market structures well after interstate branching was permitted in 1997. Federal Reserve Bank of St. Louis REVIEW May/June 2013 211 Gilbert and Wheelock Figure 4, cont’d Average Deposit Shares in Rural Counties E. South Atlantic Census Division F. East South Central Census Division 0.9 0.9 2001 0.8 2001 0.8 2006 0.7 2006 0.7 2012 2012 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Small Small-Mid Mid-Large Small Large Small-Mid Mid-Large Large Bank Size Bank Size G. West South Central Census Division H. Mountain Census Division 0.9 0.9 2001 0.8 2001 0.8 2006 0.7 2006 0.7 2012 2012 0.6 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 0.0 Small Small-Mid Mid-Large Large Small Small-Mid Mid-Large Large Bank Size Bank Size I. Pacific Census Division 0.9 2001 0.8 2006 0.7 2012 0.6 0.5 0.4 0.3 0.2 0.1 0.0 Small Small-Mid Mid-Large Large Bank Size 212 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock Table 1 States in Census Divisions New England West North Central East South Central Mountain Connecticut* Iowa Alabama Arizona* Maine* Kansas Kentucky Colorado Massachusetts Minnesota Mississippi Idaho* New Hampshire* Missouri Tennessee New Mexico Rhode Island* Nebraska Vermont* North Dakota South Dakota* Montana West South Central Arkansas Louisiana Middle Atlantic New Jersey* South Atlantic Oklahoma New York* Delaware* Texas Pennsylvania District of Columbia East North Central Indiana Illinois Michigan Ohio Wisconsin Utah Nevada* Wyoming Pacific Alaska* Florida California* Georgia Hawaii* Maryland* Oregon* North Carolina* Washington* South Carolina* Virginia West Virginia NOTE: States marked with an asterisk permitted statewide branching in 1980, subject to various restrictions on branching in the communities where banks had their headquarters (Gilbert, 2000). Large Bank Presence in Rural and Urban Markets After several years of increased branching, the number of bank branches in the United States has changed little since 2008, and several large banks have been pruning offices. Bank of America alone eliminated 193 branches in 2012, many of which were in rural markets and areas especially hit hard by falling home prices and the recent recession (Sidel, 2013). Community banks have acquired some branches shed by large banking organizations in rural and smaller urban markets that larger banks have abandoned (Peters, 2012). Still, large banking organizations continue to have a significant presence in many rural counties and, as shown in Figure 1, maintained a stable share of rural county deposits between 2001 and 2012. Table 2, which provides information on the location of offices of banking organizations that held at least $50 billion in assets in 2012, indicates that the presence of large banking organizations in rural counties varies substantially among large organizations. In general, organizations that only recently became bank holding companies or whose traditional business lines were insurance or other services besides commercial banking (e.g., Goldman Sachs Group, MetLife, and American Express Company) have banking offices in fewer counties than holding companies whose traditional business was dominated by commercial banking. Similarly, Bank of New Federal Reserve Bank of St. Louis REVIEW May/June 2013 213 Gilbert and Wheelock Figure 5 Average Deposit Shares in Rural Counties B. States that Did Not Permit Branching in 1980 A. States that Permitted Branching in 1980 0.9 0.9 2001 0.8 2006 0.7 2012 0.6 2001 0.8 2012 0.6 0.5 0.5 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0.0 2006 0.7 0.0 Small Small-Mid Mid-Large Bank Size Large Small Small-Mid Mid-Large Large Bank Size York Mellon Corporation and State Street Corporation, which focus mainly on providing financial services to other financial institutions rather than retail customers, have no offices in rural counties. Other large organizations (e.g., Ally Financial and Discover Financial) offer specific financial services to consumers throughout the country from offices in one or two urban locations using advertising and communications technology. By contrast, the large banking organizations that operate many branch offices all had their roots in commercial banking. Wells Fargo & Company and Bank of America Corporation, for example, have offices in the largest number of counties throughout the United States. The percentage of corporate banking offices and deposit shares in rural counties varies somewhat across even the traditional commercial banking organizations, however; this largely reflects the different regions of the country where the organizations are currently or were originally headquartered or where the organizations have made significant acquisitions. For example, Wells Fargo & Company, U.S. Bancorp, Regions Financial Corporation, and BB&T Corporation all have branches in numerous rural counties. Wells Fargo and BB&T also have comparatively high average deposit shares in rural counties in which they have locations. The headquarters of both Wells Fargo and BB&T are in states that have long permitted statewide branch banking— California in the case of Wells Fargo and North Carolina in the case of BB&T. Wells Fargo also has numerous branches in North Carolina as a result of its acquisition of Wachovia Corporation in 2008. Thus, the relatively high average rural deposit shares of these organizations is consistent with evidence that large organizations generally have larger rural deposit shares in states that permitted branching in 1980. 214 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Federal Reserve Bank of St. Louis REVIEW Table 2 Large Banking Organizations in Urban and Rural Counties: Number of Counties and Average Deposit Share in Counties Where They Have Offices (June 2012) Number of counties by type Banking organization Total assets ($ millions) Urban (total 1,097) Rural (total 2,017) Smallpopulation rural (total 625) Average deposit share by county type Largerpopulation rural (total 1,392) Urban Rural Smallpopulation rural Largerpopulation rural JPMorgan Chase & Co. 2,290,146 344 98 1 97 0.10 0.12 0.08 0.12 Bank of America Corporation 2,162,083 477 236 10 226 0.11 0.12 0.41 0.10 Citigroup Inc. 1,916,451 117 14 3 11 0.04 0.06 0.09 0.05 Wells Fargo & Company 1,336,204 526 404 67 337 0.17 0.21 0.42 0.17 0.12 0.26 0.10 Goldman Sachs Group, Inc. 948,981 4 0 0 0 0.06 MetLife, Inc. 825,188 1 0 0 0 0.24 U.S. Bancorp 353,136 283 313 39 274 0.09 Bank of New York Mellon Corporation 330,490 42 0 0 0 0.01 PNC Financial Services Group, Inc. 299,712 297 138 1 137 0.09 0.14 0.64 0.14 Capital One Financial Corporation 296,698 90 34 1 33 0.11 0.14 0.33 0.13 State Street Corporation 200,369 1 0 0 0 0.38 Ally Financial Inc. 178,560 1 0 0 0 0.20 BB&T Corporation 178,529 260 172 13 159 0.15 0.20 0.43 0.19 SunTrust Banks, Inc. 178,307 192 62 2 60 0.12 0.12 0.37 0.11 American Express Company 146,890 1 0 0 0 0.10 Regions Financial Corporation 122,345 226 201 12 189 0.10 0.16 0.39 0.15 Fifth Third Bancorp 0.12 0.23 0.12 155 73 2 71 0.09 94,456 41 0 0 0 0.02 Keycorp 86,741 128 75 1 74 0.07 0.11 1.00 0.10 M&T Bank Corporation 80,808 100 34 1 33 0.08 0.08 0.20 0.08 0.93 0.93 0.11 0.11 Discover Financial Services 73,256 1 1 0 1 0.00 Charles Schwab Bank 72,156 5 0 0 0 0.19 Comerica Incorporated 62,757 49 4 0 4 0.05 NOTE: Small rural counties had a total population of less than 10,000 as of 2010. The banking organizations listed had at least $50 billion in assets in 2012. 215 Gilbert and Wheelock May/June 2013 117,543 Northern Trust Corporation Gilbert and Wheelock CONCLUSION If the community bank business model is apt to succeed anywhere, it should be in rural markets. The cost of relationship lending—a hallmark of community banking—may be lower in rural areas than in urban areas, and rural markets likely have higher percentages of borrowers for whom there is limited quantitative information. Data on the shares of rural county bank deposits held by banking organizations of different sizes indicate that, on the whole, large banks have not driven small banks from rural banking markets. Although the deposit shares of small banks in rural counties and small towns declined during the 1980s and 1990s, between 2001 and 2012, the share of rural county deposits held by banks with less than $1 billion in assets changed little, as did the share held by banks with more than $50 billion in assets. Recent sales of rural-market branches by large banking organizations to smaller banks suggest that large banks may find rural markets less profitable than do small banks. Perhaps the impact of advances in information-processing technology in altering the competitive balance between large and small banks occurred prior to 2001, but other explanations for the recent stability of deposit shares of different-sized banks in rural markets are possible. Rural counties have generally experienced slower population and economic growth than urban areas in recent years, and large banks may have chosen to focus their operations in urban markets and cede business to smaller banks in slower-growing and less-profitable rural markets. If that explanation is true, then the impact of technological change may produce further shifts in the competitive balance between large and small banks as large banks eventually devote resources to building market share in the more profitable rural areas. Data on the rural banking offices of individual large banking organizations may indicate another explanation for the observations on deposit shares in rural markets: Perhaps changes in technology have made it easier over time for large organizations to provide financial services without networks of physical offices. Of course, the structure of banking markets can also be affected by regulation, antitrust enforcement, and changes in the macroeconomic environment. It is beyond the scope of this article to test alternative explanations for changes in the average market shares of banks of different sizes. However, the evidence that the average rural county deposit share of large banking organizations did not increase between 2001 and 2012 is at least suggestive that small banks remain competitive in their main market segments. The future, of course, is less certain. Advances in information technology continue to lower the cost of acquiring hard information about potential borrowers and facilitate obtaining banking and other financial services by means of the Internet. Access to a brick-and-mortar bank office has become less important for many bank customers, which may favor larger banks that can spread the fixed costs of information technology over more customers.13 However, evidence from micro studies of community banks (such as Gilbert, Meyer, and Fuchs, 2013) indicates that well-managed community banks can thrive even in challenging times. The relative stability of deposit shares of small banks over the past decade suggests further that community banks, as a group, remain competitive with larger banking organizations, at least in markets where informationally opaque borrowers are most prevalent. 216 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Gilbert and Wheelock NOTES 1 Data on the number of commercial banks and offices are from the Federal Deposit Insurance Corporation (FDIC), Table CB03 (http://www2.fdic.gov/hsob/HSOBRpt.asp). 2 Data on the total assets and deposits held by commercial banks of various sizes are available from the FDIC’s Statistics on Banking (http://www2.fdic.gov/SDI/SOB/). 3 See Stern and Feldman (2009) or Rosenblum (2011) for discussions about too-big-to-fail policy. 4 Some evidence in support of this assumption is discussed in DeYoung et al. (2012). 5 See Berger et al. (2005). 6 DeYoung et al. (2011) find that the increased use of credit scoring has been an important cause of the increase in average distance between small businesses and their lenders, especially since 1993. 7 Credit unions and other types of financial intermediaries are also present in many rural markets. However, our data include only commercial banks and savings institutions, which are the only depository institutions for which comprehensive data on the location of branch offices are available. 8 Banking organizations include bank and thrift holding companies and commercial banks and savings institutions that are not owned by a holding company. 9 A few studies use more-refined definitions of community banks. For example, the FDIC (2012) includes all banking organizations with less than $1 billion of assets (indexed for inflation to 2010), as well as some larger organizations with high ratios of loans to assets or core deposits to assets, or those that operate in few markets. However, in the absence of a standard, widely accepted definition of community banks, we group banks solely on the basis of asset size. 10 Some bank holding companies are controlled by other holding companies. In those cases, we aggregate all data to the level of the top holding company. 11 These data are available from the FDIC (http://www2.fdic.gov/sod/). 12 The FDIC (2012) found that the average rural market share of community banks remained stable during 2001-11. As noted previously (endnote 9), the FDIC (2012) defines community banks on the basis of both size and scope criteria. Hence, our study provides evidence on whether market share trends are sensitive to the use of the alternative criteria for defining community banks. 13 See Wheelock and Wilson (2012) for more discussion and evidence of increasing returns to scale in banking. REFERENCES Berger, Allen N. “The Economic Effects of Technological Progress: Evidence from the Banking Industry.” Journal of Money, Credit, and Banking, April 2003, 35(2), pp. 141-76. Berger, Allen N.; Cowan, Adrian M. and Frame, W. Scott. “The Surprising Use of Credit Scoring in Small Business Lending by Community Banks and the Attendant Effects on Credit Availability, Risk, and Profitability.” Journal of Financial Services Research, April 2011, 39(1-2), pp. 1-17. Berger, Allen N.; Miller, Nathan H.; Petersen, Mitchell A.; Rajan, Raghuram G. and Stein, Jeremy C. “Does Function Follow Organizational Form? Evidence from the Lending Practices of Large and Small Banks.” Journal of Financial Economics, May 2005, 76(2), pp. 237-69. Bernanke, Ben S. “Community Banking and Community Bank Supervision.” Remarks at the Independent Community Bankers of America National Convention and Techworld, Las Vegas, Nevada, March 8, 2006; http://www.federalreserve.gov/newsevents/speech/bernanke20060308a.htm. Brickley, James A.; Linck, James S. and Smith, Clifford W. Jr. “Boundaries of the Firm: Evidence from the Banking Industry.” Journal of Financial Economics, December 2003, 70(3), pp. 351-83. Cole, Rebel A.; Goldberg, Lawrence G. and White, Lawrence J. “Cookie Cutter vs. Character: The Micro Structure of Small Business Lending by Large and Small Banks.” Journal of Financial and Quantitative Analysis, June 2004, 39(2), pp. 227-51. Federal Reserve Bank of St. Louis REVIEW May/June 2013 217 Gilbert and Wheelock Cortés, Kristle Romero. “Did Local Lenders Forecast the Bust? Evidence from the Real Estate Market.” Working Paper No. 12-26, Federal Reserve Bank of Cleveland, November 2012; http://www.clevelandfed.org/research/workpaper/2012/wp1226.pdf. Cyree, Ken B. and Spurlin, W. Paul. “The Effects of Big-Bank Presence on the Profit Efficiency of Small Banks in Rural Markets.” Journal of Banking and Finance, September 2012, 36(9), pp. 2593-603. DeYoung, Robert; Frame, W. Scott; Glennon, Dennis and Nigro, Peter. “The Information Revolution and Small Business Lending.” Journal of Financial Services Research, April 2011, 39(1), pp. 19-33. DeYoung, Robert; Glennon, Dennis and Nigro, Peter. “Borrower-Lender Distance, Credit Scoring, and Loan Performance: Evidence from Informationally-Opaque Small Business Borrowers.” Journal of Financial Intermediation, January 2008, 17(1), pp. 113-43. DeYoung, Robert; Glennon, Dennis; Nigro, Peter and Spong, Kenneth. “Small Business Lending and Social Capital: Are Rural Relationships Different?” CBE Working Paper No. 2012-1, KU Center for Banking Excellence, University of Kansas School of Business, June 2012; http://www.business.ku.edu/sites/businessdev.drupal.ku.edu/files/docs/ CBE%20WP%202012-1%20DeYoung%20Glennon%20Nigro%20Spong.pdf. DeYoung, Robert; Hunter, William C. and Udell, Gregory F. “The Past, Present, and Probable Future for Community Banks.” Journal of Financial Services Research, April 2004, 25(2-3), pp. 85-133. Federal Deposit Insurance Corporation. FDIC Community Banking Study. Washington, DC: Federal Deposit Insurance Corporation, December 2012; http://www.fdic.gov/regulations/resources/cbi/study.html. Gilbert, R. Alton. “Nationwide Branch Banking and the Presence of Large Banks in Rural Areas.” Federal Reserve Bank of St. Louis Review, May/June 2000, 82(3), pp. 13-28; http://research.stlouisfed.org/publications/review/00/05/05ag.pdf. Gilbert, R. Alton; Meyer, Andrew P. and Fuchs, James W. “The Future of Community Banks: Lessons from Banks That Thrived During the Recent Financial Crisis.” Federal Reserve Bank of St. Louis Review, March/April 2013, 95(2), pp. 115-43; http://research.stlouisfed.org/publications/review/13/03/gilbert.pdf. Hannan, Timothy H. and Prager, Robin A. “The Profitability of Small Single-Market Banks in an Era of Multi-Market Banking.” Journal of Banking and Finance, February 2009, 33(2), pp. 263-71. Peters, Andy. “Community Banks Swoop in on B of A’s Unwanted Branches.” American Banker, December 28, 2012; http://www.americanbanker.com/issues/177_249/community-banks-swoop-in-on-b-of-a-s-unwanted-branches1055494-1.html. Petersen, Mitchell A. and Rajan, Raghuram G. “The Benefits of Lending Relationships: Evidence from Small Business Data.” Journal of Finance, March 1994, 49(1), pp. 3-37. Petersen, Mitchell A. and Rajan, Raghuram G. “Does Distance Still Matter? The Information Revolution in Small Business Lending.” Journal of Finance, December 2002, 57(6), pp. 2533-70. Pilloff, Steven J. “Does the Presence of Big Banks Influence Competition in Local Markets?” Journal of Financial Services Research, May 1999, 15(3), pp. 159-77. Rice, Tara and Strahan, Philip E. “Does Credit Competition Affect Small-Firm Finance?” Journal of Finance, June 2010, 65(3), pp. 861-89. Rosenblum, Harvey. “Choosing the Road to Prosperity: Why We Must End Too Big to Fail—Now.” Federal Reserve Bank of Dallas Annual Report 2011, pp. 2-23; http://www.dallasfed.org/assets/documents/fed/annual/2011/ar11b.pdf. Sidel, Robin. “After Years of Growth, Banks Are Pruning Their Branches.” Wall Street Journal, March 31, 2013; http://online.wsj.com/article/SB10001424127887323699704578326894146325274.html. Stein, Jeremy C. “Information Production and Capital Allocation: Decentralized versus Hierarchical Firms.” Journal of Finance, October 2002, 57(5), pp. 1891-921. Stern, Gary H. and Feldman, Ron J. Too Big to Fail: The Hazards of Bank Bailouts. Washington, DC: Brookings Institution Press, 2009. Wheelock, David C. and Paul W. Wilson. “Do Large Banks Have Lower Costs? New Estimates of Returns to Scale for U.S. Banks.” Journal of Money, Credit, and Banking, February 2012, 44(1), pp. 171-99. 218 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Foreign Currency Loans and Systemic Risk in Europe Pınar Yeşin Foreign currency loans to the unhedged non-banking sector are remarkably prevalent in Europe and create a significant exchange-rate-induced credit risk to European banking sectors. In particular, Swiss franc (CHF)-denominated loans, popular in Eastern European countries, could trigger simultaneous bank failures if depreciation of the domestic currencies prevents unhedged borrowers from servicing the loans. Foreign currency loans thus pose a systemic risk from a “common market shock” perspective. The author uses a novel dataset of foreign-currency loans from 17 countries for 2007-11 (collected by the Swiss National Bank) and builds on the method suggested by Ranciere, Tornell, and Vamvakidis (2010) to quantify this systemic risk. The author finds that systemic risk is substantial in the non-euro area, while it is relatively low in the euro area. However, CHF-denominated loans are not the underlying source of the high systemic risk: Loans denominated in other foreign currencies (probably to a large extent in euros) contribute significantly more to the systemic risk in the non-euro area than CHFdenominated loans. Furthermore, systemic risk shows high persistence and low volatility during the sample period. The author also finds that banks in Europe have continuously held more foreign-currencydenominated assets than liabilities, indicating their awareness of the exchange-rate-induced credit risk they face. (JEL F3, G2) Federal Reserve Bank of St. Louis Review, May/June 2013, 95(3), pp. 219-35. ystemic risk in the financial system is a multifaceted phenomenon. First, it can arise in the form of contagion among financial institutions: The failure of one institution can trigger cascading defaults of the others through their linkages in the interbank market. Second, systemic risk can arise in the form of joint failures of financial institutions as a result of their exposure to a “common market shock,” as was the case during the recent financial crisis S Pınar Yeşin is a senior economist at the Swiss National Bank and an economics lecturer at the University of Zurich. The author was a visiting scholar at the Research Division of the Federal Reserve Bank of St. Louis while working on this article and thanks colleagues there for their hospitality. The author thanks an anonymous referee; Silvio Contessi, Andreas Fischer, Steven Ongena, Adrian Peralta-Alva, B. Ravikumar, and Rajdeep Sengupta; in addition to seminar participants at the Swiss National Bank and Federal Reserve Bank of St. Louis; and conference participants at the 10th International Conference on Credit Risk Evaluation (“Stability and Risk Control in Banking, Insurance and Financial Markets”) in Venice, the 9th Annual NBP-SNB Joint Seminar (“Managing Capital Flows for Monetary and Financial Stability”) in Poland, the 4th International IFABS Conference (“Rethinking Banking and Finance: Money, Markets and Models”) in Valencia, and the 3rd International Conference of the Turkish Economic Association (“Debt Dynamics, Financial Instability and the Great Recession”) in Izmir for their helpful comments and discussions. The author also extends thanks to the 19 European central banks that participated in the Swiss National Bank’s CHF Lending Monitor project and commented on an earlier version of this article. © 2013, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, the regional Federal Reserve Banks, or the Swiss National Bank. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW May/June 2013 219 Yeşin when mortgage-backed securities lost significant value and simultaneously damaged the balance sheets of countless banks. And third, systemic risk can develop in the form of informational spillovers when bad news about one financial institution increases the refinancing costs of the others, which in turn increases the probability of their own collapse.1 This article attempts to measure systemic risk in European banking sectors arising from foreign currency loans to domestic non-banks.2 To do so, it adheres to the common market shock view, whereby banks could fail jointly as a result of their exposure to unhedged non-banks that borrowed in a foreign currency but have no steady income in that currency. In this case, the common market shock would be a sharp exchange rate movement triggering defaults of, for example, domestic households on their foreign-currency-denominated mortgages. Such a shock would lead to a simultaneous deterioration of the financial condition of numerous banks in Europe because they hold sizable foreign currency loans to domestic households on the asset side of their balance sheets and have short-term foreign currency liabilities in the interbank market. Foreign currency loans are remarkably prevalent in Europe. Figure 1 shows that in 2011:Q3 the majority of the outstanding loans to the non-banking sector in the non-euro area countries were denominated in a foreign currency. For example, in Hungary, Bulgaria, and Romania, more than 60 percent of the outstanding loans to non-banks are denominated in a foreign currency. Even if borrowing in euros might be considered rational for some non-euro area countries that are “closer” to adopting the euro,3 many other European countries will not officially join the euro area for several more years. Moreover, there is a considerable amount of borrowing in foreign currencies other than the euro by the non-banking sector in Europe. For example, loans denominated in Swiss francs (CHF) constitute a significant share of total loans in some countries, such as Austria, Croatia, Hungary, and Poland (see Figure 1). Remarkably, the European household sector in particular is borrowing in foreign currencies.4 Figure 2 shows the sectoral breakdown of CHF loans to the non-banking sector as of 2011:Q3. Note that in almost all countries in the sample, more than 40 percent of the outstanding CHF loans are made to the domestic household sector (the exceptions are the Czech Republic, the United Kingdom, Germany, and Luxembourg). Only in the United Kingdom and Luxembourg is a large share of CHF loans made to nonresidents,5 reflecting the status of these countries as international financial centers. Therefore, foreign currency loans in Europe are mainly considered “small men’s carry trade”: Households (and small firms) borrow in a lower-yielding foreign currency and invest in a highyielding domestic currency (e.g., in the form of a home mortgage or car loan). While doing so, they expect the domestic currency to continue to appreciate as it did in the past, a recurring violation of the uncovered interest parity condition.6 While borrowing in a foreign currency offers some immediate benefits—such as lower interest rates and longer maturities—for the non-banking sector, these loans also carry a significant exchange rate risk. The domestic household sector is unlikely to have any income in a foreign currency or to use sophisticated hedging instruments against the exchange rate risk because of the lack of hedging instruments, willing counterparties, or deep financial markets. The fact that the household sector is unhedged manifests itself, for example, in the nonexistence of forward 220 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin Figure 1 Share of Foreign Currency Loans as a Percentage of Total Loans to the Non-Banking Sector in Europe (2011:Q3) Percentage of Total Loans CHF Loans/Total Loans Non-Euro Area Slovakia Italy Slovenia Germany France Greece Austria Luxembourg Czech Rep. Denmark Poland U.K. Romania Bulgaria Hungary Croatia Serbia Other FCY Loans/Total Loans Latvia 100 90 80 70 60 50 40 30 20 10 0 Euro Area NOTE: CHF, Swiss francs; FCY, foreign currency. SOURCE: CHF Lending Monitor. Figure 2 Sectoral Breakdown of CHF Loans to the Non-Banking Sector (2011:Q3) Percentage of CHF Loans Households Nonfinancial Corporations General Government and Other Non-Banks Germany Italy Austria Slovakia Greece U.K. Euro Area Luxembourg Non-Euro Area Czech Rep. Denmark Hungary Bulgaria Serbia Croatia Romania 100 90 80 70 60 50 40 30 20 10 0 Nonresidents NOTE: CHF, Swiss francs. SOURCE: CHF Lending Monitor. Federal Reserve Bank of St. Louis REVIEW May/June 2013 221 Yeşin exchange rates of the Eastern European currencies vis-à-vis the CHF. To hedge against the exchange rate risk, a household in Eastern Europe with a mortgage denominated in CHF would need to take two hedging positions: one vis-à-vis the euro and another one between the euro and the CHF. Since each hedging instrument also entails a transaction cost, the “profit” from borrowing in a low-yielding currency would simply vanish in this case. There are no studies in the literature that document the hedging behavior of households borrowing in foreign currencies. However, Pann, Seliger, and Übeleis (2010) refer to a survey by the Austrian Central Bank and state that “foreign currency lending to naturally hedged households in CESEE [Central, Eastern, and Southeastern Europe] and the CIS [Commonwealth of Independent States] was negligible (lower than 10 percent).”7 Therefore, a sharp depreciation of the domestic currency can prevent borrowers from being able to service their foreign currency loans.8 As a result, the sizable foreign currency loans on banks’ balance sheets can pose a significant systemic risk. The European Systemic Risk Board (ESRB), an independent European Union (EU) institution monitoring financial stability within the EU, recognized the systemic risk that foreign currency loans pose to European banking sectors and made an official recommendation on lending in foreign currency on November 22, 2011.9 In particular, the ESRB stated the following: (1) Foreign currency lending to unhedged borrowers has increased in a number of Union Member States. (2) Excessive foreign currency lending may produce significant systemic risks for those Member States and may create conditions for negative cross-border spillover effects. Similarly, the Hungarian Financial Supervisory Authority voiced its concerns regarding systemic risk arising from foreign currency loans in the Hungarian banking sector in 2011 and advised the banks in Hungary to swap the CHF-denominated loans into euro loans to reduce the systemic risk (Reuters, 2011). In fact, new foreign currency loans to non-banks have been prohibited in Hungary since the summer of 2010. However, the outstanding foreign currency loans to non-banks still pose a significant risk to the economy and Hungarian authorities have been working on a plan to contain it. The European Bank for Restructuring and Development (EBRD) also acknowledged foreign currency loans in the EBRD region10 as a “key vulnerability” and therefore launched a major initiative in 2010 to develop local currency and capital markets to help reduce unhedged foreign currency borrowing (see ERBD, 2010). While policymakers in Europe repeatedly express their concern regarding the systemic risk created by foreign currency loans,11 there is a lack of literature on the exact measurement of this systemic risk. The previous literature cannot answer questions such as the following: • What is the magnitude of the systemic risk for European banking sectors arising from foreign currency loans to the non-banking sector? • To what extent would the banking sectors’ balance sheets be affected if the (unhedged) non-banking sector could no longer service its foreign currency debt because of a sudden exchange rate movement? 222 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin • Are CHF-denominated loans the main driver of a systemic risk in certain countries, as perceived by policymakers and the general public? This article attempts to fill this void in the literature by using a novel dataset of foreign currency loans in Europe and building on a method suggested by Ranciere, Tornell, and Vamvakidis (2010). The Swiss National Bank (SNB) has collected quarterly data on foreign currency loans from 19 European central banks since 2009 under its CHF Lending Monitor project. “Systemic risk measure” is defined here as the impact of writing down the unhedged foreign currency loans on banks’ balance sheets. This measure takes into account the indirect exchange rate risk that banks take on when they lend to unhedged borrowers (i.e., domestic households and nonfinancial corporations). In particular, it calculates the net unhedged foreign currency liabilities as a share of total assets. The larger this number is in a given country, the higher is the systemic risk in its banking sector. The four distinct contributions of this article to the literature are as follows: (i) Since data from the CHF Lending Monitor are reliable and consistent across countries and are more detailed than the information that has been publicly available until now, these new systemic risk measures are more accurate than those in Ranciere, Tornell, and Vamvakidis (2010). (ii) I can separately identify the aggregate systemic risk arising from all foreign currency loans and the particular systemic risk resulting from CHF loans in Europe. (iii) A sectoral breakdown of systemic risk is possible in this dataset. (iv) The quarterly data of the CHF Lending Monitor allow me to observe the evolution of the currency mismatch on a quarterly basis from 2009:Q1 until 2011:Q3. For a subset of countries, the measure can be calculated as far back as 2002:Q1.12 First, I find that the systemic risk arising from foreign currency loans is quite high in the majority of the non-euro area countries: Net unhedged foreign currency liabilities of the banking sectors correspond to more than 20 percent of their total assets. However, net unhedged CHF liabilities never exceed more than 5 percent of their total assets. Therefore, I conclude that CHF loans are not the main driver of a systemic risk in Europe, despite the common belief that CHF loans pose the greatest threat among foreign currency loans. Second, the sectoral breakdown of systemic risk shows that banks attempt to hedge against the foreign-currency-induced credit risk by persistently holding more foreign currency assets than foreign currency liabilities on their balance sheets. However, the mere size of the foreign currency loans to unhedged households exceeds this buffer in most countries, resulting in a positive and, in certain countries, sizable systemic risk. Third, I find that systemic risk has been quite persistent and nonvolatile in Europe since 2007. That is, short-term policies would be unable to swiftly reduce that systemic risk. This also confirms the findings of previous research that non-banks take long-term loans in foreign currency and are not involved in short-term speculative carry trade activity based solely on shortterm changes in the macroeconomic environment.13 Federal Reserve Bank of St. Louis REVIEW May/June 2013 223 Yeşin This article is structured as follows: The next section provides an overview of the previous literature on foreign currency loans and systemic risk. The two following sections explain the method and the data used, respectively. The findings are then laid out, with conclusions and policy implications discussed in the final section of the article. PREVIOUS LITERATURE The previous literature on foreign currency borrowing focuses mostly on household-, firm-, or bank-level survey data to understand the factors at play. A common finding is that households and small firms may engage in excessive risk-taking because (i) they do not necessarily have foreign currency income and (ii) foreign banks in Eastern Europe may be a driver of foreign currency loans in the region. For example, Brown, Ongena, and Yeşin (2011) study firm- and country-level determinants of foreign currency borrowing by small firms in transition countries. Although they find that firms with foreign currency income are more likely to borrow in foreign currency, local currency earners also borrow in foreign currency to a significant degree. Similarly, Beer, Ongena, and Peter (2010) empirically analyze foreign currency borrowing by households in Austria and find that risk-seeking, affluent, and married households are more likely to take a housing loan denominated in a foreign currency. Brown, Kirschenmann, and Ongena (2011), on the other hand, study loan application and approval data at a bank in Bulgaria; they find that foreign currency borrowing may be partly supply driven by banks hesitant to give longterm loans in the local currency.14 Furthermore, Brown, Ongena, and Yeşin (2013) show in a simple banking model that persistent violation of the uncovered interest parity may lead to more foreign currency borrowing in equilibrium if the banks have imperfect information about the revenue level and currency of borrowing firms. Similarly, Degryse et al. (2012) find that foreign banks that entered the emerging markets through greenfield investment tend to extend more loans in foreign currency, possibly because of their easier access to foreign currency funding in international money markets. All in all, the existing literature on foreign currency loans points to the emergence of a systemic risk in European banking sectors.15 In this article, systemic risk corresponds to the joint failure of several banks as a result of their exposure to a common market shock (i.e., the simultaneous default of unhedged foreign currency borrowers as the result of a sharp exchange rate movement) because banks tend to hold similar assets on their balance sheets. Undoubtedly, contagion by way of the interbank market or information spillovers among banks could represent a significant part of the systemic risk arising from foreign currency loans. However, the emerging literature on systemic risk since the financial crisis perceives common shocks as an equally important threat to the system and highlights the need to further understand their source and their impact. For example, Acharya and Yorulmazer (2008) show that banks herd and undertake correlated investments to minimize the impact of information contagion on the expected cost of borrowing. Such herding among banks would lead to greater exposure of the financial system to a common shock. Currently, available data are insufficient to conduct a contagion analysis in this regard. Cerutti, Claessens, and McGuire (2011) argue that systemic risk analysis is very difficult, even impossible, with the existing bank-level data and suggest the collection of consistent bank-level 224 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin data showing aggregate positions and linkages to appropriately measure and monitor systemic risk both within countries and across borders. Therefore, this article adopts the common shock approach rather than the contagion or informational spillover perspective to measure the systemic risk arising from foreign currency loans. METHOD I build on the approach suggested by Ranciere, Tornell, and Vamvakidis (2010) to calculate a measure of systemic risk for European countries’ banking sectors. This systemic risk measure calculates the net unhedged foreign currency liabilities as a percentage of total assets and can be characterized, in fact, as a novel currency mismatch index. This customized currency mismatch index takes into account the indirect exchange rate risk that banks assume when they lend to unhedged borrowers. In other words, it considers the “exchange-rate-induced credit risk” and is an evaluation of the currency mismatch on the balance sheets of European banking sectors in the case of a joint failure of households (and nonfinancial corporations) to service their foreign currency loans resulting from a sharp appreciation of the foreign currency. In the existing literature, currency mismatch in a banking sector is usually measured as the net foreign currency liabilities (i.e., the difference between foreign currency liabilities and foreign currency assets) as a share of the total assets of the banking sector. However, banks usually match their foreign currency assets and foreign currency liabilities so that their difference would be almost (or sometimes by regulation identical to) zero. Furthermore, this simple measure treats all foreign currency assets equally without considering the risks associated with foreign currency loans given to unhedged borrowers. Therefore, Ranciere, Tornell, and Vamvakidis (2010) propose calculating the net foreign currency liabilities as a share of total assets, but excluding the “risky” foreign currency assets from the foreign currency assets. In other words, foreign currency loans given to the domestic16 non-banking sector are not included in total foreign currency assets. Thus, in a given country the foreign currency mismatch index in the banking sector is equal to net foreign-currencydenominated liabilities plus unhedged foreign currency assets divided by total assets. Therefore, Foreign currency mismatch index = FCY liabilities − ( FCY assets − FCY loans to resident households and nonfinancial corporations ) , Total assets where FCY denotes foreign currency. As detailed information on CHF-denominated loans is available in the CHF Lending Monitor, the same method can be applied to calculate the systemic risk arising from CHF loans only. Thus, CHF mismatch index = CHF liabilities − (CHF assets − CHF loans to resident households and nonfinancial corporations) Total assets . Similarly, Federal Reserve Bank of St. Louis REVIEW May/June 2013 225 Yeşin Other FCY mismatch index = Other FCY liabilities − (Other FCY assets − Other FCY loans to resident households and nonfinancial corporations ) Total assets . Note that Foreign currency mismatch index = CHF mismatch index + Other FCY mismatch index. Furthermore, these indexes can be broken down into three components: net foreign currency liabilities, foreign currency loans to households, and foreign currency loans to nonfinancial corporations, all as a share of total assets. In other words, a sectoral breakdown of systemic risk is possible. Note also that the mismatch indexes are upper bounds for the systemic risk because they assume that none of the domestic households or nonfinancial corporations can service their foreign currency debt. More detailed information on each country regarding the characteristics of domestic borrowers can certainly refine these indexes. That is, instead of using the share of foreign currency loans by all domestic borrowers, the share of foreign currency loans by domestic borrowers with no foreign currency income can be entered into the equation to derive a more precise measure of systemic risk. However, these data are not available in the CHF Lending Monitor; therefore, in this article, I can calculate only the upper bounds for the systemic risk for each country. There are a few possible extensions and refinements to this measure of systemic risk. One possible extension would be to vary the “default rate” of the borrowers in relation to the magnitude of the depreciation and calculate an elasticity of systemic risk vis-à-vis the exchange rate. This extension would rely on the nonperforming loan ratios of individual countries in the past. Another possible extension would be to use this approach for individual banks rather than the entire banking system. National central banks and/or financial market regulators could have access to bank-level data in their own countries and easily calculate this systemic risk measure for individual banks. This extension could help policymakers identify banks that are most exposed to the common shock, as well as banks that contribute significantly to the systemic risk. Because such bank-level data are not available to the author and/or are confidential, these extensions can be analyzed only in possible future works. DATA All data are from the CHF Lending Monitor, which is an ongoing SNB project to understand the scope of CHF lending in Europe. Nineteen European central banks17 have been sharing their aggregate banking sector statistics with the SNB quarterly since 2009. The data are confidential and have not been published publicly. A quarterly report based on these data is circulated among the participating central banks. This report provides an overview of the volume, structure, and refinancing of CHF loans by banks domiciled in European countries to their non-banking clients. The CHF Lending Monitor data consist of aggregate banking sector statistics on both the asset side (such as loans and other assets) and the liability side (such as deposits, own securities 226 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin issues, and other liabilities). The sectoral breakdown of loan and deposit data is available for the following categories: resident banks,18 resident households, resident nonfinancial corporations, resident government, nonresident banks, and nonresident non-banks. Furthermore, a currency breakdown of loans, deposits, and own securities issued is available for the following categories: domestic currency, CHF, and other foreign currency. Such detailed aggregate-level information on the banking sector’s assets and liabilities makes it possible to accurately calculate the systemic risk arising from a foreign currency shock. Furthermore, the data are precise, as they rely on central banks’ reporting and do not involve an uninformed estimation.19 Central banks are asked to provide quarterly data for 92 variables in total; however, the central banks are free to decide to withhold certain variables because of confidentiality or unavailability of certain banking statistics. As a result of this data limitation, the foreign currency mismatch index can be calculated for 17 countries in the sample, and the CHF mismatch index can be calculated for 13 countries. FINDINGS Table 1 lists the measures for systemic risk across countries in Europe as of 2011:Q3. It also shows the currency breakdown of systemic risk between the CHF mismatch index and the other foreign currency (Other FCY) mismatch index.20 First, note that the foreign currency mismatch index varies significantly across European countries: from –0.8 percent in France to 44.3 percent in Latvia. It is generally higher in the non-euro area than in the euro area. For example, in all non-euro area countries except the Czech Republic and the United Kingdom, the foreign currency mismatch index is between 14 percent and 45 percent. In other words, net unhedged foreign currency liabilities of the banking sectors constitute between 14 percent and 45 percent of their total assets. Consequently, there is high systemic risk in the banking sectors of these countries. On the other hand, in the euro area countries the foreign currency mismatch index is lower by comparison. Net unhedged foreign currency liabilities always constitute less than 3.2 percent of total banking sector assets. Interestingly, the CHF mismatch index is fairly low in all European countries. For example, in Croatia the foreign currency mismatch index is 36.7 percent, but only 4.4 percent is due to loans denominated in CHF. The rest of the systemic risk is due to loans denominated in other foreign currencies (the mismatch index for other foreign currency in Croatia is 32.3 percent). Also, in other countries with a relatively high foreign currency mismatch index, the CHF mismatch index is relatively quite low. Thus, CHF loans to households and nonfinancial corporations are not the main driver behind the high systemic risk in the non-euro area countries. Consequently, loans denominated in other foreign currencies (probably denominated to a large extent in euros21) contribute far more to systemic risk in most non-euro area countries. Table 2 shows the individual components—that is, the sectoral breakdown—of the foreign currency mismatch index across European countries. These components are net foreign currency liabilities, foreign currency loans to domestic households, and foreign currency loans to domestic nonfinancial corporations (all as a share of total assets). The table reveals that most Federal Reserve Bank of St. Louis REVIEW May/June 2013 227 Yeşin Table 1 Systemic Risk Indexes in Europe Arising From Foreign Currency Loans (2011:Q3) FCY mismatch index (%) CHF mismatch index (%) Latvia 44.3 NA NA Croatia 36.7 4.4 32.3 Serbia 31.8 1.9 29.8 Romania 29.6 2.3 27.3 Bulgaria 26.6 0.0 26.6 Hungary 21.1 3.7 17.4 Poland Region/Country Other FCY mismatch index (%) Non-euro area 14.3 NA NA Czech Republic 3.5 –0.2 3.7 United Kingdom –0.1 0.0 0.0 Euro area Slovenia 3.1 3.1 0.1 Austria 2.2 0.8 1.3 Greece 1.6 –0.4 2.0 Slovakia 1.1 NA NA Germany 0.6 –0.1 0.7 Luxembourg 0.2 –0.7 1.0 Italy France 0.1 0.0 0.1 –0.8 NA NA NOTE: FCY mismatch index = CHF mismatch index + Other FCY mismatch index. A higher mismatch index indicates higher systemic risk. CHF, Swiss francs; FCY, foreign currency; NA, not available. SOURCE: Author’s calculations based on CHF Lending Monitor data. banking sectors in the non-euro area are possibly trying to hedge against the exchange rate risk by holding (significantly) more foreign currency assets than foreign currency liabilities. However, the volume of foreign currency loans to domestic households and nonfinancial corporations is fairly large relative to their total assets. Therefore, the resulting currency mismatch indexes are substantial in several non-euro area countries. The high systemic risk in the non-euro area countries is due to two factors: (i) The banking sectors in those countries have sizable foreign currency assets (and liabilities) relative to their total assets. (ii) Foreign currency loans to domestic households and nonfinancial corporations constitute a very large portion of the foreign currency assets. Similarly, Table 3 displays the sectoral composition of the CHF mismatch index in 2011:Q3 across Europe. As shown in Table 1, the CHF mismatch index, and hence the systemic risk resulting from CHF loans, is relatively low in all countries. However, in a few countries, the low mismatch index is due to banks holding significantly more CHF assets than CHF liabilities (particularly prevalent in Hungary, Croatia, and Austria). These net CHF assets compensate for 228 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin Table 2 Components of the Foreign Currency Mismatch Index (2011:Q3) Region/Country Net FCY liabilities/ Total assets (%) FCY loans to domestic households/ Total assets (%) FCY loans to domestic nonfinancial corporations/ Total assets (%) FCY mismatch index (%) Non-euro area Latvia –6.4 24.0 26.7 44.3 Croatia –6.8 23.2 20.3 36.7 Serbia –5.8 13.2 24.5 31.8 Romania –5.6 17.3 17.8 29.6 Bulgaria –12.9 8.9 30.6 26.6 Hungary –7.7 16.4 12.4 21.1 Poland –5.6 15.3 4.6 14.3 Czech Republic 0.3 0.0 3.2 3.5 United Kingdom –0.5 0.0 0.5 –0.1 Euro area Slovenia 0.0 2.4 0.7 3.1 Austria –2.9 3.8 1.2 2.2 Greece –3.9 1.6 4.0 1.6 Slovakia 0.6 0.5 0.0 1.1 Germany –0.2 0.1 0.6 0.6 0.1 0.0 0.2 0.2 Luxembourg Italy –0.3 0.1 0.4 0.1 France –1.4 0.2 0.4 –0.8 NOTE: The FCY mismatch index is the sum of the following three components: net FCY liabilities, FCY loans to domestic households, and FCY loans to domestic nonfinancial corporations (all as a share of total assets). FCY, foreign currency. SOURCE: Author’s calculations based on CHF Lending Monitor data. the sizable CHF loans to domestic borrowers in those countries. Consequently, the resulting CHF mismatch index is relatively low. Without such compensation for the risk, the CHF mismatch would have been much higher. Figures 3 and 4 show the time evolution of systemic risk in Europe. Note that both the foreign currency mismatch index and the CHF mismatch index have been fairly persistent over time for most countries and show no large fluctuations during the period for which data are available. In Figure 3, the foreign currency mismatch index reveals an upward trend in a few countries (e.g., Latvia, Bulgaria, Romania, and Poland) in the first half of the sample period. However, the index remains fairly flat or declines slowly in the second half of the sample period. In the remaining non-euro area countries, it is fairly stable during the period observed. Figure 4 similarly shows high persistence and low volatility of the foreign currency mismatch index in the euro area, albeit at a much lower level of systemic risk. Federal Reserve Bank of St. Louis REVIEW May/June 2013 229 Yeşin Table 3 Components of the CHF mismatch index (2011:Q3) Region/Country Net CHF liabilities/ Total assets (%) CHF loans to domestic households/ Total assets (%) CHF loans to domestic nonfinancial corporations/ Total assets (%) CHF mismatch index (%) Non-euro area –3.2 6.9 0.7 4.4 Hungary Croatia –12.8 14.4 2.1 3.7 Romania –1.3 3.5 0.2 2.3 Serbia –2.7 3.9 0.8 1.9 Bulgaria –0.2 0.2 0.0 0.0 United Kingdom 0.0 0.0 0.0 0.0 Czech Republic –0.2 0.0 0.0 –0.2 Euro area Slovenia 0.0 2.4 0.7 3.1 Austria –3.6 3.5 0.9 0.8 Italy –0.1 0.1 0.0 0.0 Germany –0.3 0.1 0.1 –0.1 Greece –2.3 1.5 0.4 –0.4 Luxembourg –0.7 0.0 0.0 –0.7 NOTE: The CHF mismatch index is the sum of the following three components: net CHF liabilities, CHF loans to domestic households, and CHF loans to domestic nonfinancial corporations. CHF, Swiss francs. SOURCE: Author’s calculations based on CHF Lending Monitor data. Finally, Figures 5 and 6 depict the time evolution of the CHF mismatch index across Europe. Again, persistency and low volatility of the systemic risk can be observed. A few countries, such as Croatia and Hungary, do show a slight upward trend in the CHF mismatch index over time; however, the CHF mismatch index remains at much lower levels than the foreign currency mismatch index during the sample period. CONCLUSION AND SOME POLICY IMPLICATIONS In this article, I have quantified the systemic risk in European countries resulting from foreign currency loans from a “common market shock” perspective. For this purpose, I used data from the CHF Lending Monitor and built on the method suggested by Ranciere, Tornell, and Vamvakidis (2010). These accurate, frequent, and detailed measurements of systemic risk in the form of more-precise currency mismatch indexes may give policymakers a finer gauge of the systemwide risks associated with foreign currency loans. I find that the systemic risk in the banking sector is very high in a majority of the non-euro area countries: Net unhedged foreign currency liabilities of the banking sectors correspond to 230 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin Figure 3 Foreign Currency Mismatch Index in the Non-Euro Area Percent 60 50 40 Latvia Croatia Serbia Romania Bulgaria Hungary Poland Czech Rep. U.K. 30 20 10 2011:Q3 2011:Q1 2010:Q3 2010:Q1 2009:Q3 2009:Q1 2008:Q3 2008:Q1 2007:Q3 –10 2007:Q1 0 SOURCE: Author’s calculations based on CHF Lending Monitor data. Figure 4 Foreign Currency Mismatch Index in the Euro Area Percent 6 5 4 3 Slovenia Austria Greece Slovakia Germany Luxembourg Italy France 2 1 0 –1 2011:Q3 2011:Q2 2011:Q1 2010:Q4 2010:Q3 2010:Q2 2010:Q1 2009:Q4 2009:Q3 2009:Q2 2009:Q1 2008:Q3 2008:Q4 2008:Q2 2008:Q1 2007:Q4 2007:Q3 2007:Q2 2007:Q1 –2 SOURCE: Author’s calculations based on CHF Lending Monitor data. Federal Reserve Bank of St. Louis REVIEW May/June 2013 231 Yeşin Figure 5 CHF Mismatch Index in the Euro Area Percent 5 4 3 2 1 Slovenia Austria Italy Germany 0 –1 Greece 2011:Q3 2011:Q1 2010:Q3 2010:Q1 2009:Q3 2009:Q1 2008:Q3 2008:Q1 2007:Q3 2007:Q1 –2 Luxembourg NOTE: CHF, Swiss francs. SOURCE: Author’s calculations based on CHF Lending Monitor data. Figure 6 CHF Mismatch Index in the Non-Euro Area Percent 5 4 3 2 Croatia Hungary 1 Romania Serbia Bulgaria U.K. Czech Rep. 0 2011:Q3 2011:Q1 2010:Q3 2010:Q1 2009:Q3 2009:Q1 2008:Q3 2008:Q1 2007:Q3 2007:Q1 –1 NOTE: CHF, Swiss francs. SOURCE: Author’s calculations based on CHF Lending Monitor data. 232 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin more than 20 percent of their total assets as of 2011:Q3. However, net unhedged CHF liabilities amount to less than 5 percent of their total assets. Thus, other foreign currency loans (probably to a large extent in euros) contribute to systemic risk far more than the CHF loans in the noneuro area. Switching the CHF loans into euro loans, as was recently suggested by the Hungarian Financial Supervisory Authority, would therefore not eliminate the systemic risk resulting from foreign currency loans. Furthermore, foreign currency mismatch indexes show persistence and low volatility during the sample period for which they can be calculated. That is, short-term policies would be unable to swiftly reduce that systemic risk. Encouragement of local currency borrowing, on the other hand, can be a long-run solution, as recently promoted by the EBRD initiative (see EBRD, 2010). A sectoral breakdown of systemic risk indicates that banks may be aware of the exchangerate-induced credit risk they assume when they lend to unhedged borrowers. This is evident from their balance sheets, which contain significantly more foreign currency assets than foreign currency liabilities. Last but not least, the role of the European Central Bank as the lender of last resort within the European Union (and not only within the euro area) needs to be discussed among the policymakers. In the non-euro area countries, where “euroization” took place unofficially within the financial system before the formal adoption of the euro, significant threats can materialize suddenly in the case of sharp exchange rate movements. A sizable depreciation of the non-euro area currencies vis-à-vis the euro can irreversibly damage the balance sheets of banks resident in the non-euro area. NOTES 1 See, for example, De Bandt and Hartmann (2000) and Georg (2011a,b) for an in-depth discussion of systemic risk. See Schwaab, Koopman, and Lucas (2011) for systemic risk diagnostics. 2 The non-banking sector consists of these subsectors: households, nonfinancial corporations, non-bank financial institutions, general government, and nonprofit institutions serving households (see the European System of Accounts, 1995; http://circa.europa.eu/irc/dsis/nfaccount/info/data/ESA95/en/esa95en.htm). 3 As of March 2013, there are only three currencies in the Exchange Rate Mechanism II (ERM II): the Danish krone, Latvian lat, and Lithuanian litas. According to the European Commission, “Participation in ERM II is voluntary, although, as one of the convergence criteria for entry to the euro area, a country must participate in the mechanism without severe tensions for at least two years before it can qualify to adopt the euro” (from “What is ERM II?”; http://ec.europa.eu/economy_finance/euro/adoption/erm2/index_en.htm). The Danish krone has been in the ERM II since 1999, but Denmark has no specific plans to join the euro area. Latvia is not expected to join the euro area before 2014. Its currency, the Latvian lat, has been in the ERM II since 2005. The Lithuanian litas has been in the ERM II since 2004; however, Lithuania is also not expected to join the euro area before 2014. 4 Previous literature documents that large corporations traditionally borrow in foreign currencies in many countries in East Asia and Latin America (e.g., see Allayannis, Brown, and Klapper, 2003, and Galindo, Panizza, and Schiantarelli, 2003). However, retail borrowing in foreign currencies in Europe is a more recent development. 5 Nonresidents are entities (such as households and nonfinancial corporations) residing in a country other than where their banks are domiciled. 6 See Galati, Heath, and McGuire (2007) on the recent buildup of carry trade positions resulting from low exchange rate volatility and persistent interest rate differentials. The literature on the uncovered interest parity is vast. Francis, Hasan, and Hunter (2002), for example, find persistent violation of the uncovered interest parity in emerging market economies. Federal Reserve Bank of St. Louis REVIEW May/June 2013 233 Yeşin 7 See Pann, Seliger, and Übeleis (2010) for a discussion on risks to the Austrian banking sector as a result of their foreign currency lending in Central, Eastern, and Southeastern Europe. 8 For example, this became an important issue during the financial crisis when the Hungarian forint and Polish złoty depreciated sharply and both the stock and the ratio of nonperforming loans rose in the portfolios of banks. 9 See the EBRD recommendations at http://www.esrb.europa.eu/pub/pdf/recommendations/2011/ESRB_2011_1.en. pdf?3304c9df8c9cecd453b4ecbab95d359d. 10 The EBRD region consists of the transition countries in Eastern Europe and the CIS countries (the former USSR). 11 During the financial crisis, when many Eastern European currencies depreciated significantly, many borrowers could no longer service their foreign currency debt. Nonperforming loan ratios increased during the crisis and policymakers were concerned about the systemic threat these foreign currency loans pose to the banking sector. 12 This article reports only systemic risk measures since 2007. 13 Brown, Ongena, and Yeşin (2011) find that foreign currency borrowing by small firms in transition countries is much more strongly related to firm-level foreign currency revenues than to country-level interest rate differentials. 14 Brown and de Haas (2012) also provide a similar finding. 15 On foreign currency loans in Europe, see also Fidrmuc, Hake, and Stix (2011) and Cuaresma, Fidrmuc, and Hake (2011). 16 Domestic = resident. A resident bank is a bank domiciled in that country. It can be a domestically owned bank or a subsidiary of a foreign bank. 17 The 19 contributing central banks are those in Austria, Bulgaria, Croatia, the Czech Republic, Denmark, Estonia, France, Germany, Greece, Hungary, Italy, Latvia, Luxembourg, Poland, Romania, Serbia, Slovakia, Slovenia, and the United Kingdom. 18 All central banks started reporting in 2009. At that time, some central banks also submitted their previous data going back to 2002:Q1. 19 Data received from the United Kingdom include some estimated sector split information. 20 The currency mismatch indexes cannot be calculated for Denmark as data on the banking sector’s liabilities are miss- ing. In this article, the currency mismatch indexes for Estonia are also not shown because Estonia adopted the euro during the sample period. Furthermore, CHF and other foreign currency mismatch indexes cannot be calculated for France, Latvia, Poland, and Slovakia because of missing data. 21 Further breakdown of currencies is not available in the CHF Lending Monitor; hence the euro mismatch index in the non-euro area countries cannot be calculated separately. REFERENCES Acharya, Viral L. and Yorulmazer, Tanju. “Information Contagion and Bank Herding.” Journal of Money, Credit, and Banking, February 2008, 40(1), pp. 215-31. Allayannis, George; Brown, Gregory W. and Klapper, Leora F. “Capital Structure and Financial Risk: Evidence from Foreign Debt Use in East Asia.” Journal of Finance, December 2003, 58(6), pp. 2667-709. Beer, Christian; Ongena, Steven and Peter, Marcel. “Borrowing in Foreign Currency: Austrian Households as Carry Traders.” Journal of Banking and Finance, September 2010, 34(9), pp. 2198-211. Brown, Martin and De Haas, Ralph. “Foreign Banks and Foreign Currency Lending in Emerging Europe.” Economic Policy, January 2012, 27(69), pp. 57-98. Brown, Martin; Kirschenmann, Karolin and Ongena, Steven. “Foreign Currency Loans—Demand or Supply Driven?” Working Paper No. 2011-2, Swiss National Bank, 2011; http://www.snb.ch/n/mmr/reference/working_paper_2011_02/source. Brown, Martin; Ongena, Steven and Yeşin, Pınar. “Foreign Currency Borrowing by Small Firms in the Transition Economies.” Journal of Financial Intermediation, 2011, 20(3), pp. 285-302. 234 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Yeşin Brown, Martin; Ongena, Steven and Yeşin, Pınar. “Information Asymmetry and Foreign Currency Borrowing by Small Firms.” Comparative Economic Studies. Published electronically April 25, 2013; http://www.palgrave-journals.com/ces/journal/vaop/ncurrent/pdf/ces20139a.pdf or doi:10.1057/ces.2013.9. Cerutti, Eugenio; Claessens, Stijn and McGuire, Patrick. “Systemic Risks in Global Banking: What Available Data Can Tell Us and What More Data Are Needed?” IMF Working Paper No. 11/222, International Monetary Fund, September 2011; http://www.imf.org/external/pubs/ft/wp/2011/wp11222.pdf. Cuaresma, Jesús Crespo; Fidrmuc, Jarko and Hake, Mariya. “Determinants of Foreign Currency Loans in CESEE Countries: A Meta-Analysis.” Focus on European Economic Integration, 2011, 4, pp. 69-87. De Bandt, Olivier and Hartmann, Philipp. “Systemic Risk: A Survey.” ECB Working Paper No. 35, European Central Bank, November 2000; http://www.ecb.int/pub/pdf/scpwps/ecbwp035.pdf. Degryse, Hans; Havrylchyk, Olena; Jurzyk, Emilia and Kozak, Sylwester. “Foreign Bank Entry, Credit Allocation and Lending Rates in Emerging Markets: Empirical Evidence from Poland.” Journal of Banking and Finance, November 2012, 36(11), pp. 2949-59. European Bank for Reconstruction and Development. “Local Currency Initiative: EBRD Launches Local Currency and Local Capital Markets Initiative.” May 15, 2010; http://www.ebrd.com/pages/news/press/2010/100515a.shtml. Fidrmuc, Jarko; Hake, Mariya and Stix, Helmut. “Households’ Foreign Currency Borrowing in Central and Eastern Europe.” Working Paper No. 171, Austrian Central Bank, July 2011; http://www.oenb.at/de/img/wp171_tcm14-239053.pdf. Francis, Bill B.; Hasan, Iftekhar and Hunter, Delroy M. “Emerging Market Liberalization and the Impact on Uncovered Interest Rate Parity.” Journal of International Money and Finance, November 2002, 21(6), pp. 931-56. Galati, Gabriele; Heath, Alexandra and McGuire, Patrick. “Evidence of Carry Trade Activity.” Bank for International Settlements Quarterly Review, September 2007, 33, pp. 27-41; http://www.bis.org/publ/qtrpdf/r_qt0709e.pdf. Galindo, Arturo J.; Panizza, Ugo and Schiantarelli, Fabio. “Debt Composition and Balance Sheet Effects of Currency Depreciation: A Summary of the Micro Evidence.” Emerging Markets Review, December 2003, 4(4), pp. 330-39. Georg, Co-Pierre. “Basel III and Systemic Risk Regulation—What Way Forward?” Working Papers on Global Financial Markets No. 17-2011, Friedrich-Schiller-University Jena, January 2011a; http://pubdb.wiwi.uni-jena.de/pdf/wp_hlj17-2011.pdf. Georg, Co-Pierre. “The Effect of the Interbank Network Structure on Contagion and Common Shocks.” Discussion Paper Series 2: Banking and Financial Studies, No. 12/2011, Deutsche Bundesbank, 2011b; http://www.bundesbank.de/Redaktion/EN/Downloads/Publications/Discussion_Paper_2/2011/2011_10_25_dkp_1 2.pdf?__blob=publicationFile. Pann, Johannes; Seliger, Reinhardt and Übeleis, Julia. “Foreign Currency Lending in Central, Eastern and Southeastern Europe: The Case of Austrian Banks.” Financial Stability Report, December 2010, 20, pp. 56-76. Ranciere, Romain; Tornell, Aaron and Vamvakidis, Athanasios. “Currency Mismatch, Systemic Risk and Growth in Emerging Europe.” Economic Policy, October 2010, 25(64), pp. 597-658. Reuters. “UPDATE 1—Hungary Watchdog Proposes Switch to EUR-Based Loans.” January 11, 2011; http://www.reuters.com/article/2011/01/11/hungary-loans-idUSLDE70A17S20110111. Schwaab, Bernd; Koopman, Siem Jan and Lucas, André. “Systemic Risk Diagnostics: Coincident Indicators and Early Warning Signals.” Working Paper Series No. 1327, European Central Bank, October 2011; http://www.ecb.int/pub/pdf/scpwps/ecbwp1327.pdf. Federal Reserve Bank of St. Louis REVIEW May/June 2013 235 236 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Labor Mismatch in the Great Recession: A Review of Indexes Using Recent U.S. Data Maria E. Canon, Mingyu Chen, and Elise A. Marifian Labor mismatch, also known as structural imbalance, can be defined as a poor match between the characteristics of unemployed workers and those required for vacant jobs. In the wake of the jobless recovery from the Great Recession, economists have sought to explain the coexistence of a high unemployment rate and increasing job openings as a mismatch phenomenon. This article reviews five studies that have contributed to the development of mismatch indexes and computes the corresponding indexes over the period May 2005–May 2012 using job vacancy data from the Conference Board Help Wanted OnLine® (HWOL) Data Series. For most of the indexes, mismatch increased during the Great Recession, although the indexes exhibit a range of behaviors. According to an index developed in Jackman and Roper (1987), mismatch can account for at most 2.72 percentage points of the 5.30-percentage-point increase in the unemployment rate from the beginning of the recession to the unemployment rate peak. (JEL E24, J01, J23, J63, J64) Federal Reserve Bank of St. Louis Review, May/June 2013, 95(3), pp. 237-71. n the years following the Great Recession, high unemployment rates persisted across the United States despite the steady increase in job openings.1 This unexplained dynamic has led many to believe that the U.S. labor market’s slow and jobless recovery could be explained as a mismatch phenomenon.2 For instance, supporters of the sectoral mismatch hypothesis have pointed to changes in the employment and vacancy breakdown by sector. From December 2007 to February 2011, more than 50 percent of job losses occurred in the manufacturing and construction sectors, while over 90 percent of new positions opened in other sectors, suggesting that sectoral mismatch may have increased. Proponents of geographic mismatch—characterized by job vacancies in places different from those where people are looking for work—believe that the slow labor market recovery is rooted in the real estate bust and the subsequent extreme declines in housing prices, which may have reduced the mobility of homeowners. For example, Ferreira, Gyourko, and Joseph (2010) I Maria E. Canon is an economist, Elise A. Marifian is a research associate, and Mingyu Chen is a former research associate at the Federal Reserve Bank of St. Louis. The authors thank Jeanne Shu from the Conference Board for her thorough assistance with the Conference Board Help Wanted OnLine® Data Series. The authors also thank two referees and Andrew B. Langan for their valuable comments. © 2013, The Federal Reserve Bank of St. Louis. The views expressed in this article are those of the author(s) and do not necessarily reflect the views of the Federal Reserve System, the Board of Governors, or the regional Federal Reserve Banks. Articles may be reprinted, reproduced, published, distributed, displayed, and transmitted in their entirety if copyright notice, author name(s), and full citation are included. Abstracts, synopses, and other derivative works may be made only with prior written permission of the Federal Reserve Bank of St. Louis. Federal Reserve Bank of St. Louis REVIEW May/June 2013 237 Canon, Chen, Marifian conclude that unemployed workers who owe more than their homes are worth would be less likely to apply for and accept positions that require them to sell their homes.3 Yet Şahin et al. (2012) argue that geographic mismatch has played an insignificant role. Whether mismatch has caused a significant increase in the unemployment rate is debated among economists and policymakers. In this article, we present a comprehensive review of five studies that have contributed to the historical development of mismatch indexes. For each index, we describe the theoretical framework, interpretation, and links to other indexes. We then compute the described mismatch indexes using (i) Conference Board Help Wanted OnLine® (HWOL) data, which provide information on online job advertisements nationwide by detailed occupations and industries, and (ii) Current Population Survey (CPS) data on unemployment and employment. Our analysis of the indexes includes comparisons within a given index’s different types (here industrial and occupational mismatch) and disaggregations, as well as comparisons among the different indexes. We also separate the analysis into three time periods: pre-recession (May 2005–November 2007), Great Recession (December 2007–June 2009), and post-recession (July 2009–May 2012). We find that the level of mismatch can vary significantly among indexes and within an index’s different types and disaggregations. The choice of an index ultimately depends on the question that one seeks to answer, as well as the extent to which one desires a practical interpretation and application of the index. With these points in mind, we believe that one index in particular, developed by Jackman and Roper (JR, 1987) and recently extended by Şahin et al. (2012), stands out as the most practical and intuitive measure of mismatch. What this index suggests is that industrial and occupational mismatch can account for at most 2.72 percentage points of the 5.30-percentage-point increase in the unemployment rate from the beginning of the recession to the unemployment rate peak. MEASURES OF MISMATCH Shortly after the development of the Beveridge curve,4 economists became interested in the role of mismatch in the labor market. With persistently high unemployment in Europe during in the 1970s and 1980s, researchers began examining nontemporary imbalances between supply and demand for labor across industries, skill groups, regions, and age groups.5 Over the past three decades, economists have developed various mismatch indexes to measure the level of sectoral imbalance in labor markets and the contribution of this imbalance to unemployment. In this section, we chronologically review eight mismatch indexes, providing the historical context that motivated the development of the index and summarizing the derivation. Seeking to explain the high unemployment rates observed in the 1970s, Lilien (1982) points to volatility in employment demand over the period. He first notes that the composition of industry employment shares changed dramatically from 1969 to 1980: Manufacturing’s share shrank 22.8 percent while the shares of retail trade; fire, insurance and real estate; and service industries grew a combined 47.6 percent. He then observes that (i) employment shocks to durable manufacturing employment coincided with the decade’s so-called cyclical increases in unemployment and (ii) the durable manufacturing employment share did not increase significantly 238 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian even when unemployment abated.6 Given these two observations in the data, Lilien challenges what was then a common belief: That cyclical unemployment represents a deviation from a stable natural rate. He argues instead that most postwar-era unemployment is more accurately described as stemming from workers’ attachment to sectors. He argues that this attachment— perhaps the result of wage premiums and industry-specific skills —reduces workers’ willingness to look for the available positions in other sectors. As a result, when exogenous shocks create new patterns of sectoral demand, labor markets are slow to adjust and unemployment tends to be high during the transition. To support his argument that labor’s slow adjustment to sectoral employment demand shifts is the underlying source of the decade’s high unemployment, Lilien (1982) develops an index to estimate the “dispersion of employment demand conditions throughout the labor market” or the “dispersion in hiring conditions” (p. 780).7 Specifically, Lilien’s dispersion index measures the sectoral employment share-weighted percent log deviations of sectoral employment from total employment, taking the following form: 1/ 2 n x 2 I Lilien = σˆ t = ∑ it ( ∆ log xit − ∆ log Xt ) , i=1 Xt where xit is the number of people employed in sector i at time t and Xt is the total employment for all sectors. Of note, this index has been critiqued for its correlation with both sectoral shifts and aggregate demand fluctuations (see Abraham and Katz, 1986; Neelin, 1987; and Blanchard and Diamond, 1989). Jackman and Roper (1987) produce the first study that formalizes the notion of mismatch, or “structural imbalance,”8 which they define as a situation in which the characteristics of unemployed workers, such as skill, experience, or location, differ from those required for the vacant jobs. Structural imbalance between the pattern of labor supplied and demanded yields excess unemployment, which they call “structural unemployment.” More precisely, they specify that structural unemployment exists when, given the configuration of vacancies, it would be possible to reduce unemployment (increase the job hiring rate) by moving an unemployed worker from one sector to another. Jackman and Roper (1987) use a static matching model with an unemployment/vacancy framework to develop a measure of structural unemployment, from which they derive various indexes of structural imbalance.9 In the context of their model, they define an absence of structural unemployment as the configuration of the existing stock of unemployment that maximizes aggregate hires, given the distribution of vacancies. Their maximization of aggregate hires implies that the unemployment-to-vacancies ratio is identical across sectors when the labor market is at structural balance. Structural unemployment (SU) is denoted by 1 SU = U ∑| ûi − v̂i | , 2 i where U is the total stock of unemployed workers and ûi and v̂i are sector i’s share of the total stock of unemployed workers and vacancies, respectively. From there, Jackman and Roper’s first two mismatch indexes can be derived by normalizing SU on total unemployment (JR index 1, Federal Reserve Bank of St. Louis REVIEW May/June 2013 239 Canon, Chen, Marifian JR1) or on the total labor force (JR index 2, JR2). JR1 measures the proportion of the unemployed in the wrong sector10 and is denoted by 1 ∑| ûi − v̂i | . 2 i I JR1 = SU U = JR2 measures the proportion of the labor force in the wrong sector. It is denoted by I JR 2 = SU L = 1U ∑| ûi − v̂i | , 2L i where L is the size of the labor force. Jackman and Roper (1987) note that one disadvantage of these two indexes is that they do not indicate how much of the current unemployment is due to mismatch (or, alternatively, how much unemployment would fall if structural balance were achieved). To answer this question, they develop a third index (JR3), assuming that the hiring function is Cobb-Douglas.11 They use an elasticity of substitution term a = 0.5, which is within the 0.5 to 0.7 range of estimates used in the literature (Petrongolo and Pissarides, 2001). For this index, they define the levels of unemployment and vacancies at structural balance (U * and V*) as the levels needed (i) to produce the current level of aggregate hires (which is given) and (ii) to equalize the unemployment and vacancy shares across sectors (ûi = v̂i , for every i).12 Their third index is denoted by I JR 3 = 1 − ∑ (ûi v̂ i ) 1/ 2 i and measures the proportion of the observed unemployment attributable to structural imbalance. IJR1, IJR2, and IJR3 are invariant to aggregate demand shocks that keep ût and v̂t constant. Under a non-accelerating inflation rate of unemployment environment, Jackman, Layard, and Savouri (JLS, 2008) develop a framework in which the optimal unemployment rate is obtained when unemployment rates are equalized across sectors. In the current context, the authors define mismatch unemployment as the distance between the observed unemployment rate and the equilibrium rate. The derivation of their mismatch index relies on the critical assumptions for Cobb-Douglas production functions and double logarithmic wage functions of the form logwi = bi – g logui. The index, which measures the proportional excess of actual unemployment over minimum unemployment, is defined as u 1 I JLS = var i = log u − log umin , u 2 where ui is the unemployment rate in sector i and u is the mean of the sector-specific unemployment rates (see Jackman, Layard, and Savouri, 2008). The index is a measure of the dispersion of relative unemployment rates (varui/u) across sectors. Motivated by the high levels of European unemployment in the 1980s and the shift in the U.K. Beveridge curve from 1963 to 1984, Evans (1993) examines the extent to which mismatch— which he calls “sectoral imbalance”—may have contributed to increased unemployment rates in the United Kingdom. Based on a framework with temporarily inflexible wages, Evans constructs a measure of sectoral imbalance that calculates the average deviation of the percentage difference 240 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian between the supply and demand for labor across sectors (p. 442). This measure is approximated by I Evans = 1 ∑li | (ui − vi ) − (u − v ) | , 2 i where li is the sector i’s share of labor force; ui and vi are the unemployment and vacancy rates in sector i, respectively; and u and v are the aggregate rates. Evans notes that his index measures imbalance as a share of the sectoral labor forces. Accordingly, he states that unlike many other indexes, his index has the advantage of being invariant to a neutral change in aggregate demand for labor (i.e., the demand for labor in each sector changes by the same percentage).13 In addition, he argues that his index, unlike Lilien’s (1982), measures the overall state of labor market imbalance and the contribution of the shocks to the state of imbalance. Evans’s (1993) index is similar to the JR2 index (Jackman and Roper, 1987) in that both attempt to measure sectoral imbalance as a proportion of the labor force. Yet the two diverge in their definition of structural balance: For their second index, Jackman and Roper define structural balance in terms of equalizing the ratio of unemployment to vacancies across sectors, while Evans defines sectoral balance in terms of equalizing the absolute differences between sectoral and aggregate unemployment rate deviations from the sectoral and aggregate vacancy rates. Jackman and Roper state that, given their hiring function assumptions, the definition of structural balance suggested by Evans’s index14 does not maximize the outflow from unemployment (and thus, aggregate hires). They also suggest that the definition in terms of ratios (i.e., the number of unemployed per vacancy) is more natural than one in terms of absolute differences between the unemployment rate and the vacancy rate within a sector. Most recently, Şahin et al. (SSTV, 2012) approach the mismatch topic motivated by the persistently high unemployment in the United States. Noting that the flow into unemployment has decreased to pre-recession levels, they argue that any theory to explain the recent dynamics in the labor market must explain the long-lasting decline in the rate at which workers find jobs. They hypothesize that sectoral mismatch between available jobs and unemployed workers could explain this lower aggregate job-finding rate. Building on the work of Jackman and Roper (1987), Şahin et al. (2012) develop a dynamic stochastic environment that allows heterogeneity in sectoral matching and productivity efficiencies in the generalized model. In Şahin et al.’s (2012) framework, the optimal allocation of unemployed workers is one that would be chosen by a planner who can move unemployed workers across sectors freely, subject only to the restrictions imposed by matching frictions within each labor market (sector).15 In their benchmark environment, the solution to the planner’s problem states that the planner allocates more unemployed workers to search in the markets with higher vacancies and matching efficiencies. The optimal allocation of unemployed workers across sectors is achieved when the matching efficiency-weighted vacancy-to-unemployment ratios are equalized across sectors. From the planner’s allocation rule, they derive their benchmark mismatch index, with the additional assumption that the matching function is Cobb-Douglas. The benchmark index (Mft ) is denoted by Federal Reserve Bank of St. Louis REVIEW May/June 2013 241 Canon, Chen, Marifian α ISSTV 1φ 1−α I φ v u ht = M φt = 1 − * = 1 − ∑ it it it ht i =1 φ t vt ut , 16 where vit and uit are the vacancies and unemployment in sector i at time t; vt and ut are the aggregate levels of vacancies and unemployment at time t; fit is the matching efficiency component for sector i at time t; f t is the aggregate matching efficiency component; and a is the elasticity of substitution.17 ISSTV1f measures the fraction of hires in period t that are lost as a result of misalloh cation, 1 − t∗ , where h*t is the planner’s hires and ht is the observed level of aggregate hires. ht In the absence of heterogeneities other than vacancies (i.e., matching efficiency, productivity, and job destruction rates), Şahin et al. (2012) refer to the index as Mt . Specifically, α 1−α v u I SSTV 1 = M t = 1 − ∑ it it i =1 vt ut I . Şahin et al.’s first index (ISSTV1) is mathematically equivalent to Jackman and Roper’s third index (IJR3 ) when the elasticity term a = 0.5. It is important to note, however, that IJR3 and ISSTV1 are interpreted differently. When interpreting IJR3 , Jackman and Roper (1987) define aggregate unemployment without mismatch U * as the level of unemployment needed to equalize the unemployment and vacancy shares in each sector, assuming a given aggregate level of hires and a given aggregate ratio of unemployment to vacancies. It follows that IJR3 is given by the amount of unemployment (in the aggregate) that exceeds the level that would exist at structural balance (U – U *) as a share of actual aggregate unemployment U (see Jackman and Roper, p. 14). Alternatively, ISSTV1 is given by the difference between the number of hires that would result from a planner who could freely move workers across sectors (h*t) and the observed level of hires ht , normalized by h*t . Şahin et al. (2012) mention three useful properties of their index: (i) It allows easy interpretation from 0 to 1, where 0 indicates no mismatch and 1 indicates maximum mismatch. (ii) It is invariant to aggregate shocks that raise or lower the aggregate number of vacancies and unemployed while leaving these shares unchanged across markets. And (iii) it is increasing in the level of disaggregation, which implies that statements regarding the role of mismatch should be qualified with respect to the degree of sectoral disaggregation that is used. Although we do not provide a full review or computation of it in this article, it is worth noting that Şahin et al. (2012) also construct a second index (ISSTV2) in which they abandon the assumption that productivities and job destruction rates are identical across sectors. This change yields their generalized model, where sector heterogeneity exists in the matching efficiency, vacancy, productivity, and destruction rate components. This second index measures the fraction of hires lost because of job-seeker misallocation at time t and is denoted by α 1−α φ v u I SSTV 2 = M xt = 1 − ∑ ix it it i =1 φt vt ut I 242 May/June 2013 , Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian 1−α α I 1 v x α v – where φt = ∑ φit it it with x–t = ∑ xitα it 18 and f tx is an aggregator of i =1 i =1 v t x t vt the vacancy share-weighted market-level efficiencies. Lastly, Şahin et al. (2012) attempt to quantify the contribution of mismatch unemployment to the aggregate unemployment rate. They construct a counterfactual unemployment rate u* that would exist if there were no mismatch: x I ut*+1 = st + (1 − st − ft* ) ut* , where st is the separation rate and ft* is the job-finding rate in the absence of mismatch: α ut ft ft = , (1 − I SSTV 2 ) ut* * where ft = ht /ut is the observed job-finding rate. DATA We need detailed information on employment, unemployment, and vacancies to compute the mismatch indexes. For employment and unemployment data, we use the Bureau of Labor Statistics’ (BLS) Current Population Survey (CPS) estimates for national- and state-level employment and unemployment counts.19 Our computations of the mismatch indexes use the counts of unemployed and employed workers by industries or occupations, and we seasonally adjust the data.20 For vacancy counts, we use the Help Wanted OnLine Data Series (HWOL) published by the Conference Board, which we also seasonally adjust. The HWOL provides monthly measures of new and total21 online job advertisements at various geographic levels for the entire United States. It also provides occupational and industrial classification for the job posts. Current Population Survey Data The CPS is a monthly survey of approximately 60,000 households that began in 1940. In addition to other data, the CPS provides detailed U.S. labor market statistics on employment, unemployment, earnings, labor force participation rates, and those not in the labor force. The CPS estimates are available by demographic characteristics, industry, and occupation. The monthly survey is conducted during the calendar week including the 19th day of the month, and questions are asked regarding the respondent’s labor market activity during the previous calendar week (which included the 12th day of the month)—the reference week. As noted, for the mismatch indexes we use the count of unemployed workers by industries or occupations. Using these data requires the assumption that unemployed workers are searching for jobs in the same sector as their previous job. Şahin et al. (2012) attempt to improve the accuracy of the count by following a method used by Hobijn (2012) that merges monthly CPS data to identify the sectors of the new jobs found by the previously unemployed workers. Şahin et al. Federal Reserve Bank of St. Louis REVIEW May/June 2013 243 Canon, Chen, Marifian Figure 1 HWOL Job Vacancies Data Comparison: U.S. Total Versus National Industry and Occupation Aggregations (May 2005–May 2012) Vacancies (in Millions) 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 May-05 Jan-06 Sep-06 May-07 Jan-08 Sep-08 May-09 Jan-10 Sep-10 May-11 Jan-12 U.S. Total SOC Occupation Aggregation NAICS Industry Aggregation NOTE: The shaded area indicates the most recent recession as determined by the National Bureau of Economic Research. SOC, Standard Occupational Classification system. (2012) show that this method is sufficient to infer the sectors in which unemployed workers were searching. Nevertheless, an unemployed worker could have past work experience in various sectors (i.e., he or she holds many sector-specific skills) and thus could look for jobs in many sectors, which would not be reflected in the CPS. Allocating this group of unemployed workers to the “wrong” sectors could overestimate the sector-specific mismatch indexes. HWOL Vacancy Data22 The HWOL is designed to provide monthly measures of labor demand as determined by advertised vacancies and is targeted to cover the entire universe of online job postings for the United States. Each month’s HWOL data series provides detailed information for 3 million to 4 million unique active vacancy ads. Data in HWOL are collected by WANTED Technologies Corporation from more than 16,000 online job websites, including corporate job boards and smaller job boards that serve niche markets.23 The historical data begin in May 2005 and the Conference Board publishes annual revisions with the January data. HWOL data are the sum of postings from mid-month to mid-month and are aligned with the CPS unemployment job search time period for straightforward comparison. Data are released around the first of the month following the completed mid-month to mid-month period. For example, data published on December 1 contain job ads from October 15 to November 14. 244 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian Vacancy counts are obtained through real-time database queries. Queries can be made at geographic, occupational, and industry levels (Figure 1 shows national aggregations). Although this section includes only what is most important for the indexes, more detailed information about the HWOL is included in Appendix B. Unduplication. The HWOL attempts to reflect unduplicated ads. As the geographic level increases (e.g., from county to metropolitan area), an ad may appear more than once because neighboring counties may each have ads for the same job. Accordingly, a process is used for each query to eliminate duplicated ads across job sites using information such as a company’s name, job title, and location. Approximately two-thirds of the ads are removed as duplicates. Reliability. Because the HWOL is effectively a population count,24 it is not subject to the standard sampling error and nonresponse errors embodied in the majority of statistical surveys. The HWOL data are subject to nonsampling error from (i) undercoverage resulting from missing a significant job board and smaller, local job boards and (ii) overcoverage resulting from the failure to eliminate all duplicate ads. Other errors could include ads incorrectly coded at the occupational, industrial, or geographic level. Approximately 40 percent of ads do not provide information about the employer. Accordingly, those ads do not have an industry recorded and thus could affect the reliability of calculations that use industry data. RESULTS In this section, we use the above-described data to compute six of the mismatch indexes described earlier: ILilien, IJR1, IJR2, IJR3(ISSTV1), IJLS, and IEvans.25 We calculate the indexes between May 2005 and May 2012 for two types of mismatch—industrial and occupational—using two levels of disaggregation for each (Table 1). We first review the mismatch trends by index to understand how changes in the indexes compare with one another over the course of the sample. We then analyze the data for three separate periods to determine whether mismatch has changed during the Great Recession and, if so, its implication for labor markets. The periods for our study are as follows: pre-recession (May 2005–November 2007), Great Recession (December 2007– June 2009), and post-recession (July 2009–May 2012). Lastly, we calculate the counterfactual unemployment rate described previously to infer the contribution of mismatch to the rise in the unemployment rate around the Great Recession.26 For greater tractability, we focus our analysis on the 19-industry and 22-occupation breakdowns. We follow Şahin et al.’s (2012) approach and apply a Hodrick-Prescott filter to the computed indexes to remove high-frequency movements.27 Furthermore, because approximately 40 percent of ads do not provide information on the employer (see the previous section), only around 45 percent of the unduplicated ads are assigned industry codes (see Figure 1). To address this issue, we assume that the proportion of ads that cannot be assigned a North American Industry Classification System (NAICS) code is homogeneous across all industry categories used in this article. While this assumption may not be very realistic, the results for our industry mismatch indexes show trends similar to the results of Şahin et al. (2012), who use Job Openings and Labor Turnover Survey (JOLTS) data (see Appendix B). Given this assumption, estimations of the IJR1, Federal Reserve Bank of St. Louis REVIEW May/June 2013 245 Canon, Chen, Marifian Table 1 Industrial and Occupational Mismatch at Two Levels of Disaggregation Industries (NAICS code) Occupations (SOC code) General level of disaggregation 12 Industries General level of disaggregation 10 Occupations Natural resources and mining (11, 21) Construction (23) Manufacturing (31-33) Wholesale and retail trade (42, 44-45) Transportation and utilities (48-49, 22) Information (51) Financial activities (52, 53) Professional and business services (54, 55, 56) Education and health services (61, 62) Leisure and hospitality (71, 72) Other services (81) Public administration (92) Detailed level of disaggregation 19 Industries Agriculture, forestry, fishing and hunting (11) Mining (21) Utilities (22) Construction (23) Manufacturing (31-33) Wholesale trade (42) Retail trade (44-45) Transportation and warehousing (48-49) Information (51) Finance and insurance (52) Real estate and rental and leasing (53) Professional, scientific, and technical services (54) Business services (55, 56) Education services (61) Health care and social assistance (61) Arts, entertainment, and recreation (71) Accommodation and food services (72) Other services (81) Public administration (92) 246 May/June 2013 Management, business, and financial (11, 13) Professional and related (15, …, 29) Service (31, …, 39) Sales and related (41) Office and administrative support (43) Farming, fishing, and forestry (45) Construction and extraction (47) Installation, maintenance, and repair (49) Production (51) Transportation and material moving (53) Detailed level of disaggregation 22 Occupations Management (11) Business and financial operations (13) Computer and mathematical (12) Architecture and engineering (17) Life, physical, and social science (19) Community and social services (21) Legal (23) Education, training, and library (25) Arts, design, entertainment, sports, and media (27) Health care practitioners and technical (29) Health care support (31) Protective service (33) Food preparation and serving related (35) Building and grounds cleaning and maintenance (37) Personal care and service (39) Sales and related (41) Office and administrative support (43) Farming, fishing, and forestry (45) Construction and extraction (47) Installation, maintenance, and repair (49) Production (51) Transportation and material moving (53) Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian IJR2, IEvans, and IJR3(ISSTV1) indexes are not affected since sector i’s share of total vacancies is unchanged from the missing ads. Mismatch Trends by Index Figure 2 plots each index’s change in industrial mismatch (19 industries) over the entire sample relative to its value in May 2005. The behaviors of the six indexes vary considerably over the course of the 7-year sample. Among the 19-industry indexes, the IJR1 and IJR3(ISSTV1) indexes share a similar pattern: Both increase over the course of the first two periods and then fall from their May 2005 values, yielding declines of approximately 27 percent and 35 percent, respectively. The IJR2 and IEvans indexes also share a similar long-term pattern: Although their paths generally differ in the pre-recession period, they increase through the recession and into the post-recession period, turning around in March 2010 and declining for the remainder of the sample; yet, at May 2012 their values are still considerably higher than their May 2005 values. The remaining indexes—ILilien and IJLS—exhibit trends different from all the other indexes. Figure 3, which plots occupational mismatch changes in the six indexes (22 occupations) over the sample, shows that most of the occupation mismatch indexes follow trends similar to those of their respective industry indexes. The main difference between the industry and occupation calculations in a given index is generally a function of scale: In the IEvans , IJLS , and ILilien cases, the occupation indexes appear to be somewhat muted versions of the industry index trends; the reverse is true for the others. Perhaps the most noticeable feature from Figures 2 and 3 is that the IJR2 and IEvans indexes exhibit changes of magnitude considerably greater than the other indexes. If we consider the long-term changes in the IJR2 index in context with data on unemployment and the labor force, the index would be expected to increase rapidly, while IJR1 and IJR3(ISSTV1) would not. The reason for this difference rests in the underlying calculations of the Jackman and Roper (1987) indexes. Recall that IJR1 is derived by normalizing structural unemployment by the total stock of unem1 U ∑| ûi − v̂i | 2 i ployment, , whereas IJR2 is derived by normalizing structural unemployment by U 1 U ∑| ûi − v̂i | 2 i the labor force, . The second normalization results in IJR2 being equal to IJR1 L scaled by U/L, the total stock of unemployed over the labor force. Therefore, the key to understanding the behavior of IJR2 is the U/L ratio at each point of the long-term change calculation; in May 2005, U/L = 0.051, with U = 7,651,000 and L = 149,261,000, compared with May 2012, when U/L = 0.082, with U = 12,695,000 and L = 154,998,000. With these scalars in mind, the calculation for the percentage change over the sample (May 2005–May 2012) is altered from the basic IJR1 calculation, Federal Reserve Bank of St. Louis REVIEW May/June 2013 247 Canon, Chen, Marifian Figure 2 Normalized 19-Industry Indexes: Changes Over Sample Normalized, May 2005 = 100 225 200 JR1 JR3 JR2 Evans JLS Lilien 175 150 125 100 75 50 May-05 Jan-06 Sep-06 May-07 Jan-08 Sep-08 May-09 Jan-10 Sep-10 May-11 Jan-12 Figure 3 Normalized 22-Industry Indexes: Changes Over Sample Normalized, May 2005 = 100 225 200 JR1 JR3 JR2 Evans JLS Lilien 175 150 125 100 75 50 May-05 Jan-06 Sep-06 May-07 Jan-08 Sep-08 May-09 Jan-10 Sep-10 May-11 Jan-12 248 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian %∆I JR1 = −12 − 05 I 05 − I 05 JR 1 JR 1 −05 I 05 JR 1 , to ( ) ( ) I 05−12 * 0.082 − I 05−05 * 0.051 JR1 JR 1 %∆I JR 2 ≅ . 05− 05 I JR1 * 0.051 ( ) Adding in the scalars (what the IJR2 index adds to IJR1) adjusts the behavior of the IJR1 index such that the percentage change for IJR2 over the sample switches from negative to positive; the May 2012 index value is multiplied by 0.082 and the May 2005 index value is multiplied by 0.051. This analysis explains why IJR2 differs from its related IJR1 and IJR3(ISSTV1) indexes. Comparing underlying formulas can also help shed light on why IEvans increases more than IJR1. The IJR1 index is calculated with just one deviation—namely, sectoral unemployment share deviations from sectoral vacancy shares. Alternatively, IEvans is calculated with two deviations: the deviation of sectoral differences from aggregate differences between the unemployment and vacancy rates. Therefore, IEvans adds an additional element of variation. If sectoral deviations are significantly different from the aggregate deviation, IEvans will capture additional volatility that IJR1 does not. In other words, the additional deviation can make IEvans more sensitive to unemployment or vacancy rate outliers, skewing the index relative to an index that is calculated with only one deviation, such as IJR1. We now review the behaviors of each index over three periods—pre-recession, Great Recession, and post-recession—to understand each index’s behavior in the context of its theoretical interpretation (i.e., its implication for labor markets). Pre-Recession Mismatch Over the pre-recession period (May 2005–November 2007), the behaviors of the indexes, both among the different indexes and within a particular index (industrial versus occupational mismatch), vary significantly. The most important point from the pre-recession period is that, despite the diverging trends, most of the industry and occupation indexes share a common element. Namely, regardless of the overall behavior over the period, most indexes began to increase in the months preceding the recession. This point is important to remember when considering increases in mismatch during the Great Recession. An increase in mismatch during the Great Recession with no preceding mismatch increase could suggest a cyclical component to the indexes. On the other hand, an increase in mismatch preceded by other such increases could be more indicative of other labor market shifts underway even before the Great Recession began. The IJR1 and IJR3(ISSTV1) industry mismatch indexes exhibit similar behaviors during the pre-recession period. Figure C128 shows that the IJR1 industry index, which is essentially flat in the 19-industry breakdown, declines until January 2007, when it turns around and increases for the remainder of the period. Over the entire pre-recession period, the index falls a slight 1.12 percent, from a May 2005 value of 0.418 to a November 2007 value of 0.413. In other words, around 41 percent of the unemployed are in the wrong industry. Federal Reserve Bank of St. Louis REVIEW May/June 2013 249 Canon, Chen, Marifian Alternatively, the IJR3 19-industry index, which measures the proportion of unemployment attributable to structural imbalance, offers perhaps a more intuitive mismatch index. The IJR3(ISSTV1) (Figure C2) index declines very modestly in the first half of the period and then begins to increase in June 2006, eight months prior to IJR1. In contrast to the net decline in the IJR1 index, IJR3(ISSTV1) increases 2.93 percent over the period. The 19-industry values over the period for IJR3 imply that structural unemployment accounts for approximately 11 percent of the total unemployment. The behaviors of the corresponding IJR1 and IJR3(ISSTV1) occupation indexes (22 occupations) diverge from the marginally downward trend of the corresponding industry indexes. The IJR1 22-occupation index exhibits an upward trend during the pre-recession period, with an overall period increase of 12.52 percent (Figure C3). The index values indicate that approximately 36 percent of the unemployed were in the wrong occupation in May 2005 (compared with approximately 42 percent in the wrong industry) and that percent increases to 40 in November 2007. The pattern of the IJR3(ISSTV1) 22-occupation index (Figure C4) is a more pronounced version of the pattern seen in the IJR1 occupation index, with monthly percentage increases for IJR3(ISSTV1) averaging 2.5 times those of IJR1. The IJR3(ISSTV1) index begins the first half of the period with a steady incline, slightly picking up speed from June 2006 until January 2007, when it begins to decelerate. Over the entire period, the IJR3(ISSTV1) 22-occupation index increases 33.87 percent, which is more than 2.5 times the increase seen in the IJR1 occupation index. The values of the IJR3 occupation index indicate that by December 2007 (the start of the Great Recession), approximately 15 percent of the unemployment across 22 occupation groups would be attributed to structural imbalance, up from 11 percent in May 2005. Clearly, both the IJR1 and IJR3(ISSTV1) indexes indicate that occupational mismatch was on the rise even before the recession had begun. On the other hand, these indexes would suggest that industrial mismatch was little changed over the period leading up to the recession and began to increase slightly only in the six months preceding the recession. We next consider the behavior of the IJR2 and IEvans indexes during the pre-recession period.29 The net changes of the 19-industry indexes contrast with one another, while the occupation indexes both exhibit net increases over the period. The IJR2 industry index, which measures the share of the labor force in the wrong sector, demonstrates a slightly downward trend for the majority of the pre-recession period, falling 5.38 percent from May 2005 to November 2007 (Figure C5). This industry index reaches its minimum value of 1.652 in February 2007, after which it begins a period of acceleration that lasts for the remainder of the period. As described previously, the IJR2 index calculates imbalance as absolute deviations of sectoral unemployment shares from sectoral vacancy shares, weighted by one-half the aggregate unemployment-to-labor ratio. While IEvans is similar to IJR2 in that both indexes measure imbalance as a share of the labor force, IEvans calculates imbalance as one-half the deviation of the sectoral difference between unemployment rates and vacancy rates from the aggregate difference between the unemployment and vacancy rate, with each sector weighted by its labor force share (labor supply weights). The difference in calculations yields differing behaviors in the industry indexes, which do not share the same trends over the pre-recession period. In contrast to IJR2, which falls on net from 250 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian May 2005 to November 2007, the IEvans 19-industry index increases consistently over the period (Figure C6), rising 18.77 percent from a value of 0.798 in May 2005 to a November 2007 value of 0.948. For occupational mismatch, both the IJR2 and IEvans indexes increase on net over the period (7.19 percent and 16.81 percent, respectively), with paths different from each other but largely consistent with their corresponding industry indexes. The IJR2 occupation index (Figure C7) follows a softened version of the pattern for the industry index during the pre-recession period, declining slightly from its May 2005 value of 1.670 until it reaches its sample minimum of 1.564 in October 2006. Following this trough, the IJR2 occupation index begins to increase, with acceleration beginning around July 2007 and lasting through the recession period, mirroring the behavior of the corresponding industry index. Worth noting is that the IJR2 occupation index values are less than the industry values throughout the period. Among the six indexes, IJR1 and IJR2 are the only indexes for which the industry values are consistently greater than the occupation values; for both of these indexes, this trend reverses in the beginning of the post-recession period. Furthermore, the spread between the IJR2 industry and occupation indexes, which averages 0.13 percentage points over the period, narrows from a spread of 0.27 percentage points in May 2005 to a spread of 0.04 percentage points in November 2007 (see Figures C5 and C7). As Figure C8 shows, the IEvans occupation index follows a trend similar to the IEvans industry index (see Figure C6). From May 2005 to November 2007, the index increases 16.81 percent (compared with the industry index’s 18.77 percent increase), with an average monthly increase of less than 1 percent until August 2007. The IEvans occupation index values, which are 1.205 in May 2005 and 1.407 in November 2007, are greater than the values of the corresponding industry index over the period, with an average spread of 0.41 percentage points, but the trends of the two indexes are consistent. The IJLS indexes, which are calculated as half the variance of the ratios of sectoral unemployment rates to the mean unemployment rate, provide an indication of the dispersion of unemployment rates across sectors. The industry and occupation indexes exhibit slight concavity over the period, increasing slightly from May 2005 until September 2006 (industry) and November 2006 (occupation) and subsequently falling for the remainder of the period (Figures C9 and C10). Thus, the index indicates that in the first half of the pre-recession period, the dispersion of sectoral unemployment rates from the aggregate rate was slightly increasing, while over the second half it was decreasing to its original May 2005 level. On net, over the period the IJLS industry and occupation indexes increased only 4.12 and 4.14 percent, from May 2005 values of 0.100 and 0.134, respectively, to values of 0.104 and 0.140 in November 2007 (see Figures C9 and C10). Unlike the other indexes, which require unemployment and vacancy data to measure the degree of mismatch in the economy, ILilien measures mismatch through calculations of employment dispersion across sectors.30 Figure C11 plots the ILilien 19-industry index, which experiences a sharp decrease in the first year of the pre-recession period, falling from a May 2005 value of 1.745 to a sample low of 1.410 in October 2006. Following this trough, the index recovers slightly to 1.489 by November 2007. This index’s decline of 14.65 percent from May 2005 to November Federal Reserve Bank of St. Louis REVIEW May/June 2013 251 Canon, Chen, Marifian Table 2 Indexes that Increased During the Great Recession Index Type of mismatch Increase (%) preceding Great Recession Increase (%) over Great Recession period IJR1 Industrial 1.13 3.49 IJR3(ISSTV1) Industrial 4.96 8.79 IJR2 Industrial 10.93 92.83 IJR2 Occupational 14.43 88.79 IEvans Industrial 18.77 73.82 IEvans Occupational 16.89 39.69 IJR1 Occupational 12.52 0.80 IJR3(ISSTV1) Occupational 33.87 1.98 NOTE: All but two of the indexes increased during the Great Recession. For the IJR1 and IJR3(ISSTV1) 22-occupation indexes, most increases were completed before the recession with only minimal increases during the Great Recession. 2007 is the largest of the 19-industry indexes over the pre-recession period. At the 22-occupation level, the ILilien index also declines, falling 15.27 percent over the pre-recession period (Figure C12). Like its corresponding industry index, the ILilien occupation index reaches its sample low during the pre-recession period, with a minimum value of 1.425 in July 2007, about 9 months after the industry index’s minimum value. Great Recession Mismatch Similar to the pre-recession period, the behaviors of most indexes differ over the Great Recession period (December 2007–July 2009). Despite their different paths, four of the six 19industry indexes increase, and the remaining two indexes (IJLS and ILilien) are approximately flat. The 22-occupation IJR2, IEvans, and ILilien indexes increase over the period, while IJR1, IJR3(ISSTV1), and IJLS are essentially flat, exhibiting only minimal increases, if any. As during the pre-recession period, the IJR1 and IJR3(ISSTV1) industry indexes follow a comparable trend, while the corresponding occupation indexes are flatter, with IJR3(ISSTV1) exhibiting slightly more concavity. The IJR2 industry index behavior is similar to that of IEvans: Both increase significantly from values that are relatively low for the entire sample. The trends for the IJR2 and IEvans occupation indexes are also similar to each other and their corresponding industry indexes over the period. It is important to note that while many of the indexes did increase over the Great Recession, most already had begun increasing before the recession began (the ILilien 19-industry index and the IJLS indexes are the exceptions) (see Figures 2 and 3). Table 2 shows that all but two of the indexes increased during the Great Recession; the percentage increase over the Great Recession is listed in the far-right column. The table also shows the percentage by which each index increased before December 2007, when the Great Recession began.31 As is clear from Table 2, most indexes began to increase in the pre-recession period. Six of the eight indexes, including IJR1 and IJR3(ISSTV1) 19-industry and both types of the IJR2 and IEvans 252 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian indexes, exhibited increases over the Great Recession of greater magnitude than the increases in the pre-recession period. For the other two indexes—namely, the IJR1 and IJR3(ISSTV1) 22-occupation indexes—most of the increases were completed before the recession and increased only minimally during the Great Recession. Even though the Great Recession accelerated the degree of mismatch for six of the eight indexes, it was not the initial catalyst of the increases; positive increases preceded the increases during the Great Recession for all eight indexes listed in the table. The 19-industry IJR1 index values, which began to increase at the end of the pre-recession period, continued to increase through the first half of the Great Recession period; the index takes on a concave shape throughout the majority of the period. The index reaches a sample maximum value of 0.432 in January 2009 (see Figure C1), but this value is only 3.52 percent greater than the index’s initial value of 0.418 in May 2005. Following its maximum value, the index declines for the remainder of the period, yielding an overall increase of 3.49 percent from December 2007 to June 2009. The corresponding occupation index (see Figure C3) takes a different path during the Great Recession, increasing a mere 0.80 percent. Although this index also reaches its maximum value in January 2009, most of the movement responsible for the index’s generally concave shape occurs during the pre- and post-recession periods. It is interesting to note that the IJR1 industry and occupation values, which appeared to be converging over the course of the pre-recession period, began to diverge, given that the industry index increases while the occupation index remains constant. Also worth noting is that this is one of two indexes for which the industry values are consistently greater than the occupation values during the Great Recession period. The IJR3(ISSTV1) industry index (see Figure C2) mirrors the IJR1 industry index, exhibiting a concave shape throughout the majority of the period. It increases to a sample maximum of 0.127 in February 2009 and declines for the remainder of the period. This industry index increased 10.14 percent from December 2007 to February 2009, compared with 13.36 percent from May 2005 to February 2009. Because of the declines during the second half of the recession, however, the index’s overall increase from December 2007 to June 2009 was slightly less at 8.79 percent. As for occupational mismatch, Figure C4 shows that the IJR3(ISSTV1) 22-occupation index is significantly greater than the 19-industry index, with occupation values ranging from 0.151 to 0.155 during the Great Recession, compared with the industry index’s range of 0.116 to 0.127. Although the IJR3(ISSTV1) occupation index reaches a sample maximum of 0.155 in November 2008, the increase from the value at the beginning of the recession to the maximum value is only a slight 3.15 percent. As with the IJR1 occupation index, most of the increase in the IJR3(ISSTV1) index from May 2005 to the sample maximum occurs during the pre-recession period. Figure C5 shows the IJR2 industry index during the Great Recession. Like most of the other industry indexes, the increase in this index begins a few months before the recession and continues to accelerate until August 2008, approximately the midpoint of the recession. After this inflection point, the index decelerates slightly for the remainder of the period, increasing to the period maximum of 3.535 by June 2009. The total increase over the period is a substantial 92.83 percent. It is important to note why IJR3 demonstrates this rapid incline while IJR1 and IJR3(ISSTV1) do not. As stated previously, IJR2 is effectively the same as IJR1 scaled by U/L, the total stock of Federal Reserve Bank of St. Louis REVIEW May/June 2013 253 Canon, Chen, Marifian unemployed over the labor force. Since the unemployment level increases rapidly (specifically, by 7,465,000) from December 2007 to June 2009 and the labor force increases considerably less (875,000) over the same period, it is no surprise that the net effect is that IJR2 mirrors the unemployment level, while IJR1 and IJR3(ISSTV1) increase only slightly over the period. As shown in Figure C7, the behavior of the IJR2 22-occupation index during the Great Recession is very similar to the corresponding industry index (see Figure C5). Like the industry index, it accelerates throughout the recession period, slowing in approximately November 2008. Over the period, the occupation index increases 88.79 percent. In contrast to their differing trends during the pre-recession period, over the Great Recession period the IEvans indexes follow a pattern very similar to that of the IJR2 indexes. Figure C6 shows the 19-industry IEvans index, which increases 73.32 percent from the beginning of the recession to the end. Like the IJR2 industry index, the IEvans index reaches its peak in December 2009, after the Great Recession ends. The trend of the IEvans occupation index (see Figure C8) is somewhat similar to the corresponding industry index, although the period increase of 39.69 percent is considerably less than the 73.32 percent increase at the industry level. Nevertheless, the IEvans occupational mismatch (22 occupations) value is 47 percent higher than the IEvans industrial measure (19 industries) at the beginning of the recession, although it falls to 20 percent by the end of the recession. The IJLS indexes, shown in Figures C9 and C10, exhibit little change over the period. The 19-industry index declines a modest 2.50 percent over the period, while the 22-occupation index changes even less, with an increase of 0.52 percent. As previously noted, both the ILilien 19-industry and 22-occupation indexes declined over the pre-recession period. Yet over the Great Recession period, the behaviors of the ILilien indexes diverge. For the last 9 months of the pre-recession period and the first 5 months of the Great Recession, the industry index values are greater than those of the occupation index. For the remainder of the Great Recession period, however, the opposite is true. The ILilien 19-industry index is essentially flat over the Great Recession period, increasing a minimal 0.44 percent from December 2007 to June 2009 (see Figure C11). Figure C12 shows that, in contrast, the ILilien 22-occupation index grows steadily over the period, resulting in a 14.09 percent increase from December 2007 to June 2009. Post-Recession Mismatch The labor market dynamics after the recession (July 2009–May 2012) were such that a variety of measures would offer comparable trends in their behavior. Over the post-recession period, all 19-industry and 22-occupation mismatch indexes exhibit modest to significant declines, with the exception of the ILilien indexes, which increase steadily over the period. For the 19-industry mismatch calculations, the IJR1, IJR2, IJR3(ISSTV1), IEvans , and IJLS indexes all decline during the post-recession period. While the IJR2 and IEvans indexes (see Figures C5 and C6) reach their peaks a few months after the recession ends, IJR1 and IJR3(ISSTV1) (see Figures C1 and C2) had already reached their sample maximums a few months before the end of the recession. As shown in Figure C1, the IJR1 industry index declines 28.54 percent, while the IJR2 index declines 37.74 percent (see Figure C5). The IJR3(ISSTV1) industry index declines the most over the 254 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian period, at 41.95 percent (see Figure C2). Falling somewhere in the middle of the indexes, the IEvans industry index fell 31.87 percent (see Figure C6), while the IJLS 19-industry index declined 32.24 percent (see Figure C9). The only index that diverged from the industry group was the ILilien 19-industry index; Figure C11 shows that this index rose 18.7 percent over the period, an increase that began immediately following the recession. This index indicates that sectoral employment’s dispersion from the aggregate was increasing throughout the post-recession period, which implies that a greater number of workers would need to change sectors to adjust to shifts in employment demand. Yet despite the index’s significant increase over the entire period, it began to taper in October 2011, and for the subsequent 8-month period it remained at a constant 1.777, suggesting a possible stagnation and perhaps the beginning stages of a trend reversal. Like their corresponding 19-industry indexes, the 22-occupation indexes fall during the post-recession period in differing magnitudes. The IJR1, IJR3(ISSTV1), IJR2, IEvans, and IJLS indexes fall 15.22 percent, 33.47 percent, 26.51 percent, 13.39 percent, and 17.10 percent, respectively, from July 2009 to May 2012. The exception is the 22-occupation ILilien index, which increases 5.36 percent over the period (see Figure C12). Figure C3 shows that the IJR1 occupation index continues to exhibit slight concavity over the first half of the post-recession period, switching to a linear path in the second half of the period. This index’s decline of 15.22 percent is over half the magnitude of the industry decline (see Figure C1). The IJR3(ISSTV1) occupation index maintains its concave shape, as well as the declines that begin halfway through the recession (see Figure C4). Of all the indexes, the pattern of the IJR2 occupation index (see Figure C7) most resembles that of its corresponding industry index (see Figure C5). Like the industry index, in the beginning of the post-recession period the IJR2 occupation index increases at a decelerated rate before reaching its sample peak, although the occupation index peaks a few months later in February 2010, compared with November 2009 for the industry index. From the beginning to the end of the post-recession period, the IJR2 22-occupation index falls 26.51 percent. Of note, the convergence of the IJR2 index trends with other indexes such as IJR1 and IJR3(ISSTV1) can be explained by the fact that U/L falls between 0.081 and 0.100 over the entire post-recession period. The IEvans occupation indexes in Figure C8 follow a trend similar to the IEvans industry indexes. Also like the IJR2 occupation index (see Figure C7), the IEvans 22-occupation index increases in the first few months of the period before hitting its sample maximum of 2.076 in March 2010. The IEvans occupation index (see Figure C8) is much flatter than its corresponding industry index, falling only 13.39 percent, compared with the industry index’s decline of 31.87 percent (see Figure C6). The IJLS occupation index (see Figure C10) falls 17.10 percent over the period, less than the 32.24 percent decline in the corresponding industry index (see Figure C9). Part of this difference stems from the tapering off of the occupation index in the last four months of the sample, in contrast to the consistent declines seen in the industry index throughout the period. The last occupation index, ILilien, takes on trends somewhat similar to those of its corresponding industry index. Figure C12 shows that the 22-occupation index continues its increases from the recession period, reaching a sample maximum of 1.854 in March 2011. It subsequently Federal Reserve Bank of St. Louis REVIEW May/June 2013 255 Canon, Chen, Marifian declines for the remainder of the period, falling 7.16 percent from its maximum to its final value of 1.721 in May 2012, but still 5.36 percent higher than its value at the beginning of the period. Counterfactual Unemployment In addition to computing the indexes for industrial and occupational mismatch, we use ISSTV1 to compute the counterfactual unemployment rate (the unemployment rate that prevails when there is no mismatch) presented in the previous section.32 Following Şahin et al. (2012), we calculate the counterfactual unemployment rate, which, when subtracted from the actual unemployment rate, indicates how much of the recent rise in the unemployment rate can be attributed to mismatch. Figures 4 and 5 show the actual unemployment rate plotted with our two calculations of the counterfactual unemployment rates, which use (i) industries disaggregated into 12 and 19 categories and (ii) occupations disaggregated into 10 and 22 categories. The spread between the actual and the counterfactual unemployment rates is always greater than zero, even when the unemployment rate is low, indicating that mismatch unemployment exists throughout the sample. The counterfactuals tend to follow the path of the actual unemployment rate, with fluctuations appearing to lag the actual rate slightly. Figure 4 shows that the lesser-disaggregated indexes appear superimposed throughout the majority of the sample; alternatively, with the greater level of disaggregation, the industrial counterfactual unemployment rate generally falls at higher values than those for the occupational rate (see Figure 5). Again, we focus our description of the data on the greater disaggregations, 19 industries and 22 occupations (see Figure 5). At the beginning of the pre-recession period, both the industrial and occupational counterfactual unemployment rate calculations have the same value at 3.60 percent. Over the entire period, both rates exhibit convex behavior, declining to lows of 2.63 percent (19-industry) and 2.47 (22-occupation) percent in April 2007. From then until the beginning of the recession, the values increase by 11.08 percent and 9.40 percent, respectively, to 2.93 percent and 2.70 percent unemployment in the absence of mismatch. The counterfactual unemployment rates begin their rapid growth during the Great Recession, following the trend of the actual unemployment rate. Over the entire recession period, the actual, industrial counterfactual, and occupational counterfactual unemployment rates increase 102.13 percent, 127.06 percent, and 133.07 percent, respectively, from their initial values in November 2007, which suggests that the counterfactual unemployment rates (the rates in the absence of mismatch) responded more sensitively during the recession than the actual values. Also worth comparing is the slope of the counterfactuals versus the slope of the actual unemployment rate, specifically over the six-month period when the paths appear linear. The actual unemployment rate exhibits a steep increase of 42.62 percent from its September 2008 value of 6.10 percent to its March 2009 value of 8.70, yet this increase is considerably less than the 65.21 percent increase exhibited by the occupation counterfactual unemployment rate from the October 2008 value of 3.55 percent to the April 2009 value of 5.87 percent. From the beginning of the Great Recession to the unemployment rate peak in October 2009, the mismatch unemployment rate increased 1.28 percentage points in the industrial type and 1.44 percentage points in the occupational type. This result implies that the rise in industrial 256 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian Figure 4 Comparison of Actual and Counterfactual Unemployment Rates (lesser disaggregation) Percent 11 10 9 8 7 6 5 Actual Unemployment Rate 12 Industries 10 Occupations 4 3 2 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 Figure 5 Comparison of Actual and Counterfactual Unemployment Rates (greater disaggregation) Percent 11 10 9 8 7 6 5 Actual Unemployment Rate 19 Industries 22 Occupations 4 3 2 May-05 Jun-06 Federal Reserve Bank of St. Louis REVIEW Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 May/June 2013 257 Canon, Chen, Marifian and occupational mismatch, as measured by ISSTV1, can account for at most 2.72 percentage points of the 5.30-percentage-point increase in the unemployment rate from the start of the recession to the unemployment rate peak. CONCLUSION In this article, we provide an overview of five studies of labor market mismatch and use new U.S. vacancy data to calculate six of the corresponding indexes and a counterfactual unemployment rate. Similar to two decades ago, when the mismatch literature indicated that industrial and occupational mismatch played a nonnegligible role in the increase in European unemployment, we find notable increases in both industrial and occupational mismatch in the United States from May 2005 to November 2009. Our calculations of the counterfactual unemployment rates revealed that the total increase in industrial and occupational mismatch unemployment can account for at most 2.72 percentage points of the 5.30-percentage-point increase in the unemployment rate from the beginning of the Great Recession to the unemployment rate peak in October 2009. Perhaps the most suitable indexes for describing mismatch in the United States are the IJR3 and ISSTV1 indexes, thanks to their intuitive interpretations and their ability to offer a measurement of the contribution of mismatch to unemployment. Alternatively, the IJR2 index perhaps is the least effective in providing an indication of mismatch. Because the IJR2 index is scaled by the unemployment and labor market level, it is susceptible to volatility in these statistics, which could overshadow important trends that can be gleaned from the dispersion of unemployment rates from vacancy rates. Furthermore, although computing the IJLS index is simple and requires only unemployment data, Padoa-Schioppa (2008b) concludes from Freeman (2008) and Bean and Pissarides (2008) that the assumption on wage functions is not realistic for countries such as the United Kingdom and United States, and thus its results should be approached with caution. Some general limitations of mismatch indexes are noteworthy. First, they do not provide any information on the sources of mismatch. As a result, the indexes have limited relevance for the design of demand stimulation policies because the root of the problem is not identified. As Diamond (2011) affirms, “Insofar as direct measures of frictional or structural unemployment are dependent on the tightness of the labor market, they have limited relevance for the design of demand stimulation policies” (p. 1069). Another important drawback to existing mismatch indexes is that they do not capture any newly emerged labor sectors, which may bias the observed level of mismatch. Almost every theoretical framework of mismatch indexes assumes that the aggregate labor market is divided into a fixed number of sectors that differ significantly in terms of skills, locations, and so on. None of the indexes, however, accounts for potential changes in the number and the composition of sectors. As an example from recent history, the emergence of information technology over the past three decades may have created some new sectors while destroying some old ones. Empirically, it is currently difficult for mismatch indexes to capture this issue because of data limitations, but it is worthwhile to consider this issue in the theoretical framework for future research. 258 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian We end by considering a recent alternative explanation of the forces that resulted in the jobless recovery in the United States. In their 2012 paper, Jaimovich and Siu argue that the jobless recovery is due to job polarization and that job polarization is a business cycle phenomenon. Job polarization describes the scenario in which employment grows at the tails of the skills distribution (high-skill and low-skill) but disappears for occupations in the middle of the skills distribution (middle-skill). Central to their analysis is the classification of occupations as either routine (middle-skill) or nonroutine (high-skill and low-skill). They believe that the jobless recovery is caused by the downturn-induced job losses in middle-skill occupations over the past 30 years that result from technological advances and labor offshoring. If Jaimovich and Siu (2012) are correct, then our understanding of industrial mismatch would need to be altered: Industrial mismatch has little or nothing to do with vacancies existing in growing industries with unemployment in other, distressed industries. Instead, the industrial mismatch that exists would be characterized by a scenario in which the unemployed belong to a certain occupational type (routine jobs) and the job vacancies are for the opposite occupational type (nonroutine jobs), regardless of the industry. Assuming Jaimovich and Siu’s (2012) conclusions are accurate, such analysis would capture most effectively the degree of mismatch in the economy, because it approaches the mismatch analysis with a framework based on the changing dynamics of developed labor markets. We believe such examination is a worthwhile avenue for future research in mismatch using unemployment and vacancy data. Federal Reserve Bank of St. Louis REVIEW May/June 2013 259 Canon, Chen, Marifian APPENDIX A: INDEXES The Lilien Index In “Sectoral Shifts and Cyclical Unemployment,” Lilien (1982) responds to work by Lucas and Prescott (1974), pointing out that their derivation of the stationary equilibrium unemployment rate is dependent on the assumption that market-specific demands do not vary over different time periods. Lilien’s analysis begins with the rejection of this assumption, which results in a time-variant equilibrium unemployment rate that moves with the quantity of labor reallocation (see p. 778). To craft his dispersion index, Lilien (1982) first constructs hiring, separation, layoff, and accession functions that allow him to derive a model of the aggregate layoff rate. He assumes that the net hiring rate of a typical firm, ht , can be expressed as the sum of the aggregate hiring rate Ht and a random disturbance ⑀t . Specifically, the hiring function takes the form ht = H t + t , where ⑀t is assumed to be distributed with mean zero and variance st2 according to the density function f(⑀t |st); st is the measure of sectoral dispersion of employment demand, whereby an increase in st occurs when a shock has a varying impact on firms. Lilien (1982) proposes a proxy for st , ŝt ,33 which takes the following form: 1/ 2 n x 2 I Lilien = σˆ t = ∑ it ( ∆ log xit − ∆ log Xt ) , i=1 Xt where xit is the number of people employed in sector i at time t and Xt is the total employment for all sectors. The index measures the dispersion of changes in sectoral employment from changes in aggregate employment. Lilien then uses the index to estimate the effect of dispersion on unemployment and aggregate layoffs. The unemployment equation is derived from a simple flow model, where the change of unemployment is equal to unemployment inflow34 minus unemployment outflow. The Jackman and Roper Indexes Jackman and Roper’s (1987) job hiring function takes the form H i = Vi h (U i Vi ) (h' > 0 ,h'' < 0) , where Ui and Vi are the number of unemployed and vacancies in sector i, respectively. They maximize the sum of Hi across all sectors, subject to the constant ∑U i i = U and the given Vi. When the first-order condition of the maximization problem is satisfied, the derivative of the hiring function is equal to a constant, indicating that the unemployment-to-vacancies ratio is identical across sectors, and they define this as a labor market at structural balance. From this maximization, a measure of structural unemployment directly follows, given by the number of 260 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian unemployed workers needed to move from one sector to another to maximize aggregate hires. Specifically, SU = 1 ∑ U i − (U V )Vi , 2 i or, alternatively, 1 SU = U ∑ ûi − v̂i , 2 i where ûi = (Ui /U) and v̂i = (Vi /V) are sector i’s shares of unemployed and vacancies, respectively. The Evans Index Evans’s (1993) framework assumes short-run disequilibrium and looks at the “aggregate impact of sectoral imbalance on labour markets” (p. 440, footnote 1). Evans states that his theoretical orientation differs from that of other papers discussing labor market imbalance, such as those by Jackman, Layard, and Pissarides (1989) and Jackman, Layard, and Savouri (2008), who focus instead on equilibrium labor markets’ frictional factors. Deviating from conventional macroeconomic models with aggregate supply and aggregate demand equations,35 Evans’s framework assumes that wages are predetermined; as a result, labor market vacancies and unemployment enter the model. Under this framework, a partial equilibrium analysis of the aggregate supply side (see Hansen, 1970) shows that the degree of sectoral imbalance affects the position of the Beveridge curve (unemployment-vacancy [UV] curve). Accordingly, a neutral variation in aggregate demand results in movement along the UV curve, while a change in sectoral imbalance alters the position of the UV curve. However, Evans notes that if (i) the demand for labor exhibits a nonneutral change and (ii) there is simultaneously a correlation across sectors between the elasticities of labor demand with respect to aggregate output and the net unemployment rates, then the index of sectoral imbalance can exhibit a cyclical pattern. Specifically, he notes that positive correlations would result in countercyclical behavior by the index, which could result in fluctuations in the unemployment rate being attributed to changes in sectoral imbalance when they are actually a function of demand changes. In such a case, Evans attempts to correct for the problem by first estimating the sectoral elasticities of labor demand with respect to aggregate output, and then using those estimates to remove the effects of variations in aggregate output on the index of sectoral imbalance. This method is at a disadvantage because the cyclical pattern is not corrected within the model itself. The Şahin et al. Indexes Şahin et al.’s (2012) benchmark economy36 is composed of a large number of i distinct labor markets/sectors. New production opportunities arise exogenously in each sector and correspond to job vacancies. Individuals in the economy are risk neutral and can either be employed in sector i(ei ) or unemployed and searching for work in the same sector (ui ). The authors do not Federal Reserve Bank of St. Louis REVIEW May/June 2013 261 Canon, Chen, Marifian allow for on-the-job search. Labor market frictions exist, and the number of new hires hi in sector i is given by the matching function F . fi . m(ui ,vi ), which matches unemployed workers ui with vacancies vi . The variable F represents the aggregate efficiency component, which is the same for all sectors, while the sector-varying matching efficiency is given by fi ; together, F and fi measure the matching efficiency in sector i. The component m is assumed to be strictly concave, strictly increasing, and homogeneous of degree one. The authors also assume that (i) existing matches produce units of output Z (homogeneous sectoral labor productivities) and (ii) the rates of match-destroying Δ are the same for all sectors (homogeneous sectoral job-destruction rates). Therefore, in the benchmark model, the only two sources of heterogeneity across sectors come from the number of vacancies vi and the matching efficiency fi . In this environment, the model operates as follows: Z, Δ, and F are exogenous aggregate shocks, which are observed at the beginning of the period with aggregate vacancies v and matching efficiencies f. Also given at the beginning of the period is the distribution of active matches across labor markets, e = {e1,…,eI }, and the total number of unemployed workers u. Unemployed workers freely choose a labor market, and then the matching process takes place according to the matching function, yielding hi in each market. Production occurs with existing matches ei and the new matches hi ; a share Δ of the matches are destroyed in each market i. Thus, at the end of the period, the employment distribution and the number of unemployed workers are determined for the next period. Following Jackman and Roper (1987), Şahin et al. (2012) assume that the individual-market matching function m(uit ,vit) is Cobb-Douglas, taking the form hit = Φt φit vitα uit1−α , where hit are hires in sector i at date t and a ∈ (0,1) is the vacancy share for all sectors. Then, the first index (for the benchmark environment) is I SSTV 1φ α 1−α I ht φit vit uit = M φ t = 1 − * = 1 − ∑ , ht i =1 φ t vt ut where ht* is the planner’s hires and ht is the observed level of aggregate hires. APPENDIX B: DATA HWOL Vacancy Data General. Ad-scraping is very common and the HWOL program identifies and eliminates from collection any job boards that are simply aggregators of ads from other job boards (and therefore do not offer any new or unique ads). Each year the job board sources change to (i) stay up-to-date with new sources and (ii) drop existing sources that primarily maintain postings from other job boards. New job boards are identified based on independent research and recommendations from industry sources. 262 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian Geographic, Occupational, and Industry Levels. Geographic levels available for the job posts include metropolitan area, state, regional, and national levels. Geographic identities are recorded using Federal Information Processing Standards (FIPS) codes for ads at the county level, and metropolitan levels use the 2005 Office of Management and Budget (OMB) countybased definitions for metropolitan statistical areas (MSAs); regional levels use U.S. Census Bureau region definitions. Counts by Census regions are the direct sum of the respective state counts (including the District of Columbia). The national count is a direct sum of the state counts, plus ads designated as nationwide for their location requirement. A job post’s area code is assigned based on the location cited in the text of the ad itself. Approximately 93 percent of the ads are coded to a county or city level, 5 percent are coded as statewide ads, and 2 percent are coded as nationwide. The job posts are also classified by occupation and industry. Occupation codes use the 2000 OMB Standard Occupational Classification (SOC) system. The codes are assigned to the ads by an autocoder37 at the six-digit and the eight-digit O*Net level based on the job titles and job descriptions. The Conference Board periodically updates the O*Net classification of HWOL data and reclassifies the entire HWOL database with new federal government revisions to the SOC manual. Industry codes are determined using the advertiser’s name listed on the ads. The advertiser’s name is then searched in the Dun & Bradstreet database of companies, and six-digit North American Industry Classification System (NAICS) codes are taken from there. Because approximately 40 percent of ads do not provide information about the employer, it is difficult to identify the industry for these ads. For employers that can be identified, industry code assignments are very accurate. Unduplication. Ads are first unduplicated at the Census Bureau’s place code level (the lowest level of geographic coding). Additional unduplication is performed as geographic scope increases. If the identical ad appears in two different counties in the same MSA, it would be counted separately in each of the counties but would be counted only once at the MSA level (i.e., one of the ads would be treated as a duplicate in the MSA total). The same procedure applies to counting ads at the state level. Accordingly, there is no direct additivity from the county level to the MSA, state, or national level because summing the lower levels of geography (i.e., counties) produces a larger number of vacancies than will be retrieved by a query at higher geographic (i.e., MSA or state) levels. In a limited number of instances, duplication can also occur from inaccurate coding of occupations. According to the Conference Board, the autocoder is able to assign occupation codes to 96 percent to 99 percent of all unduplicated ads. If two ads that are unduplicated in the U.S. total query are assigned different six-digit SOC codes, then they would both appear in their independent SOC code tabulations and thus would fail to be identified as duplicates in the occupation code query.38 As such, the number of ads aggregated by adding all occupations is slightly larger (on average 1 percent) than the number obtained in the U.S. total query, even though not all ads are assigned an occupation code (see Figure 1). This problem is unlikely to have a significant effect on computation of mismatch indexes. Other Data Series Relating to Vacancies. HWOL’s national ads count trend is very similar to that indicated by another vacancy dataset, the Bureau of Labor Statistics’ (BLS) Job Openings Federal Reserve Bank of St. Louis REVIEW May/June 2013 263 Canon, Chen, Marifian Figure B1 HWOL Data Compared with JOLTS Data (U.S. Totals, May 2005–May 2012) Vacancies (in Millions) 5.00 4.50 4.00 3.50 3.00 2.50 2.00 1.50 HWOL U.S. Total JOLTS U.S. Total 1.00 May-05 Jan-06 Sep-06 May-07 Jan-08 Sep-08 May-09 Jan-10 Sep-10 May-11 Jan-12 and Labor Turnover Survey (JOLTS) (Figure B1).39 JOLTS is a monthly series that attempts to provide a broad national picture of U.S. hiring activity. JOLTS collects monthly data of job openings, separations, and hires from a sampling of 16,000 of 7 million nonfarm business establishments. JOLTS data are released roughly a month after their reference period and are available for December 2000 onward.40 The definition of a job opening in JOLTS also allows it to provide an accurate estimate for national job vacancies.41 The JOLTS definition of hires includes rehires and hires of people who were previously not in the labor force. A disadvantage of JOLTS data is that they can be broken down into only 4 Census regions and 18 broad industries at the national level. In contrast, HWOL’s geographic completeness and occupational detail enable the computation of regional and occupational mismatch; HWOL also provides sufficient vacancy information for computing mismatch at more-disaggregated levels. An additional alternative source of vacancy data is the Conference Board’s Help Wanted Index (HWI), which offers an index for the number of ads in 51 leading newspapers from their respective cities.42 It is the only known source that provides estimates for national job vacancies before 2000. The HWI has data from 1951 to the second quarter of 2010, when it was discontinued. The series does not provide actual counts of newspaper ads and it does not offer any industrial or occupational information. 264 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian APPENDIX C: INDUSTRY AND OCCUPATION INDEXES Figure C1 Figure C2 IJR1 Industry Indexes IJR3(ISSTV1) Industry Indexes Share of Unemployment in the Wrong Sector Proportion of Unemployment Attributable to Mismatch 0.45 0.17 0.40 0.15 0.13 0.35 0.11 0.30 0.09 0.25 0.07 0.20 0.05 12 Industries 19 Industries 0.15 0.10 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 12 Industries 19 Industries 0.03 0.01 May-05 Jun-06 Jul-07 Aug-08 Figure C3 Figure C4 IJR1 Occupation Indexes IJR3(ISSTV1) Occupation Indexes Sep-09 Oct-10 Share of Unemployment in the Wrong Sector Proportion of Unemployment Attributable to Mismatch 0.45 0.17 0.40 0.15 Nov-11 0.13 0.35 0.11 0.30 0.09 0.25 0.07 0.20 0.05 10 Occupations 22 Occupations 0.15 0.10 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Federal Reserve Bank of St. Louis REVIEW Oct-10 Nov-11 10 Occupations 22 Occupations 0.03 0.01 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 May/June 2013 265 Canon, Chen, Marifian Figure C5 Figure C6 IJR2 Industry Indexes IEvans Industry Indexes Percent of Labor Force in the Wrong Sector Average Deviation of the Percentage Difference Between Labor Supply and Demand Across Sectors 4.10 2.20 3.60 2.00 1.80 3.10 1.60 2.60 1.40 2.10 1.20 1.60 1.00 12 Industries 19 Industries 1.10 0.80 12 Industries 19 Industries 0.60 0.60 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 0.40 May-05 Jun-06 Jul-07 Aug-08 Figure C7 Figure C8 IJR2 Occupation Indexes IEvans Occupation Indexes Percent of Labor Force in the Wrong Sector Sep-09 Oct-10 Nov-11 Average Deviation of the Percentage Difference Between Labor Supply and Demand Across Sectors 4.10 2.20 3.60 2.00 1.80 3.10 1.60 2.60 1.40 2.10 1.20 1.60 1.00 10 Occupations 22 Occupations 1.10 0.80 10 Occupations 22 Occupations 0.60 0.60 May-05 266 Jun-06 Jul-07 May/June 2013 Aug-08 Sep-09 Oct-10 Nov-11 0.40 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian Figure C9 Figure C10 IJLS Industry Indexes IJLS Occupation Indexes Variance of Unemployment Rates Across Sectors Variance of Unemployment Rates Across Sectors 0.20 0.20 12 Industries 19 Industries 0.18 0.18 0.16 0.16 0.14 0.14 0.12 0.12 0.10 0.10 0.08 0.08 0.06 0.06 0.04 0.04 0.02 May-05 Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 0.02 May-05 10 Occupations 22 Occupations Jun-06 Jul-07 Aug-08 Figure C11 Figure C12 ILilien Industry Indexes ILilien Occupation Indexes Sep-09 Oct-10 Nov-11 Percent Log Deviations of Sectoral Employment from Total Employment (Scaled by 100) Percent Log Deviations of Sectoral Employment from Total Employment (Scaled by 100) 2.00 2.00 1.80 1.80 1.60 1.60 1.40 1.40 1.20 1.20 1.00 0.80 May-05 12 Industries 19 Industries Jun-06 Jul-07 Aug-08 Sep-09 Federal Reserve Bank of St. Louis REVIEW Oct-10 Nov-11 1.00 0.80 May-05 10 Occupations 22 Occupations Jun-06 Jul-07 Aug-08 Sep-09 Oct-10 Nov-11 May/June 2013 267 Canon, Chen, Marifian NOTES 1 See the Conference Board Help Wanted OnLine (HWOL) job advertisements data series shown in Figure 1 (see http://www.conference-board.org/data/helpwantedonline.cfm). 2 Among economists and policymakers, some frequently discussed reasons for this jobless recovery include labor mismatch, extended unemployment insurance benefits, and low aggregate labor demand. For examples, see Şahin et al. (2012), Farber and Valletta (2011), and Elsby et al. (2011). 3 They conclude that negative equity significantly reduced the mobility of homeowners. 4 The Beveridge curve is the empirical and negative relationship between the unemployment rate (typically plotted on the x-axis) and job vacancy rate (y-axis). If the Beveridge curve moves outward (inward), then a given vacancy rate is associated with a higher (lower) unemployment rate. 5 See Padoa-Schioppa (2008a) for a collection of papers on this issue. 6 Lilien sees this point—combined with the fact that service industry employment increased in all three downturns (1970-71, 1975, 1980)—as evidence that sectoral shifts contributed to the economic downturns, as opposed to the downturns causing temporary shifts in employment (see p. 779). 7 From Lilien’s point of view, at that time in the literature’s development, the importance of hiring condition dispersion in determining the aggregate layoff rate was not well understood. However, he simultaneously noted that it was widely observed that some firms issue layoffs when others are hiring new workers. To emphasize the importance of capturing this variance in hiring in the layoff function (and, ultimately, in his unemployment model), he uses the following example: Suppose that in one economy, all the firms are growing at 2 percent, while in another economy, half of the firms are growing at 8 percent and half are shrinking by 4 percent. Even though these two economies have identical aggregate employment growth rates, the first economy would have many fewer layoffs than the second. 8 Whereas Jackman and Roper define labor market frictions as being characterized by the “coexistence of unemployment and unfilled vacancies within a sector,” they believe that structural imbalance is characterized by unemployment in some sectors (beyond any frictional unemployment) coexisting with vacancies in other sectors (beyond any frictional vacancies). 9 They assume that the labor market is divided into a number of sectors, viewing job characteristics as homogeneous within each sector but significantly different across sectors. Their theoretical basis stems from a job hiring function, which is assumed to be convex, linear homogeneous, and the same in all sectors. 10 They note that “wrong” is only in relation to the definition of structural balance (see p. 12). 11 Petrongolo and Pissarides (2001) conclude from studies of matching functions that the Cobb-Douglas form is a good approximation. 12 Aggregate hires is assumed to equal separations at all times (i.e., aggregate employment is unchanged). 13 See Appendix A for Evans’s commentary of the effect of nonneutral changes. 14 They are referencing Evans (1985b). 15 Because the authors assume there are no frictions between sectors, unemployment at the optimal allocation is caused purely by the frictions within sectors. The matching function governs this relationship between unemployed workers and vacancies within a given market i. 16 Computation of I SSTV1f requires hiring data by sector to estimate the sector-specific matching efficiencies f. We choose instead to compute ISSTV1, which is described below. 17 φ m vi gives the matching efficiency-weighted vacancy-to-unemployment ratio in sector i, where u * is the optii U ∗ i ui mal allocation of unemployment in sector i implied by the first-order condition and mU is the partial derivative of m with respect to ui . 18 Data on matching efficiency, labor productivity, and job destruction rates by sector are needed to compute I . SSTV2 19 According to the BLS, the lowest geographic levels of reliable estimates provided by the CPS are the state and 12 of the largest metropolitan statistical areas (MSAs). The Local Area Unemployment Statistics program provides monthly 268 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian data for employment and unemployment at geographic areas smaller than states (U.S. Department of Labor, Bureau of Labor Statistics, 2009). 20 We seasonally adjust all series because some indexes, such as I JLS and ILilien, use only unemployment and employment data, respectively. Therefore, these indexes could exhibit seasonal patterns that might not allow for comparison. 21 New ads are defined as unduplicated ads that did not appear in the previous reference period. Total ads are defined as unduplicated ads that appear in the reference period. Total ads include both new ads and ads from the previous month that have been reposted. 22 Data sources for this section are the Conference Board Help Wanted OnLine® Data Series Technical Notes (August 2012; http://www.conference-board.org/pdf_free/press/TechnicalPDF_4560_1343756152.pdf) and personal email communications with Jeanne Shu (May 2011, June 2011, July 2011, April 2012, and May 2012). 23 However, the Conference Board states that it may not target smaller local Internet job boards in an area with a lim- ited number of ads. 24 The number of online ads from HWOL provides an estimate of the actual count of job openings. When an ad is posted for a position that has multiple openings, it is counted as only one ad in HWOL. 25 As mentioned previously, I JR3 is mathematically equivalent to ISSTV1 because of the assumption that vacancies are the only source of heterogeneity across sectors. 26 Unlike Şahin et al. (2012), who use I SSTV2 to calculate the counterfactual unemployment rate, we use ISSTV1. 27 As an alternative to the Hodrick-Prescott filter, we also calculated a 3-month moving average for each of the indexes and the trends are consistent. We chose to use the Hodrick-Prescott filter because the trends are smoother. 28 Figure numbers that begin with “C” can be found in Appendix C, which provides a figure containing each index’s 12- and 19-industry breakdowns, as well as a figure containing each index’s 10- and 22-occupation breakdowns. 29 We multiply the I JR2 values by 100. 30 We multiply the I Lilien values by 100. 31 The starting point for this calculation is the pre-recession month when the index began an increase that lasted for the remainder of the period. The endpoint is November 2007. For example, the increase preceding the Great Recession for the IJR1 19-industry index is measured as the percentage change over the period May 2006–November 2007, because in May 2006 the index began to increase and continued this increase for the remainder of the prerecession period (see Figure C1). 32 In their paper, Şahin et al. use I SSTV2 to calculate the counterfactual unemployment rate. Since we do not compute ISSTV2 , we use ISSTV1 instead. 33 Padoa-Schioppa (2008b) refers to ŝ t as the “turbulence index” (see p. 2). 34 The term s t (dispersion) enters Lilien’s equation of unemployment inflow through the layoff function. Layoffs, quits (including those who did not find employment before leaving their most recent job), and labor force entrants complete the flow into unemployment. 35 Evans explains that in traditional macroeconomic models (with aggregate supply and aggregate demand equa- tions), sectoral imbalance functions through the aggregate supply side. Referencing Evans (1985a) and Evans (1989), he writes: For example, in a stochastic disaggregated model with temporary sectoral wage floors and a gradual movement of labor between sectors…the position of the aggregate supply curve, in both the short-run and the long-run, depends upon the extent of imbalance between sectors. The degree of imbalance, which is increased from random sectoral shocks and diminished by equilibrating changes in sectoral wages and labour supplies, is appropriately measured, in the stochastic steady state, by the standard deviation of sectoral excess demands for labor, computed at the (temporary) nominal wage floors. [A]n increase in aggregate demand leads, in the short run, to a decrease in unemployment in excess supply sectors and an increase in wages and prices in excess demand sectors. There is no role for vacancies because of the assumed upward flexibility of wages (pp. 440-41). Federal Reserve Bank of St. Louis REVIEW May/June 2013 269 Canon, Chen, Marifian 36 Shimer (2007) develops an alternative theoretical framework for mismatch. 37 The Conference Board states that the autocoder software is selected for its accuracy. The HWOL time series will be converted to the 2010 SOC standard with the release of the 2013 annual revision. 38 This issue will be corrected with the release of the 2013 annual revision, which implements a methodology change where all HWOL counts will now be created by directly summing unduplicated city-level counts. 39 The trend of ad counts by broad industry categories in HWOL is also very similar to what is shown by JOLTS, although the levels are quite different because of the missing industry information for online ads. 40 Job openings for a certain month are defined as the number of openings on the last day of the reference month. 41 See Shimer (2005) for a discussion. 42 See Abraham (1987) for a detailed description of these data. REFERENCES Abraham, Katharine G. “Help-Wanted Advertising, Job Vacancies, and Unemployment.” Brookings Papers on Economic Activity, 1987, 18(1), pp. 207-48. Abraham, Katharine G. and Katz, Lawrence F. “Cyclical, Unemployment: Sectoral Shifts or Aggregate Disturbances?” Journal of Political Economy, June 1986, 94(3 Part 1), pp. 507-22. Bean, Charles R. and Pissarides, Christopher A. “Skill Shortages and Structural Unemployment in Britain: A (Mis)matching Approach,” in Fiorella Padoa-Schioppa, ed., Mismatch and Labour Mobility. Cambridge, UK: Cambridge University Press, 2008, pp. 325-53. Blanchard, Olivier J. and Diamond, Peter A. “The Beveridge Curve.” Brookings Papers on Economic Activity, 1989, 20(1), pp. 1-76. Diamond, Peter. “Unemployment, Vacancies, Wages.” American Economic Review, June 2011, 101(4), pp. 1045-72. Elsby, Michael W.; Hobijn, Bart; Şahin, Ayşegül and Valletta, Robert G. “The Labor Market in the Great Recession—An Update to September 2011.” Brookings Papers on Economic Activity, Fall 2011, pp. 353-84. Evans, George W. “Bottlenecks and the Phillips Curve: A Disaggregated Keynesian Model of Inflation, Output and Unemployment.” Economic Journal, June 1985a, 95, pp. 345-57. Evans, George W. “A Cyclically Adjusted Index of Structural Imbalances,” Unpublished manuscript, London School of Economics, 1985b. Evans, George W. “The Conduct of Monetary Policy and the Natural Rate of Unemployment.” Journal of Money, Credit, and Banking, November 1989, 21(4), 498-507. Evans, George W. “Sectoral Imbalance and Unemployment in the United Kingdom: 1963-84.” Oxford Economic Papers, July 1993, 45(3), pp. 440-56. Farber, Henry S. and Valletta, Robert. “Do Extended Unemployment Benefits Lengthen Unemployment Spells? Evidence from Recent Cycles in the U.S. Labor Market.” NBER Working Paper No. 19048, National Bureau of Economic Research, May 2013; http://www.nber.org/papers/w19048.pdf?new_window=1. Ferreira, Fernando; Gyourko, Joseph and Joseph, Tracy. “Housing Busts and Household Mobility.” Journal of Urban Economics, July 2010, 68(1), pp. 34-45. Freeman, Richard B. “Labour Market Tightness and the Mismatch between Demand and Supply of Less-Educated Young Men in the United States in the 1980s,” in Fiorella Padoa-Schioppa, ed., Mismatch and Labour Mobility. Cambridge, UK: Cambridge University Press, 2008, pp. 360-81. Hansen, Bent. “Excess Demand, Unemployment, Vacancies, and Wages.” Quarterly Journal of Economics, February 1970, 84(1), pp. 1-23. Hobijn, B.“Industry-Occupation Mix.” Working Paper No. 2012-09, Federal Reserve Bank of San Francisco, September 2012; http://www.frbsf.org/publications/economics/papers/2012/wp12-09bk.pdf. 270 May/June 2013 Federal Reserve Bank of St. Louis REVIEW Canon, Chen, Marifian Jackman, Richard A.; Layard, Richard and Pissarides, Christopher A. “On Vacancies.” Oxford Bulletin of Economics and Statistics, November 1989, 51(4), pp. 377-94. Jackman, Richard A.; Layard, Richard and Savouri, S. “Mismatch: A Framework for Thought,” in Fiorella Padoa-Schioppa, ed., Mismatch and Labour Mobility. Cambridge, UK: Cambridge University Press, 2008, pp. 44-101. Jackman, Richard A. and Roper, Stephen. “Structural Unemployment.” Oxford Bulletin of Economics and Statistics, February 1987, 49(1), pp. 9-36. Jaimovich, Nir and Siu, Henry E. “The Trend Is the Cycle: Job Polarization and Jobless Recoveries.” NBER Working Paper No. 18334, National Bureau of Economic Research, August 2012; http://www.nber.org/papers/w18334.pdf?new_window=1. Lilien, David M. “Sectoral Shifts and Cyclical Unemployment.” Journal of Political Economy, August 1982, 90(4), pp. 777-93. Lucas, Robert E. Jr. and Prescott, Edward C. “Equilibrium Search and Unemployment.” Journal of Economic Theory, February 1974, 7(2), pp. 188-209. Neelin, Janet. “Sectoral Shifts and Canadian Unemployment.” Review of Economics and Statistics, November 1987, 69(4), pp. 718-23. Padoa-Schioppa, Fiorella, ed. Mismatch and Labour Mobility. Cambridge, UK: Cambridge University Press, 2008a. Padoa-Schioppa, Fiorella. “A Cross-country Comparison of Sectoral Mismatch in the 1980s,” in Fiorella PadoaSchioppa, ed., Mismatch and Labour Mobility. Cambridge, UK: Cambridge University Press, 2008b, pp. 1-43. Petrongolo, Barbara and Pissarides, Christopher A. “Looking into the Black Box: A Survey of the Matching Function.” Journal of Economic Literature, June 2001, 39(2), pp. 390-431. Şahin, Ayşegül; Song, Joseph; Topa, Giorgio and Violante, Giovanni L. “Mismatch Unemployment.” Staff Report No. 566, Federal Reserve Bank of New York, August 2012; Updated September 2012; http://www.newyorkfed.org/research/economists/sahin/USmismatch.pdf. Shimer, Robert. “The Cyclical Behavior of Equilibrium Unemployment and Vacancies.” American Economic Review, March 2005, 95(1), pp. 25-49. Shimer, Robert. “Mismatch.” American Economic Review, September 2007, 97(4), pp. 1074-101. U.S. Department of Labor, Bureau of Labor Statistics. “How the Government Measures Unemployment.” October 16, 2009; http://www.bls.gov/cps/cps_htgm.htm. Federal Reserve Bank of St. Louis REVIEW May/June 2013 271 272 May/June 2013 Federal Reserve Bank of St. Louis REVIEW