The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 26, No. 4 ECONOMIC REVIEW 1990 Quarter 4 Bank Capital Requirements and Leverage: A Review of the Literature 2 by William P. Osterberg Expectations and the Core Rate of Inflation 13 by Richard H. Jefferis, Jr. The Case of the Missing Interest Deductions: Will Tax Reform Increase U.S. Saving Rates? by David Altig FEDERAL RESERVE BANK OF CLEVELAND 22 1990 Quarter IV Vol. 26, No. 4 Bank Capital Requirements and Leverage: A Review of the Literature 2 by William P. Osterberg Requiring banks to increase their capital-asset ratios continues to be viewed as a policy that would improve the safety of the com mercial banking system. However, relatively little is known about how banks adjust to increased capital requirements. This paper reviews the existing literature on the subject and addresses a key complication: the need to disentangle the influences of market and regulatory forces on banks' capital decisions. In order to illustrate the interaction between these forces, the author also presents a model of a bank’s choice of optimal leverage. Expectations and the Core Rate of Inflation 13 by Richard H. Jefferis, Jr. Inflation rates associated with different price series are both volatile and weakly correlated, properties that make realized inflation an un attractive guide for monetary policy. In contrast, the expected infla tion series generated by a wide variety of econometric models are less volatile than actual inflation and are highly correlated. This cor relation suggests that the different series are tracking a common trend, or core rate, and makes expected inflation a suitable bench mark for monetary policy directed toward controlling inflation. The Case of the Missing Interest Deductions: Will Tax Reform Increase U.S. Saving Rates? 22 by David Altig As of the coming tax year, U.S. taxpayers may no longer deduct per sonal interest expense when calculating taxable income. Will this change, resulting from the Tax Reform Act of 1986, increase the saving rate in the nation? This paper suggests that the answer is yes: An examination of private saving rates among several OECD countries shows that saving rates are, on average, higher in coun tries that have not historically subsidized borrowing through interest deductibility. The author also finds that the divergence of U.S. and Canadian saving rates over the past several decades appears to be significantly related to differential tax treatment of interest expense. Economic Review is published quarterly by the Research Depart ment of the Federal Reserve Bank of Cleveland. Copies of the Review are available through our Public Affairs and Bank Relations Depart ment, 216/579-2157. Coordinating Economist: Randall W. Eberts Editors: Tess Ferg Robin Ratliff Design: Michael Galka Typography: Liz Hanna Opinions stated in Economic Review are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted provided that the source is credited. Please send copies of reprinted material to the editors. ISSN 0013-0281 Bank Capital Requirements and Leverage: A Review of the Literature by W illia m P. Osterberg William P. Osterberg is an economist at the Federal Reserve Bank of Cleveland. The author would like to thank Randall Eberts, James Thomson, and Haluk Unal for helpful comments and suggestions, and Kyle Fleming for valuable research assistance. Introduction Recognition of the extensive losses inflicted on the Federal Savings and Loan Insurance Corpo ration (and thus indirectly on the taxpayers) by the thrift industry crisis has led to increased scru tiny of the safety and soundness of commercial banks and other financial institutions. It has become obvious that some of the factors respon sible for excessive risk-taking by savings and loans may also be relevant to commercial banks. In particular, the current system of fixed-rate deposit insurance and supervision and regula tion interacts in complex ways with the market forces that may ordinarily discipline banks. An understanding of these interactions is crucial to financial institution reform. This article reviews the literature relevant to assessing one proposed regulatory reform — increased capital requirements for banks. The arguments for higher capital requirements rely primarily on the premise that they will strengthen market discipline, and secondarily on the desire to provide a greater cushion for the deposit insurance agency. In theory, increased capital requirements can at least partially compensate for the weaken ing of market discipline that may result from the continued presence of fixed-rate deposit insur ance. However, the magnitude of the impact of past changes in capital requirements on banks’ capital decisions is unclear, mainly because mar ket forces also affect such decisions. Any analysis of the impact of capital require ments must take into account the current system of fixed-rate deposit insurance. The presence of non-risk-related insurance premiums and govern ment guarantees influences the capital decision by blunting the effect that increased capital-asset ratios would otherwise have on banks’ cost of funds. The current system complicates identifying the impact of changes in required capital-asset ratios, because the subsidy itself may be influ enced by the ratios and other factors. Fixed-rate insurance is also widely viewed as subsidizing risk-taking, thus providing a ra tionale for capital regulation. In the absence of government guarantees, shareholders would need higher levels of capital as a buffer against losses in order to avoid risk-related increases in their cost of funds. These guarantees thus lead to a substitution of deposits for equity, thereby lowering capital ratios. Although this distortion has led to reform pro posals that emphasize reductions in the scope of government guarantees, proposals to increase capital requirements continue to emerge despite the introduction of risk-based requirements. This has occurred in part because changes in capital requirements are seen as relatively easy to imple ment. In addition, as noted above, capital 3 requirements may induce shareholders to evaluate risk more carefully and to submit to the market’s evaluation when they attempt to raise capital. O n the surface, capital requirements seem to be effective, because almost all banks increase their capital-asset ratios (book value) after the requirements are increased. However, other fac tors may influence the ratios, especially if they are calculated in terms of book value. For exam ple, suppose that the regulatory standards were increased in response to a general market per ception that capital is inadequate. In this case, the subsequent adjustment may be partly due to banks’ desire to avoid an increase in the risk premium in their cost of funds. Clearly, in order to disentangle such influences, investigators must have a model of the factors that determine bank capital-asset ratios. The remainder of this paper is organized as follows. Section I reviews the theoretical litera ture on the determination of banks’ capital struc ture. Section II covers a closely related topic, the impact of capital requirements on portfolio risk. Section III presents a model of a bank’s choice of capital structure. Section IV analyzes the model’s implications for the impact of market and regula tory forces on bank leverage.1Section V reviews the empirical findings on the effects of capital requirements, contrasting various results in terms of the implications presented in section IV. Sec tion VI presents suggestions for future research and concludes. I. Optimal Capital Structure Theory for Financial Institutions I first discuss the theoretical findings relevant to nonfinancial corporations, since to some extent these may extend to banks, and then review the limited number of analyses of how and why the capital structure decisions of banks may differ from those of nonfinancial institutions.2 I then focus on specific analyses of banks’ capital ■ 1 Leverage is often defined as the ratio between debt and equity, measured in book or market values. In the model presented here, the bank chooses the level of promised payments to depositors, given an exoge nous asset portfolio. This is equivalent to choosing the debt-to-equity ratio directly. In Osterberg and Thomson's (1990) empirical study, the measure of leverage is the ratio between the book value of debt and the total of the book value of debt and the market value of equity. This is close to another often-analyzed measure of leverage, the debt-to-asset ratio. structure, most of which assign a prominent role to deposit insurance. The literature analyzing the capital structure decisions of nonfinancial corporations is so broad as to defy easy description.3 However, one of the strongest conclusions to emerge from the empirical studies is that optimal capital struc ture is influenced by the tax code, possibly in combination with leverage-related costs. For example, the ability of corporations to deduct interest on debt may encourage an increase in leverage. O n the other hand, higher levels of non debt tax shields, such as depreciation and tax credits, may reduce optimal leverage by increas ing the probability that not all interest expenses will be deductible. Taxes on personal equity and interest income may also decrease the net tax advantage of debt and optimal leverage. Leverage-related costs include the expected costs of bankruptcy, and agency costs associated with conflicts among creditors, stockholders, and managers. The direct costs of bankruptcy are mini mal (mainly involving administrative and legal fees), but agency costs, which include any decrease in firm value associated with contractual arrange ments to protect one party from actions taken by another party with conflicting interests, can be sig nificant. Bond covenants that restrict cash-flow usage may impose agency costs. However, such covenants may be a part of optimal contracts recon ciling bondholders with stockholders. The theory of optimal financial structures for financial intermediaries differs somewhat from the theory for nonfinancial firms. First, in analyz ing capital structure for either financial or non financial firms, it is convenient to assume that operating and financing decisions can be separated. This assumption is harder to defend for financial intermediaries. The existence of complete markets, which makes separation ■ 2 In this section, we review the theoretical analyses relevant to un derstanding the impact of changes in bank capital requirements. Because few analyses of bank capital structure are available (relative to the number that deal with the capital structure of nonfinancial institutions), it is not useful to attempt to categorize various approaches. In addition, the dis sim ilarities in approach prevent the development of a general model to which all others specialize. ■ 3 Among several useful surveys is one by Harris and Raviv (1990), who categorize the forces that may influence capital structure into desires to 1) ameliorate conflicts of interest, 2) convey private information to markets, 3) influence product or input markets, and 4) affect corporate control contests. The authors exclude tax-driven theories that they admit are of great empirical importance. Although few analyses of the capital structure of banks consider these four forces, several take into account taxes and other considerations discussed here. more likely, makes it difficult to explain the exis tence of intermediaries: If markets were com plete, lenders and borrowers could transact without them. In addition, deposits seem to play a role in both the real and financial decisions of banks, because deposits are not only an input into banks’ production, but a component of debt in their capital structure. Another reason that analyses of banks’ capital structure differ is that regulatory forces aimed directly at capital struc ture (for example, capital-asset ratios) must be considered. Although most studies of the impact of cap ital requirements on banks do not view these institutions as fundamentally different from non financial entities, many others have examined the role of informational asymmetries and con tracts in explaining the existence of intermedi aries. Early examples are Boyd and Prescott (1986) and Diamond (1984). Sealey (1985) ana lyzes a model of incomplete markets and inter mediaries, showing the conditions under which shareholder unanimity holds and under which unanimity implies separation. Sealey (1983) ex amines a model of incomplete markets in which economies of scale in the provision of deposit services influence bank leverage. Chen, Doher ty, and Park (1988) utilize an option-pricing framework to analyze the capital structure deci sions of depository financial intermediaries in the presence of deposit insurance, reserve requirements, liquidity effects, and taxation. They conclude that no clear separation exists between operating and financial decisions, and that this finding even applies to analyses of the impact of taxation on leverage decisions. As noted in Santomero (1984), most studies of bank capital structure assume that real and financial decisions can be separated, and try to explain leverage choice conditional on a given portfolio of assets. One example is Orgler and Taggart (1983), who show how personal and cor porate taxes, reserve requirements, and econ omies of scale influence intermediaries’ optimal leverage. Applications of the option-pricing framework also assume that portfolio composi tion is held constant. Pyle (1986) shows that the use of book values in capital regulation is inap propriate when combined with closure rules that deviate from an economic solvency condition. The conclusions of theoretical analyses of the impact of capital requirements are closely related to the treatment of deposit insurance and government guarantees. If deposit insurance is underpriced and unresponsive to risk, then stock holders are being subsidized by the insurer, and the size of the subsidy is a function of portfolio risk and leverage. This subsidy has a direct im pact on banks’ responses to changes in capital requirements. Buser, Chen, and Kane (1981) ex amine how the combination of capital regulation and flat-rate deposit insurance jointly influences bank leverage. They note that because capital regulation encompasses more than just numerical standards for capital-asset ratios, such regulation can be seen as imposing an implicit risk-related insurance premium that discourages banks from exploiting the subsidy implied by flat-rate deposit insurance. II. The Impact of Capital Requirements on Portfolio Composition Although most studies of bank capital structure assume a given portfolio of assets, several authors have examined the impact of capital re quirements on portfolio risk, assuming that leverage is at the regulatory maximum.4 An overall assessment of the impact of capital re quirements on bank capital structure would have to allow for possible feedback from varia tion in portfolio changes. Koehn and Santomero (1980) conclude that increased numerical capital requirements lead banks that are risk-averse expected utility maxi mizers to reshuffle their portfolios so as to in crease the probability of bankruptcy. Lam and Chen (1985) and Kim and Santomero (1988) use similar approaches. Keeley and Furlong (1987), who employ a value-maximization framework, point out that Koehn and Santomero ignore the impact of changes in leverage and portfolio risk on the deposit insurance subsidy. Osterberg and Thomson (1988) show how the impact of capi tal requirements on portfolio shares is altered by allowing the cost of funds to be influenced by leverage. ■ 4 Flannery (1989) shows why insured banks may have a preference for safe individual loans but still prefer risky overall portfolios. Capital ade quacy standards and loan examination procedures are key elements of his analysis. Lucas and McDonald (1987) study the impact of capital regula tion on bank portfolio choice when banks have private information about loan quality. EQUATIONS (1) AN A A = (X - - Y) (1 --tc) + cp (l) A 9+ 1 H 1 = (1- A.)[(X 81- if X > Y + A if Y + * < x ■ *c A A = ( 1 - X) (X - Y) if Y < X < Y + = 0 if X < Y A 5 - (p A A = Y if Y < X = X (1 - k ) if 0 < X < Y A otherwise. where X = end-of-period value of bank assets, Ys , Yb — gross end-of-period cash flows accruing to bank stockholders and depositors, respectively, A Y = total end-of-period promised payment to depositors, cp = total end-of-period after-tax value of nondebt tax shields when fully utilized, X = regulatory penalty, 8 = capital requirement, k = cost of financial distress to depositors, and d = proportion between 8 and Y (a capital requirement proxy). III. A Model of Market and Regulatory Influence on Bank Capital Structure To aid in this review, I present a model in which market and regulatory influences on banks’ cap ital structure are intertwined. The model also provides a limited synthesis of the theoretical lit erature. However, influences that could explain the existence of intermediaries, such as the pres ence of incomplete markets, are not incorporated. The only factor included that distinguishes banks is capital regulation. In addition, the model main tains the separation of real and financial decisions by holding constant the bank’s asset portfolio and return variance. Although I initially assume that there is no deposit insurance, such insurance is easily introduced (see Osterberg and Thomson [1990] and the following discussion).5 Equations (1) and (2) describe the uncertain outcomes facing stockholders and depositors. I view the bank as attempting to maximize the total of the values of equity and deposits, each of which depends on the uncertain outcomes and their associated probabilities. Pre-tax returns to stockholders depend on the uncertain end-of-period value of bank assets, X. The first line of equation (1) indicates the return when income (asset values) is high enough that the capital guideline is not violated. I assume in this case that all nondebt tax shields can be util~ A ized (X > Y+ (p/ic); however, the results are not significantly affected by this assumption. In the second case, when income is high enough to use all the shields but the capital requirement is not met {X[\- tc]+ cp - Y < 8 ), regulators impose a tax of X on stockholder returns. In the third case, income is positive but insufficient to utilize nondebt tax shields, and the guidelines are not met. ■ 5 This model is a variant of the one developed by Bradley, Jarrell, and Kim (1984), hereafter referred to as BJK. Detailed assumptions un derlying the model are given in appendix 1. Equation (2) indicates the end-of-period pre tax flows to depositors. A crucial distinction between stockholders and depositors is readily apparent: Depositors only receive Y, even if income greatly exceeds promised payments. O n the other hand, if income is positive but in sufficient to meet promised payments, the bank is in financial distress and incurs real costs that reduce the return to depositors by the fraction k. The bank is assumed to know 1) the relevant tax rates, 2) the amount of nondebt tax shields, 3) the required capital-asset ratio, d, 4) the regu latory response, embodied in X, 5) the costs of financial distress, 6) the average income, X, and 7) the standard deviation of income, (p. The bank A chooses Y to maximize the market value of its debt plus equity (see appendix 1). Equation (3) is the derivative of V with respect to Y: V*(dV/dY). (3) 1 - t.p b A A A V£ = ----— [ l - F ( Y ) ~ k Y f ( Y ) ] 1 - tp s A r CP A [ - ( l - f c ) [ l - F ( y + Y)\ (P [F(Y + j A )-/r(K)]-X{ , A Cp [F(Y+ j ) <p A Cp d ( 8 - cp) - f < r + 7 )]+ [8 + - 7 3 7 — ) A 8 - cp , where F ( ) is the cumulative probability density function of X . If banks in fact choose Y so as to satisfy equation (3), then this expression indi cates how both market and regulatory forces influence bank leverage. If X = 0, the model’s implications are consis tent with theories of optimal capital structure in which the assumed tax advantage of debt bal ances the expected cost of bankruptcy (see BJK). These implications are as follows. First, an in crease in f raises optimal leverage by increas ing the cost of equity. Analogous reasoning implies that an increase in tpb reduces optimal leverage. Second, an increase in tc raises opti mal leverage by increasing the tax advantage of debt. For this reason, an increase in cp reduces op timal leverage by increasing the probability that not all interest expenses will be deductible. Third, an increase in k reduces optimal leverage by increasing the expected cost of a bank’s inability to make all promised payments. The model is also consistent with theoretical approaches that assign deposit insurance a role in distorting market discipline. The effect of fixedrate deposit insurance on optimal leverage can be seen by comparing equation (3) with under full insurance (see Osterberg and Thom son [I99O]). Optimal leverage is higher with fixedrate deposit insurance by the amount (1 ) [F (Y) + kYf\Y)VrQ. Fixed-rate deposit insur ance increases the optimal Y by insuring that depositors are always paid in full and by shifting the cost of financial distress from depositors to the Federal Deposit Insurance Corporation. In the context of the model presented above, these two^influences are equivalent to assuming that F ( Y ) = 0 and k = 0. Although this model does not allow the higher leverage to influence bank riskiness, the increase in leverage induced by deposit insurance provides a rationale for capital regulation. The impact of capital regulation on leverage can be seen by examining the last term in equa tion (3), X (• ). The first two components of X ( •) comprise the expected after-tax regulatory penalty resulting from issuing the last dollar of deposits. As equations (1) and (2) demonstrate, the possibility of a regulatory penalty affects the return to equity, which one would expect to be reflected in the rate of return demanded by stockholders and thus in the bank’s leverage decision. In fact, the last component of X ( ■) is the increase in the cost of equity capital that results from issuing one more dollar of deposits, [8 + ¿ ( 8 - c p ) / ( 1 - tc )] f [ K + ( 8 - c p ) / ( l - t c )] Because all of the components are positive, the possibility of a regulatory penalty reduces a bank’s optimal leverage. IV. The Impact of Regulatory and Market Forces on Optimal Leverage Although equation (3) clearly shows that both market forces and regulatory variables influence leverage with signs consistent with theory, it is more important for our purposes to note that this expression also implies that the impact of an increase in X (the regulatory penalty) on leverage depends on market forces entering X ( •). Empirical studies of capital requirements vary in their treatment of the influence of such market forces ( cp, k, tc, tf)S, tph, and o, where a is the standard deviation of X ) on Y. To show how market influences affect the impact of capital regulation on bank leverage, one can differentiate the optimality condition (equation 131) with respect to the regulatory vari ables. The derivatives with respect to the marketforce variables are indicated in appendix 2. Further details can be found in Osterberg and Thomson (1990). Equation (4) gives the impact of a change in d on optimal leverage. The ratio d is closely related to a required capital-asset ratio, because it is the minim um level of the end-of-period equity value and because 5 = Yd. a 5-9 f ( Y + 1— “ ) { 25 —cp —[ F5 + (F+ —- 5(5-9) 1 - i. — - X ) / o 2} ^ 0 The impact of d on leverage clearly depends on market forces, implying that such forces in fluence leverage even if a bank fails to meet the guidelines. As discussed below, some studies imply that such banks are influenced only by regulation, while banks meeting the guidelines are influenced only by market forces. No such dichotomy emerges here. Equation (4) implies that V$d is negative whenever X > Y+ (5 - cp ) / ( l - tc); that is, an increase in d reduces leverage when the bank expects to meet the capital requirements. How ever, if a bank does not expect to meet the re quirements, an increase in d may induce it to increase leverage and thus move even further below the guidelines. Equation (5) shows that an increase in the regulatory penalty, A,, reduces bank leverage. Here, as in the response of leverage to d, the impact of capital regulation depends on market factors. Equation (6) shows that an increase in the costs of financial distress, k, also reduces optimal leverage. Although k is referred to above as a market factor, the cost of financial distress can be influenced by regulatory policies pertaining to bank closure. (5) Vfa = 1 - 1p s - A Cp F {Y + j)} ¿ /(5 -c p ) A 5 — cp + (8 +^ T ^ ) / ( V +-r r r )] < 0 1 - t.p b (6) Vpk = - A [y f{ A y )] < o V. Evidence on the Impact of Capital Requirements on Bank Leverage Separating market forces from regulatory forces has been a major difficulty in ascertaining the effectiveness of capital guidelines. Dietrich and James (1983) criticize earlier studies by Peltzman (1970) and Mingo (1975) for ignoring depositrate ceilings in their analyses of the impact of capital requirements. Under such ceilings, banks can influence risk-adjusted returns on bank debt by augmenting capital. However, only under less-than-full deposit insurance would more capital benefit stockholders, by inducing unin sured depositors to accept lower interest rates. Dietrich and James conclude that the guidelines have no effect on bank capital changes. Although the model presented here does not directly consider the possibility of interest-rate ceil ings, capital levels influence the returns to stock holders and thus the rate of return required on equity. The latter can be calculated as the ratio between E ( Y ) , the expected returns to stock holders, and S, the market value of equity (see appendix 1 and BJK). Equation (1) indicates that returns to stockholders are influenced by several market forces that must be controlled for in any analysis of the impact of capital requirements. Marcus (1983), Wall and Peterson (1987), and Keeley (1988a, 1988b) examine bank holding companies rather than independent banks. Wall and Peterson apply a switching regression tech nique to movements of equity values in an attempt to distinguish a regime in which capital ratios exceed the requirements (and are thus influenced by market forces) from a regime in which ratios are at the regulatory limit. They conclude that most banks are influenced by regulation. The model presented here implies that 1) banks may respond to market forces even if the guidelines are not being met and 2) regulatory forces may influence leverage even if the bank exceeds the guidelines. In addition, equation (4) indicates that banks below the guidelines may actually respond to stiffer requirements perversely. Keeley (1988a) examines the response of bank holding companies to the increased capital requirements of the 1980s. Although capitaldeficient banks increased their book-value ratios more than capital-sufficient banks did, market ratios increased for both classes. However, regu latory subsidies or taxes can influence the response of market-value ratios to increased capi tal guidelines, because the value of the subsidy may vary with leverage or risk. Keeley (1988b) claims that increased competition erodes the value of bank charters and thus raises incentives to increase leverage or to reduce capital ratios. Marcus (1983) utilizes a time series crosssectional approach, measuring regulatory pres sure to increase capital by the holding company’s capital ratio relative to the average (in terms of book or market value). He finds that the incen tive to decrease capital varies positively with the level and variability of interest rates, as well as with the tax disadvantage of equity finance. Regulation seems to have no effect. However, his regulatory measure does not incorporate risk. In the model presented above, d is close to a statutory capital-asset ratio. However, analyz ing banks’ capital ratios relative to the average may be a more useful way to isolate the impact of capital regulation. There are at least two rea sons for this. First, relatively few banks are below the statutory guidelines. Second, evi dence suggests that capital regulation is based on a peer-group standard. In fact, a peer-group capital standard may be a useful proxy for the regulatory penalty variable, X. The relevance of taxes to the capital struc ture of banks is discussed in more detail by Wall and Peterson (1988) and Gelfand and Hanweck (1987). Wall and Peterson argue that taxes do not influence the capital structure of banks affil iated with holding companies, because the tax consequences of the parent issuing debt to buy subsidiary equity are similar to those ensuing when the bank itself issues debt. Gelfand and Hanweck examine the financial statements of 11,000 banks and find strong evidence for mar ket influences (tax rates, risk, and municipal securities [munis] as proxies for nondebt tax shields) on leverage. Osterberg and Thomson (1990) investigate the influence of capital regulation on bank hold ing company leverage empirically, drawing on the implications of the model presented above. The authors find that market forces influence leverage through three channels: a direct chan nel, a channel in which market forces interact with risk ( a ), and a channel in which market forces interact with capital regulation. In addi tion, their analysis explicitly allows for the simul taneous determination of leverage and muni holdings. Although the latter may no longer be an important channel through which banks manage their tax liability, this may not have been the case during the period examined (1986-1987).6 The interactive capital regulation measures, taken as a whole, are significant, as are the interactive risk measures. In addition, muni holdings appear to be significant deter minants of leverage, as do market forces. VI. Conclusions and Suggestions for Future Research This article reviews the literature relevant to assess ing the impact of increased bank capital require ments. Although researchers have suggested various proposals to correct the distorted incen tives facing bankers, raising required capital ratios continues to emerge as a possible means of strengthening market discipline. However, pre vious studies have failed to clarify the impact of numerical guidelines on banks’ capital-asset ratios. The primary difficulty in discerning the influ ence of such guidelines lies in disentangling the impacts of regulatory and market forces. In order to illustrate the way in which these forces interact, I present a model of a bank’s choice of leverage ratio where, in the absence of capital regulation, tax considerations and bankruptcy costs imply an interior solution. W hen capital regulation is introduced, it becomes clear that the impact of such regulation depends on market forces. These results may provide useful insight for regulators. For example, the response of bank leverage to capital regulation may depend on ■ 6 Scholes, Wilson, and Wolfson (1990) present evidence that banks’ muni holdings responded to changes in the tax code between 1983 and 1987, and that capital regulation seemed to influence banks' timing of capital loss realization. This seems to suggest that capital regulation and the tax code interact in a manner sim ilar to that suggested inthisoaner. the market factors considered in this paper, such as tax rates, nondebt tax shields, and muni hold ings, not just on the capital position of the bank. This implies that evaluations of banks’ leverage and capital-asset ratios should take into account market influences on the leverage decision. The model may also explain previous empiri cal findings regarding the impact of capital requirements. Most studies do not control for many of the market influences on banks’ capital decisions. The analysis presented here thus implies that theoretical examinations of bank capital structure may further improve our under standing of the influence of capital require ments. In this regard, it may be particularly useful to analyze capital requirements through models that incorporate informational asymme tries and market imperfections to explain the existence of financial intermediaries. Appendix 1 11. The capital constraint, 5 , is not binding unless X is such that the tax shields are being fully utilized. Assumption 10 allows us to separate the ef fects of capital requirements from the effects of deposit insurance. Thomson (1987) shows that this is equivalent to assuming 100 percent deposit insurance if the insurance is fairly priced. The case in which all liabilities are covered by fixedrate, zero-premium deposit insurance is analyzed in appendix A of Osterberg and Thomson (1990). Assumption 11 is made for convenience only; my results are not materially affected by the al ternative assumption that 8 is binding for values of X where (p > (X - Y) tc . Under the assumption of risk neutrality, and given the uncertain outcomes detailed in the text, the after-tax market value of the banking firm is the sum of the market values of deposits and equity: (ia ) v = j r [ ] i ( i - /„ ) [ ( £ - i b u - O 0 Detailed Assumptions and Structure of the Model m t + <p]+ ( l - t pb) Y \ f ( X ) d X izi 1 -/ The main assumptions of the model presented in the text are as follows: 1. Investors are risk-neutral. 2. The personal tax rates on returns from bank debt and bank equity are tph and tps, respectively. 3. Bank income is taxed at the corporate rate, tc . 4. All taxes are levied on end-of-period wealth. 5. The firm’s end-of-period tax liability can be reduced through nondebt tax shields, (p , such as investment tax credits and depreciation. 6. Unused tax credits cannot be transferred across time or across firms. 7. If banks cannot meet their end-of-period promised payments to depositors, Y, costs of financial distress are incurred that reduce bank equity value by a factor of k. 8. The end-of-period capital requirement is 8 = Yd. 9. If X - Y< (S - cp ) / l - tc), a regulatory penalty reduces stockholders’ returns by a con stant fraction X ( X is the end-of-period value of assets). 10. All bank liabilities are uninsured deposits. - J £+s> / n o - o + q) ] f ( X ) d X t + J [(1 - $ » ) ( ! - X ) ( i - Y) Y + ( 1 - f ^ ) Y if(X )d X A Y + \ ( l - t pb) { l - k ) x f ( x ) d x ] , 0 where f ( X ) = the probability density of X , and rQ= one plus the rate of return on a risk free tax-exempt bond. The four integrals in equation (1A) are, respectively, 1) the expected value of the bank over the range of X where the bank fully utilizes its nondebt tax shields, 2) the expected value of the regulatory tax over the range of X where the bank fully utilizes its nondebt tax shields but fails to meet its capital guideline, 3) the expected value of the bank over the range of X where nondebt tax shields are no longer fully utilized, and 4) the expected value of the bank when X is not large enough to meet promised payments to the depositors and k percent of the firm value is lost to financial distress. Appendix 2 The Impact of Market Forces on Optimal Bank Leverage The effect of an increase in nondebt tax shields, q>, on optimal leverage is indicated by equation (2A). when X > Y + (8 - <p ) / ( l - tc ) arises be cause the capital constraint is assumed to be binding when the bank’s net tax bill is positive. There are two offsetting effects. First, an in crease in tc raises the value of the interest deduction on debt, which induces the bank to issue more deposits. This is the familiar effect discussed in the finance literature on optimal capital structure for nonfinancial entities. The second effect is a reduction in the after-tax value of equity and an associated increase in the probability that the bank will violate the capital constraint and reduce leverage. (3A ) 1 - t. Vp,= — ^ { \ - F ( Y + j ) 'o (2A) i/ a Yip l - ‘ps = lc A 8 - cp + \[f ( y + j r y ) - [(1+ X ) f ( y + S ) a 9 F(<Y + 7 > ] , r a 8 9 , d + Xf(Y + - — r ) { 1 + 1 - tc 1 - tc -[8 + ¿(5-q>) r d X (S - d ( 8 - cp ) erage increases if X > Y + ( 8 - cp )/( 1 - tc). This possibility is created by the combination of the capital requirement being based on the after tax value of equity, which includes the value of the shields, and the fact that the capital require ment is binding when the tax shields are being fully utilized. For high-enough values of X , an additional dollar of tax shields reduces the prob ability that the bank will violate the capital con straint and incur the regulatory penalty. The effects of changes in the various tax rates on the optimal level of debt are shown in equations (3A), (4A), and (5A). In equation (3A), the response of bank leverage to an in crease in the marginal corporate tax rate is posi tive when X > Y + (8 - (p)/(l - tc ) . In other words, if expected end-of-period income is large enough to meet the capital requirements, then an increase in tc reduces the optimal level of debt. The ambiguous sign for equation (3A) . 18 + lc If there were no regulatory penalty ( X = 0 ), I would obtain the same results as BJK; that is, a higher level of nondebt tax shields would reduce leverage ( < 0). Here, however, lev \ + [<1 + i 3 T 1- h 1 <P) 1- L 8 - cp _ U r+ j . , - * ) ] a 5 — cp , ■ A Y + T— )} ^ 0 c If there were no costs of financial distress (k —0), equation (4A) would be unambiguously negative at the optimal level of debt. In addition, if all of the bank’s deposits were insured, V<st would be clearly negative. However, more ph generally, equation (4A) is negative when the probability that Y is less than X exceeds the marginal expected leverage-related costs. This result is similar to the findings of BJK. Note that we have assumed that the costs of financial dis tress facing the depositors (k ) are distinct from the regulatory penalty. As in BJK, Vfa is unam ps biguously positive. However, here the response depends on the regulatory penalty, X. (4A) Vfr = - — [1 - F ( Y ) - k Y f(Y ) ] < 0 pb rn (5A) { a -< c) t i - f ( 7 + ^ ) ] = - 7 Ps References Boyd, J., and Edward Prescott. “Financial Inter- + [ F ( f + y ) - F (Y) ] + X [[F(Y+ j ) - mediary-Coalitions,” Jo u rn a l o fEconom ic Theory, vol. 38, no. 2 (April 1986), pp. 22132. F (Y) ] C a 8 - (p A <P + ( 1 - tc ) l F ( Y + T— )- F(y+ - ) ] «c d ( 5 - cp) A c Ô - (p + (8 + - r T^ ) / ( r + — -, ! Bradley, Michael, Gregg A. Jarrell, and E. Han Kim. “O n the Existence of an Optimal Capital Structure: Theory and Evidence,” Jo u rn a l o f Finance, vol. 39, no. 3 (July 1984), pp. 857-80. )]}>o Buser, Stephen A., Andrew H. Chen, and Ed ward J. Kane. “Federal Deposit Insurance, Finally, the optimal level of deposits is a func tion of the variability of X. Equation (6A) shows that an increase in a has an ambiguous effect on optimal leverage. The sign on depends on the proximity of Y, Y + (Ô - cp ) / ( I — tc ), and F+ c p /ic to the mean of X , as well as on the magnitudes of k, cp, d, and X. BJK find that, even without a regulatory penalty, the impact of an increase in a on F is ambiguous. Regulatory Policy, and Optimal Bank Capi tal,” Jo u rn a l o f Finance, vol. 35, no. 1 (March 1981), pp. 51-60. Chen, Andrew H., Neil A. Doherty, and Hun Y. Park. “The Optimal Capital Structure Deci sion of Depository Financial Intermediaries: A Contingent-Claim Analysis,” Research in Finance, vol. 7 (1988), pp. 91-111. Diamond, Douglas W. “Financial Intermedia tion and Delegated Monitoring,” Review o f Econom ic Studies, vol. 51, no. 3 (July 1984), pp. 393-414. I - t.pb (6A) r,X5 ( Y - X ) - k Y [ ( ^ - ^ ) 2- U 1 / ( F ) 1 - t. + I [ (1 — X ) t — 2X ] i rAo , <P (7 + / ( K A — y- x ) y ) - ( l - X )f(Y )(Y - X ) • ( £ + f — - * ( - [ 8 + d(?> ~ 'P) ] 1- t 1 - t. [ <— 6 - cp — “Regulation and the Determination of Bank Capital Changes: A Notq ,”Jo u rn a l o f Finance, vol. 38, no. 5 (December 1983), pp. 1651- 58 . Flannery, Mark J. “Capital Regulation and In + ¥ ( y + jz 7X1-U A Dietrich, J. Kimball, and Christopher James. _ ) 2- n } ^ 0 sured Banks’ Choice of Individual Loan De fault Risks "Jo u rn a l o f M onetary Economics, vol. 24, no. 2 (September 1989), pp. 235-58. Gelfand, Matthew D., and Gerald A. Hanweck. “Tax Shields, Bankruptcy Risk, and O p timal Corporate Capital Structures: Evidence and Implications for Commercial Banking,” Washington, D.C.: Board of Governors of the Federal Reserve System, Committee on Bank ing and Financial Structure, November 1987. Harris, Milton, and Artur Raviv. “The Theory of Capital Structure,” Graduate School of Business, University of Chicago, W orking Paper No. 81, September 1990. IU Keeley, Michael C. “Bank Capital Regulation in the 1980s: Effective or Ineffective?” Federal Reserve Bank of San Francisco, Econom ic Review, Winter 1988a, no. 1, pp. 3-20. ________ . “Deposit Insurance, Risk, and Market Power in Banking,” Federal Reserve Bank of San Francisco, W orking Paper No. 88-07, September 1988b. Osterberg, William P., and James B. Thom son. “Optimal Financial Structure and Bank Capital Requirements: An Empirical Investiga tion,” Federal Reserve Bank of Cleveland, W orking Paper No. 9007, July 1990. ________ , and__________ .“Capital Requirements and Optimal Bank Portfolios: A Reexamina tion,” Federal Reserve Bank of Cleveland, W orking Paper No. 8806, August 1988. _______ , and Frederick T. Furlong. “Bank Capital Regulation: A Reconciliation of Two Viewpoints,” Federal Reserve Bank of San Francisco, W orking Paper No. 87-06, Sep tember 1987. Peltzman, Sam. “Capital Investment in Com mercial Banking and Its Relationship to Port folio Regulation,”Jo u rn a l o f P o litical Econ omy, vol. 78, no. 1 0anuary/ February 1970), pp. 1-26. Kim, Daesik, and Anthony M. Santomero. “Risk in Banking and Capital Regulation,” Jo u rn a l o f Finance, vol. 43, no. 5 (Decem ber 1988), pp. 1219-33. Pyle, David H. “Capital Regulation and Deposit Koehn, Michael, and Anthony M. Santomero. Santomero, Anthony M. “Modeling the Bank “Regulation of Bank Capital and Portfolio Risk/'Jo u rn a l o f Finance, vol. 35, no. 5 (December 1980), pp. 1235-44. Lam, Chun H., and Andrew H. Chen. “Joint Effects of Interest Rate Deregulation and Capital Requirements on Optimal Bank Port folio Adjustments,”Jo u rn a l o fFinance, vol. 40, no. 2 (June 1985), pp. 563-75. Lucas, Deborah, and Robert L. McDonald. “Bank Portfolio Choice with Private Informa tion about Loan Quality: Theory and Implica tions for Regulation,”Jo u rn a l o f B anking an d Finance, vol. 11, no. 3 (September 1987), pp. 473-97. Marcus, Alan J. “The Bank Capital Decision: A Time Series-Cross Section Analysis,”Jo u rn a l o f Finance, vol. 38, no. 4 (September 1983), pp. 1217-32. Mingo, John J. “Regulatory Influence on Bank Capital Investment,”Jo u rn a l o f Finance, vol. 30, no. 4 (September 1975), pp. 1111-21. Insurance,”Jo u rn a l o f B anking a n d Finance, vol. 10, no. 2 (June 1986), pp. 189-201. ing Firm: A Survey /'Jo u rn a l o f Money, Credit, a n d B anking, vol. 16, no. 4 (November 1984, part 2), pp. 576-602. Scholes, Myron S., G. Peter Wilson, and Mark A. Wolfson. “Tax Planning, Regulatory Capi tal Planning, and Financial Reporting Strategy for Commercial Banks,” Review o f F in a n c ia l Studies, vol. 3, no. 4 (1990), pp. 625-50. Sealey, C. W. Jr. “Portfolio Separation for Stock holder Ow ned Depository Financial Interme diaries,” Jo u rn a l o f B anking a n d Finance, vol. 9, no. 4 (December 1985), pp. 477-90. _ _ _ _ _ . “Valuation, Capital Structure, and Shareholder Unanimity for Depository Finan cial Intermediaries,”Jo u rn a l o f Finance, vol. 38, no. 3 (June 1983), pp. 857-71. Thomson, James B. “The Use of Market Infor mation in Pricing Deposit Insurance,” Jo u r n a l o f Money, Credit, a n d B anking, vol. 19, no. 4 (November 1987), pp. 528-37. Wall, Larry D., and David R. Peterson. “Capi Orgler, Yair E., and Robert A. Taggart, Jr. “Implications of Corporate Capital Structure Theory for Banking Institutions,”Jo u rn a l o f Money, Credit, a n d B anking, vol. 15, no. 2 (May 1983), pp. 212-21. tal Changes at Large Affiliated Banks,”Jo u r n a l o f F in a n c ia l Services Research, vol. 1, no. 3 (June 1988), pp. 253-75. ________ , and________ . “The Effect of Capi tal Adequacy Guidelines on Large Bank Holding Companies,”Jo u rn a l o f B anking a n d Finance, vol. 11, no. 4 (December 1987), pp. 581-600. Expectations and the Core Rate of Inflation by Richard H. Jefferis, Jr. Richard H. Jefferis, Jr., is a profes sor of finance in the School of Business Administration at the University of Colorado, Boulder, and was formerly a visiting schol ar at the Federal Reserve Bank of Cleveland. The author would like to thank Michael Bryan, William Gavin, Jeffrey Hallman, and James Hamilton for helpful dis cussions and advice. Introduction Policymakers seeking to control inflation are confronted by a bewildering array of price statis tics that often provide conflicting signals about the current inflation rate. The disparity among different measures of inflation is illustrated by fig ure 1, which depicts quarterly inflation rates im plied by movements in several well-known price series between 1954 and 1987, and by table 1, which displays the correlation among inflation rates associated with a broader group of indices over the same period.1 Although the CPI , the PCE deflator, and the PPI trend together, there is a wide variation in the movements of these price indices over periods as long as a quarter. The discrepancy among inflation rates asso ciated with different price indices has important implications for the conduct of monetary policy linked to inflation targets. If long-term increases in the price level are masked by statistical noise that is a consequence of changing circumstances in individual markets, then monetary policy linked to any index of current inflation will be affected by transient shocks as well as by the ■ 1 The price indices are the Consumer Price Index (CPI), the serv ice component of the CPI (CPIS), the Producer Price Index (PPI), the PPI without food and energy (PPIWF), and the Personal Consumption Expen diture (PCE) deflator. Only the CPI, PPI, and PCE deflator appear in the figure. http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis secular trend in prices. Although shocks to the price of individual commodities or groups of commodities do affect the cost of living, they do not necessarily reflect the impact of money growth on the price level. Nor is the appropriate policy response to these two types of inflation necessarily the same. Overall, both the source of noise and the amount of noise in individual price indices make them a poor choice for infla tion targets. A related problem associated with using real ized inflation as a guide for monetary policy is the timing of inflation signals. The inflation rate of last quarter or even last month is a poor guide for policy that seeks to influence the future course of the economy, yet this is the type of information provided by direct examination of any historical record. Forward-looking policies linked to a measure of current inflation should be based on what the past can tell us about the present and the future. To make the historical record useful, we need to extract from it infor mation about current inflation and expected future inflation. Inflationary expectations address both of these problems. Expectations are, by their nature, linked to long-term trends in the price level rather than to transient movements. They are forwardlooking. Moreover, there is a remarkable degree of correlation among the inflation forecasts ta FIGURE 1 Different Measures of Inflation, 1954-1987 Percent, a n n u a l rate 25 1954 1959 1964 1969 1974 1979 1984 SOURCES: U.S. Department o f Commerce, Bureau o f Economic Analysis, and U.S. Department o f Labor, Bureau o f Labor Statistics. generated by different price indices and forecast methodologies. Realistic models of inflation dis count current innovations in the inflation rate, which are largely noise, and focus instead on movements that tend to persist over time. As a result, different models of expected inflation agree on what is likely to occur in the immediate future, even when individual price series give conflicting signals about the current inflation rate. The common trend in the different series is an indicator of the pervasive price growth, or core inflation, that is of interest to policymakers. I. Expected Inflation We usually think of expected inflation as the ex pected rate of change in a particular price index, and judge different models of expectations by their ability to project movements in that index. The two criteria most commonly used to judge model performance are the mean squared error and bias of the inflation forecast. Both statistics are informative, since an unbiased forecast that fails to identify large, predictable movements in inflation will have a larger mean squared error than a forecast that is, on average, less accurate but better able to predict significant changes in the inflation rate. Two types of statistical models used to forecast inflation have, in the past, performed equally well in terms of both bias and mean squared error.2 Time series models identify tem poral patterns in the inflation rate and use those patterns, combined with information about the recent history of inflation, to predict future infla tion. These models are capable of identifying very complex relationships among inflation rates at different points in time, but tend to ignore other contemporaneous information that might be useful in forecasting. Econometric models that incorporate informa tion about interest rates or money growth attempt to remedy this shortcoming. Although the history of money growth is correlated with inflation, in terest rates often take the place of money in fore casting models. The motivation for this choice is the notion that, in an efficient capital market, the ■ 2 Fama and Gibbons (1984) compare pure time-series models and interest-rate models, and find that the interest-rate models yield a smaller root mean squared error in out-of-sample forecasts. The differences in forecast performance increase with the forecast horizon. TABLE 1 The Correlation among Quarterly Inflation Rates Based on Different Price Indices between 1954 and 1987 CPI CPIS PPI PPIW F PCED1 CPI 1.00 0.83 0.74 0.72 0.82 CPIS 0.83 1.00 0.48 0.56 0.56 PPI 0.74 0.48 1.00 0.79 0.76 PPIW F 0.72 0.56 0.79 1.00 0.78 PCEDEF 0.82 0.56 0.76 0.78 1.00 SOURCE: Author’s calculations. In this model, 7 ( 0 is the inflation rate at time t and e ( O is an impulse that affects that rate.3 Conceptually, the impulse comes either from expansion of the money stock or from some change in market conditions, such as a drought or the threat of war in the Middle East. The cur rent change in the inflation rate is determined by current and past impulses, where the weight assigned to the past is 0. These models have an appealing interpretation in terms of expected and unexpected inflation.4 From equation (1), we know that (2) A/(0 = e(0-8e(f-l). This implies that (3) J(O = 7 ( f - l ) - 0 e ( f - l ) or that inflation premium in nominal interest rates is a sufficient statistic for expected inflation. In practice, money may have some incremental predictive power, because the decomposition of nominal rates into an expected real return and an inflation premium is not observable, but is imposed on the data by the econometrician. To the extent that this decomposition is imperfect, the economet ric model will fail to uncover the market’s inflation forecast, even if movements in the nominal interest rate are completely determined by changes in the expected real rate and expected inflation, as theory would suggest. The merits of econometric models that extract inflation forecasts from interest rates and the empirical relevance of monetary growth for predicting inflation are issues that may be resolved only by examining the data. II. Time Series Models Time series models express current inflation as a weighted sum of past inflation and past changes in the inflation rate. The manner in which this history is translated into forecasts depends on the properties of the inflation process. W hen movements in the inflation rate tend to be transient, current innovations play a marginal role in the formation of expectations, and the historical record receives more empha sis in the inflation forecast. If, on the other hand, increases in inflation tend to persist, the infla tion forecast will be closely linked to the behav ior of prices during the recent past. A time series model of inflation that has been found to forecast well is http://fraser.stlouisfed.org/ (1 ) 7 (f) - IQ-l) = e (t) Federal Reserve Bank of St. Louis 0 8 (t—1 ) . (4) A 7(i) = A / ( / - 1) - 0 A e ( i - 1). Using the definition of A I ( t ) from equation (2) and the fact that A [0 e (t- 1) ] = 0 [ e ( t —1) - e (?- 2) ], we obtain (5) A 7 ( 0 = (1 - 0 ) e (t- 1). Expression (5) states that expected inflation follows a random walk, with an innovation var iance that is (1 - 0)2 times the variance of 8 (t). Values of 0 close to 1 imply that most of the variance in inflation is accounted for by tran sient shocks, so that current innovations are not reflected in expected future inflation, while values of 0 close to 0 imply that most of the variance is accounted for by movements in infla tion that are expected to persist.5 Estimation of equation (1) for the different series described in table 1 and figure 1 yields val ues of 0 that range from 0.45 for the PCE deflator to 0.70 for the PPI. 6 The evidence from the econ ometric model is therefore in accord with the in tuition suggested by the data: A modest fraction of the quarterly innovation in inflation is reflected ■ 3 The model described here is examined by Fama and Gibbons (1982). ■ 4 Jeffrey Hallman suggested this interpretation. ■ 5 Ansley (1980) provides an alternative interpretation that yields the same inference. ■ 6 Maximum likelihood estimates are based on a sample of inflation rates from the first quarter of 1954 to the fourth quarter of 1987. The Breusch-Pagan Lagrange multiplier test for autoregressive conditional heteroscedasticity (ARCH) effects reveals that the data are conditionally heteroscedastic. All estimates involve an ARCH (2,0) model of the condi tional variance, although this is found to have only a minimal impact on estimated parameter values and forecasts. TABLE 2 The Correlation among Expected Quarterly Inflation Rates Generated by a One-Parameter Time Series Model. Inflation Is Assumed to Follow an IMA (1,1) Process PPI PPIW F 0.92 0.93 0.90 0.90 1.00 0.82 0.81 0.71 0.93 0.82 1.00 0.93 0.89 PPIW F 0.90 0.81 0.93 1.00 0.87 PCEDEF 0.90 0.71 0.89 0.87 1.00 CPI CPIS CPI 1.00 CPIS 0.92 PPI PCED1 SOURCE: Author’s calculations. in expected future inflation. In the case of the PCE deflator, a 1 percent increase in quarterly in flation is associated with a 0.55 percent increase in expected inflation. That fraction is 0.30 in the case of the PPI. An alternative perspective on the estimated value of 0 is provided by examining the fraction of the variance of changes in quarter ly inflation accounted for by changes in expected inflation. This number ranges from 10 percent in the case of the PPI to 30 percent in the case of the PCE deflator. The effect of filtering the inflation-rate series with this model, and focusing on the expected inflation series implied by equation (3), is illus trated in table 2. The correlation among expected inflation rates inferred from the different price series is substantially greater than the correlation in realized inflation rates, even when expecta tions are generated by the parsimonious oneparameter time series model. For example, the correlation between the expected inflation rate inferred from the CPI and the expected inflation rate inferred from the PPI over 35 years of quar terly data is 0.93, while the correlation between the realized rates of inflation implied by these same indices is 0.74. Thus, the different price series yield highly correlated inflation forecasts, even though there is substantial disagreement about the current inflation rate among these series. III. Econometric Models Inflation forecasts based exclusively on the tem poral pattern of past inflation ignore a great deal of potentially useful data. Information about money growth or interest rates will be without value only in the event that the history of infla tion is a sufficient statistic for its expected future course. Both the tremendous amount of noise in the various inflation series and common sense suggest that this is unlikely. Nominal interest rates are an especially appeal ing source of information, since the yield on fixedrate debt instruments contains a premium that compensates the investor for expected deprecia tion in the purchasing power of money over the life of the instrument. The advantage of using interest rates to identify expected inflation, rather than modeling the link between money and prices directly, is that the inflation premium found in bond yields represents a consensus forecast of inflation over a fixed time interval known to the observer. In contrast, the history of money growth provides little information about when an increase in money will be reflected in prices, or even whether it will be reflected in prices rather than output. Focusing on bond yields rather than on money growth makes it unnecessary to consider the complex lag structures typical of macroeco nomic models that attempt to characterize directly the link between money and prices. Extracting inflationary expectations from bond yields is not a trivial exercise: Variations in nominal yields reflect changes in expected real returns as well as changes in expected inflation. (Yields may also contain a risk premium when inflation is uncertain, but this feature of returns is rarely modeled.) Neither component of nominal yields is observed directly, and models that ex ploit interest-rate data rely on auxiliary assump tions to separate expected real rates from expected inflation. The models discussed below are distinguished by the assumptions about the real rate process that are used to identify these components of the nominal interest rate. One method of identifying the model is to assume that the expected real rate of return fol lows a random walk. This implies that (6) R (t) = R ( t - l) + i( t ) . Then, if the realized real return is equal to the expected real return plus a noise term r\(t ), the first difference of the observed real return takes the form (7) A /? (0 = S ( 0 +Tl(0-Tl(f-1). TABLE 3 The Correlation among Expected Quarterly Inflation Rates when the Expected Real Rate Follows a Random Walk and the Nominal Yield Is the Sum of the Expected Real Rate and Expected Inflation PCEDI CPI CPIS PPI PPIW F CPI 1.00 0.95 0.96 0.93 0.92 CPIS 0.95 1.00 0.88 0.87 0.78 PPI 0.96 0.88 1.00 0.96 0.92 PPIW F 0.93 0.87 0.96 1.00 0.89 PCEDEF 0.92 0.78 0.92 0.89 1.00 (8) SOURCE: Author’s calculations. TABLE 4 The Correlation among Expected Quarterly Inflation Rates Generated by a Regression-Based Model. The First Difference in Inflation Is Projected onto the First Difference in the 90-Day Treasury Yield CPI CPIS PPI PPIW F CPI 1.00 0.95 0.94 0.93 0.91 CPIS 0.95 1.00 0.85 0.88 0.78 PCEDI PPI 0.94 0.85 1.00 0.95 0.90 PPIW F 0.93 0.88 0.95 1.00 0.88 PCEDEF 0.91 0.78 0.90 0.88 1.00 SOURCE: Author’s calculations. This has a first-order moving average repre sentation identical to that of equation (1). Esti mation of this model yields an expected real return series.7 Quarterly inflation forecasts are then constructed by subtracting the expected real return series corresponding to a particular price index from the yield on 90-day Treasury bills. The correlation among the inflation forecasts created in this manner is described in table 3The more sophisticated model of expecta tions yields inflation forecasts that are both more accurate and more highly correlated with each other than those from the time series model, even when the dynamics of the ex pected real interest rate are extremely simple.8 The increased correlation is especially notice able in situations where the correlation between the time series forecasts is lowest; for example, in the service component of the CPI and PPI. The high correlation among the fitted values from the interest-rate-based models suggests that all of the forecasts are tracking some under lying trend. The natural interpretation of that trend is the core rate of inflation. This interpretation is reinforced by estimates from a closely related model. If expected real rates are constant or nearly constant between adjacent quarters, the main source of variation in Treasury yields is the inflation premium. This suggests a regression-based model of the form A 7i ( 0 = (3 o + A i ( t ) Pi + e (t), where A n (t) is the change in inflation from one quarter to the next and A i ( t ) is the change in Treasury yields from the beginning of quarter t- 1 to the beginning of quarter t . Estimation of this model indicates a statistically significant relationship between the change in Treasury yields and the change in inflation.9 The correlation among fitted values obtained by estimating equation (8) is documented in table 4. The strong resemblance between these results and those presented in table 3 suggests that whether interest rates are included in the model is a more important consideration than the manner in which they are incorporated. As before, the expected inflation forecasts track each other quite closely. Adding lagged values of either the growth rate of money or the change in the growth rate of money to the regression equation has almost no impact on the fitted values for expected infla tion, even though the regression coefficients as sociated with these variables are statistically ■ 7 Application of the Breusch-Pagan test to the residuals from maxi mum likelihood estimates reveals ARCH effects. The figures in table 3 are based on fitted values from a maximum likelihood model where the condi tional variance is ARCH (2,0). It is also worthwhile noting that the mag nitude of the moving-average parameter is considerably less than in the results reported by Fama and Gibbons for monthly data. In other words, monthly data contain even more noise. ■ 8 Fama and Gibbons (1984) document the superiority of this model relative to the time series model, using monthly data. ■ 9 The model is estimated by maximum likelihood with an MA(1) error structure and an ARCH correction for conditional heteroscedasticity. The regresssion coefficient (ii is statistically significant at 1 percent for all of the inflation series when the parameter covariance matrix is esti mated from the information matrix, with or without the Newey-West cor rection for heteroscedasticity. 1 TAB L E 5 V. Hamilton’s Model The Correlation among Actual and Predicted Series for the CPI A potential shortcoming of the econometric IMA Actual (1,1) Real rate is a random walk Regres sion w/int. rates 1.00 0.76 0.81 0.80 0.82 0.76 1.00 0.96 0.97 0.95 0.81 0.96 1.00 0.99 0.98 Regression 0.80 with interest rates 0.97 0.99 1.00 0.98 Same with 0.82 interest rates and money 0.95 0.98 0.98 1.00 Actual IMA (1,1) Real rate is a random walk Same w/int. rates and money methodologies that I have considered is the ex tremely simple dynamics that are imposed on expected real interest rates and expected infla tion in order to identify these components of the nominal rate process. Hamilton (1985) has proposed and estimated a model that permits richer dynamics in both components, and for malizes the intuition that the observed rate is equal to a signal (expected inflation) plus noise. The model, which contains the random-walk formulation (6) as a special case, assumes that the following relations among inflation, ex pected inflation, and real interest rates are stable over time: (9) SOURCE: Author’s calculations. IV. Correlation among Forecasts from Different Methodologies The results discussed above concern the correla tion among the predicted values of different inflation series obtained with a specific econo metric methodology. Inspection of the predicted values for a given series and different methodol ogies suggests that three observations are in order. First, the inflation forecasts from the dif ferent models are highly correlated; they appear to be tracking a common element. Second, the forecasts track each other more closely than they track actual inflation, consistent with my interpre tation of the inflation series as signal plus noise. Third, the forecasts that incorporate interest-rate data are both more accurate than the forecasts generated by the time series model and more highly correlated with each other than with the time series model. Although table 5 describes the correlation among forecasts only for the CPI, similar results obtain for the other price series. A A r ( t ) = kr+ 0 ( L ) r ( t ) + ' ¥ ( L ) n ( t ) + Ç ( Z ) 7 c ( f ) + er ( f ) > (10) significant in all of the models. Indeed, the cor relation among fitted values cannot be distin guished from the results presented in table 4. This is consistent with results reported by Fama (1982), who finds that interest rates contain most of the information about expected infla tion that may be extracted from the history of money and output. A n (t) = k n + a (Z) r ( t ) + (3 (Z) n (t) + Y(Z)7t ( O + M O , (11) 71 ( t ) = n (t) + e(t). Expected real rates and expected inflation are described by linear projections of these variables on their own past values and on the past values of actual inflation. The difference between ex pected inflation and actual inflation is a noise term, as in the simpler models discussed above. These assumptions, along with the assumption that the nominal rate is equal to the real rate plus the expected inflation rate, are sufficient to iden tify expected real rates and expected inflation. Note that equations (9) and (10), like equations (7) and (8), are statistical models of the relation ships among these variables; there is no presump tion that the lag polynomials O (Z ), \ j/ (Z ), £, (Z ), a (Z ), (3 (Z ), and y (Z ) represent the decision rules that agents use to form expectations about real rates and inflation. Hamilton’s model enjoys a second advantage relative to the simple models in addition to encompassing a wider variety of time series be havior. In equations (9), (10), and (11), the dis tinction between errors in expectations and errors that result from the econometrician’s ina bility to observe expected real rates or expected inflation is modeled explicitly. The error terms £r and £k represent innovations in the expected real rate and expected inflation rate that are not captured by the linear projections of equations (9) and (10). These innovations arise because TABLE 6 The Correlation among Expected Quarterly Inflation Rates Generated by Hamilton’s Kalman Filter Model of Expected Inflation and Interest Rates CPI CPI CPIS PPI PPIW F PCED1 0.57 0.60 1.00 0.57 0.40 CPIS 0.57 1.00 0.51 0.69 0.69 PPI 0.40 0.51 1.00 0.65 0.62 PPIW F 0.57 0.69 0.65 1.00 0.85 PCEDEF 0.60 0.69 0.62 0.85 1.00 SOURCE: Author’s calculations. we are unable to observe expectations. The error term e(t ) represents the difference be tween what agents thought would occur and what did in fact occur. Estimation of these para meters allows us to evaluate explicitly the con tribution of these different sources of noise to the difference between expected inflation and actual inflation, making it unnecessary to assign an economic interpretation to the movingaverage parameter in a time series model. The estimated series are consistent with those produced by the other econometric models, in that innovations in the inflation rates appear to contain a substantial noise component.10 One indicator of this phenomenon is the set of coeffi cients that represents the projection of expected inflation onto past values of inflation and ex pected inflation. In general, the sum of the coef ficients for the four lagged values of expected inflation tends to be near one, while the sum of the coefficients for the four lagged values of ac tual inflation tends to be near zero. At the first two lags, the effect is even stronger; estimated parameter values imply that inflationary expec tations tend to persist, while inflationary shocks tend to be reversed. This pattern, which is con sistent with the time series properties of the er rors in the simpler econometric models, is characteristic of all of the series except for the PCE deflator.11 It suggests that expectations of ■ 10 1estimate the state space version of the model described in Burmeister, Wall, and Hamilton (1986). By doing so, I avoid dealing with the moving-average error terms that characterize the earlier formulation. ■ 11 My estimates for the deflator series are qualitatively similar to those reported by Hamilton (1985) and Burmeister, Wall, and Hamilton (1986). inflation tend to persist, even in the face of sig nificant changes in the current inflation rate. A second indicator of the noise in the series for realized inflation is the fraction of the varia tion in the inflation rate accounted for by the expectation error series e ( t ). This ranges from 20 percent in the case of the PCE deflator to 60 percent in the case of the PPI. The expected inflation series from Hamilton’s model differ from the estimates produced by the simpler econometric models in one important respect: The substantial increase in the number of explanatory variables yields a significant im provement in fit. As a result, the predicted values bear a stronger resemblance to the actual values and a weaker resemblance to each other. This fact is evidenced by the correlation among predicted values described in table 6. VI. A Multiple Indicator Model A multiple indicator model based on Hamilton’s methodology incorporates the flexible dynamics of that model, but focuses on the common com ponent of the different series rather than on the expected component of a particular series. Inter est rates and a set of realized inflation series are driven by a single expected inflation series. This series is distinguished from the expected inflation series generated by Hamilton’s model in that it provides information about pervasive price growth rather than about the behavior of a particular index. I estimate the model by projecting expected inflation and the expected real interest rate onto their own past values and onto past values of the PPL The realized values of the PCE deflator and the CPI both serve as indicators of the core rate. The realized value of inflation for each index is presumed to be equal to expected inflation plus a noise term. The expected inflation series for this model is presented in figure 2, along with the actual series for the CPI and the PCE deflator. Expected infla tion exhibits the same time-series properties as do the individual series described above. Innovations in realized inflation are reflected only weakly in current expected inflation, which nonetheless dis plays a great deal of persistence. F I GU R E 2 Expected and Realized Inflation, 1955 -1987 Percent, a n n u a l rate SOURCES: U.S. Department o f Commerce, Bureau o f Economic Analysis, and U.S. Department o f Labor, Bureau o f Labor Statistics. VII. Conclusion Inflation targets may contribute significantly to the credibility of a monetary policy that is oriented toward controlling inflation. A potential problem with inflation targets is that inflexible mles would couple money growth to random shocks in the price level; the substantial noise in individual inflation series suggests that this con cern is more than academic. Building flexibility into policy rules is one means of dealing with this problem, but flexibility tends to undermine the credibility of the commitment to control infla tion. An inflation target that filters out these tran sient shocks, combined with a tight feedback mle from the filtered inflation rate to money growth, is an alternative that maintains credibility while mitigating the problems as sociated with noise in the policy targets. Expected inflation is an indicator of the perva sive price growth, or core inflation, that interests the architects of monetary policy. The correlation among expected inflation rates from different price series and forecast methodologies suggests that these series are tracking the core rate. Signal extrac tion models formalize this intuition. Policy rules linked to the expected inflation series from any of the econometric models examined here are both forward-looking and reasonably insulated from index-specific shocks. Moreover, such broadly based targets would be difficult to manipulate. All of these properties suggest that expected inflation may serve as an effective guide to monetary policy. El References Ansley, Craig F. “Signal Extraction in Finite Series and the Estimation of Stochastic Regres sion Coefficients,” Washington, D.C.: Ameri can Statistical Association, Proceedings of the Business and Economic Statistics Section, 19 8 0 . Burmeister, Edwin K., Kent D. Wall, and James D. Hamilton. “Estimation of Unob served Expected Monthly Inflation Using Kal man Filtering,” Jo u rn a l o f Business a n d Econom ic Statistics, vol. 4, no. 2 (April 1986), pp. 147-60. Fama, Eugene F. “Inflation, Output, and Money,” Jo u rn a l o f Business, vol. 55, no. 2 (April 1982), pp. 201-31. ________ , and Michael R. Gibbons. “A Com parison of Inflation Forecasts, "Jo u rn a l o f M onetary Economics, vol. 13, no. 3 (May 1984), pp. 327-48. ________ , and________ . “Inflation, Real Returns, and Capital Investment,” Jo u rn a l o f M onetary Economics, vol. 9, no. 2 (May 1982), pp. 297-323. Hamilton, James D. “Uncovering Financial Market Expectations of Inflation,” Jo u rn a l o f P o litical Economy, vol. 93, no. 6 (December 1985), pp. 1224-41. The Case of the Missing Interest Deductions: Will Tax Reform Increase U.S. Saving Rates? by David A ltig David Altig is an assistant profes sor of business economics and public policy at Indiana University and is an economist at the Federal Reserve Bank of Cleveland. The author gratefully acknowledges helpful comments from Michael Bryan, Chris Carroll, Randall Eberts, and Jagadeesh Gokhale and excellent research assistance from Sharon Parrott. Introduction this percentage increased steadily, from 0.85 per cent in 1977 to 1.7 percent in 1986 (see figure 1). The period subsequent to 1976 was also dis tinguished by a downward trend in personal, private, and national saving rates (see figure 2). The coincidence of decreasing personal saving rates and increasing personal interest deductions can also be seen in figure 3, which plots per sonal saving (as a percentage of GNP) against nonhousing interest deductions (as a percent age of GNP). While the negative relationship that appears in figure 3 does not necessarily imply that elim inating the deductibility of nonhousing interest Beginning in tax year 1991, U.S. taxpayers may no longer deduct personal interest expense when calculating taxable income, thus complet ing the transition from the unlimited deductibil ity provisions that existed prior to enactment of the Tax Reform Act of 1986 (TRA86). In tax-speak, personal interest expense comprises interest pay ments not associated with mortgages on qualified residences or certain income-generating ac tivities. Generally speaking, personal interest ex pense amounts to interest payments on consumer loans not secured by real estate. Although a large share of household interest payments are associated with mortgage-related interest payments, which remain deductible under TRA86, disallowing deductions for per sonal interest expense is likely to have a sub stantial impact on consumer behavior.1 Indeed, eliminating the deductibility of personal interest expense may, in the final analysis, be one of the more important legacies of TRA86. It is certainly obvious that personal interest deductions had been increasingly exploited in the years preceding passage of TRA86. After trending upward during the 1950s, the growth of nonhousing interest deductions stabilized through the mid-1970s, fluctuating between 0.8 and 1.1 percent of GNP. After 1976, however, ■ 1 The ratio of housing to nonhousing interest deductions on per sonal tax returns was 1.19 in 1966,1.78 in 1976, and 1.78 again in 1986. The largest value of this ratio over the 1964-1986 period was 1.94, which was realized in 1983. Unfortunately, the Internal Revenue Service's Statistics o f Income, from which these numbers are calculated, does not generally distinguish among the categories of nonhousing interest deduc tions. The nonhousing interest measures used in this paper therefore in clude interest expense associated with personal investment. Fortunately, available data suggest that investment interest expense claimed by indi viduals is small relative to personal interest expense. In 1977, for exam ple, 65 percent of total household interest deductions were associated with home mortgages, 34 percent were associated with personal interest expense, and only 1 percent was associated with interest expense from in vestment activity. 23 FIGURE 1 Nonhousing Interest Deductions Percent o f GNP 1950 I9 6 0 1970 1980 1990 SOURCES: U.S. Department o f Commerce, Bureau o f Economic Analysis, and Internal Revenue Service. F I G U R E 2 Saving Rates Percent o f GNP SOURCE: Carroll and Summers (1987). F I G U R E 3 Personal Saving vs. Nonhousing Interest Deductions U nadjusted p e rso n al saving, p e rce n t o f GNP N o n h o u sin g interest d e d uctio n s, pe rce n t o f GNP SOURCES: U.S. Department o f Commerce, Bureau o f Economic expense will cause an increase in the U.S. saving rate, it is commonly believed that removing in centives to dissave does indeed result in higher savings relative to income. To a large extent, this belief arises from the simple intuition that increasing the price of an activity— in this case, borrowing— will naturally lead to a decrease in that activity. Economic theory thus leads us to conclude that more restrictive tax treatment of personal interest expense will lead to less con sumption and more saving. Although empirical evidence is limited, it ap pears that the negative relationship between household borrowing subsidies and saving be havior suggested by economic theory can be found in real-world economies. Tanzi (1987) has shown that personal saving as a percentage of disposable income has tended to be lower in countries with the most generous tax treatment of personal interest expense (this evidence is also presented in Sheshinski [1990])- In a provocative comparison of U.S. and Canadian saving rates, Carroll and Summers (1987) argue that part of the historical divergence between observed saving rates in these two very similar economies is likely because, unlike taxpayers in the United States, Canadian taxpayers were un able to deduct personal interest expense.2 In this paper, I consider further some of the evidence presented by Tanzi and Carroll and Summers. Specifically, I ask two simple ques tions. First, do private saving rates tend to be higher, on average, in countries that prohibit the deductibility of personal interest expense? Second, do tax subsidies to borrowing help ex plain U.S.-Canadian saving rate differentials? The empirical evidence I present gives affir mative answers to both questions. With respect to the first question, I examine private saving rates from 1975 to 1986 in a sample of 15 mem ber countries of the Organisation for Economic Co-operation and Development (OECD). I find that private saving rates were indeed higher on average in countries without tax subsidies to consumption loans. These results confirm for private saving the observations made by Tanzi with respect to personal saving.3 ■ 2 Limitations on interest deductions available to Canadian tax payers also apply to interest expense from home mortgages. See the dis cussions in Carroll and Summers (1987) and Tanzi (1984). Analysis, Internal Revenue Service, and Carroll and Summers (1987). ■ 3 Private saving is the sum of saving by households, or personal saving, and saving by corporations. Et O f course, simply comparing aggregated cross-country saving rates provides only casual evidence. Like the relationship in figure 3, such comparisons do not control for other causal fac tors. A more detailed analysis, which builds on the Carroll and Summers work, is provided in section III. The empirical models in this section add proxies for the U.S. subsidy rate on con sumption loans to the Carroll and Summers regression equations for U.S.-Canadian saving differentials. The subsidy variables consistently appear with statistically and economically sig nificant negative effects on private saving, a result that is remarkably robust across different specifications of the empirical model. Even the more sophisticated analysis of sec tion III has serious limitations — the data include only 24 annual observations, no attempt is made to control for simultaneity bias, and the subsidy proxies are admittedly crude, to name just a few. Furthermore, the effect of the borrowing subsidy variable is not consistently significant in regres sion models of the U.S. saving rate alone. None theless, the results reported here are generally supportive of the assertion that consumptionloan subsidies may have important negative ef fects on saving behavior, and hence important implications for the long-mn performance of the U.S. economy in the wake of TRA86. Savers in the model have access to two types of assets: physical capital, denoted by a {. for an age i individual of generation j, and private debt, which takes the form of consumption loans between generations.4 To make the model interesting, it is necessary that some generation chooses to borrow. I there fore assume that each generation is endowed with an identical, exogenous life-cycle labor productivity profile given by (Ej, £2, 0), where £2 is sufficiently larger than £: to ensure that the young always choose to borrow. Let borrowing by a young household born at time t be given by slr Abstracting from population growth, mar ket clearing in the consumption loans market requires that s1; = h2 t _x , where h71A is lending by the generation that is middle-aged in time t. 5 With these definitions in hand, the budget constraints for each generation are defined as (2) C u = E1wt+ s u , (3) ^ 2 / = £ 2tti + l _ [l +r+ d l{\-bt+l) ] s lt- a 2t- h 2t, and (4) C^t = (1 + rt+2) a 2t + [1 + r f 2(1 - p f+2) ] h 2t, I. A Simple Analytical Framework Although the intuition for a negative relation ship between favorable tax treatment of house hold borrowing and personal saving is readily apparent, introducing a simple analytical framework will help to organize the issues. The framework presented here is a simple, perfect-certainty, overlapping generations model in which each generation lives three periods. Every generation consists of identical individuals who inelastically supply one unit of labor in the first two periods of life, retiring in the third. Utility is assumed to be a logarithmic, time-separable function of consumption given by where r is the rate of return to physical capital, r d is the return to private debt, 8 is the subsidy rate on borrowing (or, alternatively, the marginal tax rate on nonwage income for age 2 individ uals), and p is the tax rate on interest income earned from the purchase of private debt. Equa tions (2), (3), and (4) embody the assumption that the young choose to borrow, the condition that all generations will consume their full life time resources (so that only middle-aged individ uals save), and the simplifying assumption that the marginal tax rate on income from physical capital is zero. Assuming interior solutions for individual saving and dissaving decisions, utility maximiza tion implies the first-order conditions 3 (1) M C„). i= 1 The variable (3 is the individual subjective timediscount factor, and the subscript t indexes each generation by date of birth. ■ 4 Because the analysis here abstracts entirely from transaction costs, nothing essential is lost by ignoring the role of intermediaries and assuming that loan contracts are directly traded between generations. ■ 5 The model abstracts from bequest motives and uncertainty, so all generations choose to “die" with no assets. The old will therefore never choose to accumulate capital or lend in the consumption loans market. 25 TABLE 1 Crowding-Out Effects of increasing the Subsidy to Consumption Loans Percentage R e d u c tio n in Steady-State C a p ital 8 B e n ch m ark P o p u la tio n g ro w th = 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.10 0.11 0.12 0.13 0.14 0.15 0.5 1.0 1.6 2.1 2.6 3.2 3.7 4.3 4.8 5.4 5.9 6.5 7.0 7.6 8.2 0.6 1.1 1.7 2.3 2.9 3.5 4.1 4.7 5.3 6.0 6.7 7.3 8.0 8.7 9.4 P ro d uctivity p ro file = (0, 10, 0) 0.8 1.6 2.4 3.3 4.2 5.0 6.0 6.9 7.8 8.8 9.8 10.8 11.9 12.9 14.0 NOTE: Each entry gives the percentage reduction in the steady-state capi tal stock w hen the subsidy rate on borrowing, 8, is increased from zero. The benchmark case assumes (3 = 0.778, 0 = 0.25, zero population growth, (Ej, e2, e3) = (1.5,8.5, 0), and p = 0.11. The other cases maintain the benchmark assumptions, w ith the exception o f the indicated parameters. SOURCE: Author’s calculations. (5) C 2,= p[l + r f , ( l - 8 , + 1) ] C 1,, (6 ) C 3, = p [ l + r f 2( l - p , + 1) ] C 2, , and (7) C 3,= p ( l + r,+ 2) C 2,. Equations (6) and (7) imply that, in assetmarket equilibrium, r t = r dt( 1 - p t ). The long-run effect of changes in the subsidy variable 8 can be demonstrated by a few simple simulation exercises. Table 1 reports the reduc tion in the steady-state capital stock caused by increasing the subsidy rate 8 for particular parameterizations of the model.6 In the benchmark case, which is described in table 1, increasing 8 from 0 to 10 percent causes the steady-state capi- ■ tal stock to fall by 5.4 percent.7 By extension, interest rates rise and per capita income falls.8 Table 1 also shows how factors that increase the demand for consumption loans amplify the crowding-out effects of allowing personal interest expense to be deducted for tax purposes. Thus, an increase in either the rate of population growth or the steepness of the productivity profile be-, tween young and middle ages results in larger percentage decreases in steady-state capital for a given change in 8. (See Bryan and Byrne [1990] and the references therein for a general discus sion of the effects of demographics on aggregate saving in a life-cycle context.) Because substantial disagreement persists among economists concerning the appropriate model of aggregate saving behavior, it is impor tant to note that the qualitative results of the model presented here are not dependent on life cycle assumptions. Altig and Davis (1989) show that changes in the subsidy rate on consumption loans can also have significant long-run negative effects on aggregate savings in models where parents and children are altruistically linked, as in Barro (1974). In fact, under the plausible assump tion that the tax rate on interest income exceeds the subsidy rate on borrowing, Barro-type models predict that changes in subsidy rates can have large long-run effects on the size of the capital stock even when changes in the tax rate on inter est income do not (see Altig and Davis [19891 for a full treatment of this issue). II. Do Private Saving Rates Tend to Be Higher in Countries Without Borrowing Subsidies? Table 2 answers this question directly. The answer is yes, at least for the subset of OECD countries examined here.9 The results in table 2 ■ 7 In general, the direction of change in aggregate savings depends on the nature of the assumed preference structure. Under “standard” preferences, however, changes in the subsidy rate will have effects that are qualitatively the same as the ones reported here. The seminal discus sion of this issue in an overlapping generations framework can be found in Diamond (1965). ■ 8 The simulations reported in table 1 assume a Cobb-Douglas production technology, expressed in effective labor units as y = k d.The steady-state rate of return to capital is therefore given by 0 /re ‘ 1. Thus, y is increasing in k, and r and r d (by the asset-market clearing condi tion) are decreasing in k. ■ 6 The simulations reported in table 1 assume that all revenues raised (or lost) through distortionary taxation are rebated (or recovered) via lump-sum subsidies (or tax levies) to the affected generations. 9 The countries are Australia, Austria, Belgium, Canada, Denmark, France, Ireland, Japan, the Netherlands, Norway, Sweden, Switzerland, the United Kingdom, the United States, and West Germany. The data are from OECD National Accounts. □ TABLE 2 Average Private Saving Rates, 1975-1985 G ro u p Averages C onsum e r interest n o t de d uctible C o n sum e r interest d eductible Simple average 10.68 8.65 Weighted average 11.14 8.38 In d iv id u a l C o u n try Averages C o n sum e r interest n o t de d uctible Average saving rate Australia 5.65 Austria 9.83 Belgium 12.73 Canada 11.81 France 9.74 Ireland 14.80 Japan 15.73 United Kingdom 7.33 West Germany 8.53 C o n sum e r interest d e d uctible Denmark Netherlands Norway Sweden Switzerland United States Average saving rate 7.45 12.24 5.04 5.23 13.72 8.22 NOTE: Entries represent averages for subsets o f 15 OECD countries. C oun tries are classified into deductible and nondeductible groups according to the information provided by Tanzi (1984). Weighted averages are con structed using within-group relative shares o f real GDP. Real GDP figures are obtained from Summers and Heston (1988). Saving rates are expressed as percentages o f GNP. SOURCE: Organisation for Economic Co-operation and Development, National Accounts o f OECD Countries, 1975-1987, Volum e II. were obtained by first averaging private saving as a percentage of gross domestic product (GDP) over the sample period 1975 to 1985 for each of the 15 countries considered.10 The countries were then grouped according to whether tax subsidies were provided to interest expense from general (nonhousing) consumer credit.11 Two sets of group-average measures are reported in table 2— one based on simple averaging and one obtained by weighting the individual country averages by within-group relative shares of real GDP.12 The average private saving rate for the sample period was 10.68 percent in countries without favorable tax treatment of personal inter est expense and 8.65 percent in countries with favorable tax treatment of personal interest ex pense (11.14 percent and 8.38 percent, respec tively, when country-specific saving rates are weighted by GDP shares). To put the magnitude of this difference in some perspective, the U.S. current account deficit was 5 percent of GDP in 1988. A 2 percent increase in the private saving rate for 1988 could therefore have financed more than one-third of the U.S. current account deficit, an amount equivalent to about $44 bil lion in 1988 dollars. Table 2 also clearly shows that, in the chosen sample, average saving rates varied substantial ly among countries with similar tax treatment of personal interest expense.13 It is impossible to know how much of the variation can be ac counted for by economic, demographic, and policy variables without a more detailed inves tigation of the data. Unfortunately, the informa tion that is necessary to conduct a more detailed ■ 10 The savings measures used here are net of depreciation. See Aghevli et al. (1990) for a general discussion of the OECD saving measures. ■ 11 Countries are classified into subsidy and nonsubsidy groups according to information reported in appendix III of Tanzi (1984). Up dated information in Tanzi (1987) indicates that these classifications were still valid in 1985. ■ 12 Relative GDP shares are obtained using real GDP at internation al prices calculated by Summers and Heston (1988). ■ 13 There were also significant differences in the trend of saving rates for countries within the two groups. In the subsidy group, for instance, Norway, Sweden, and the United States experienced declining saving rates over the sample period, while Denmark, the Netherlands, and Switzerland all experienced fairly strong upward trends. a inquiry is difficult to come by.14 Because of this difficulty, the balance of this paper focuses on a comparison between two countries for which data are more readily available: the United States and Canada. III. Has the Subsidy Rate on Consumer Loans Reduced U.S. Saving Relative to Canadian Saving? Following Carroll and Summers (1987), the start ing point of the analysis in this section is a simple saving equation given by (8) St = a 0 + a 17i1+ a 2UNt + a 3SURPt + a 4SHELTt + a 5NWt + a 6R f + r\t , where St is the time t differential between the U.S. and Canadian private saving rate (as a per cent of GNP), n t is the differential in inflation rates for consumer prices, UN( is the differential in unemployment rates (as a percent of the total labor force), SURPt is the differential in net gov ernment saving (as a percent of GNP), SHELTt is the differential in the level of saving in taxsheltered assets (as a percent of personal dispos able income), NWt is the differential in household net worth (as a percent of GNP), and Rtat is the differential in weighted averages of after-tax returns to sheltered and nonsheltered saving. Before proceeding to a discussion of my em pirical work, it will be useful to introduce the ra tionale for including the particular regressors shown in equation (8). The inflation variable is included to control for the tendency of national income-account saving measures to overstate actual saving when inflation increases. The idea is that standard measures of income are dis torted by changes in nominal interest rates that arise solely from changes in the rate of inflation or, more precisely, from the expected rate of in flation. This issue is examined in detail by Jum p ■ 14 I did examine many cross-sectional regressions with variations ot the empirical specification employed by Feldstein (1980). In particular, I attempted to find whether this type of cross-sectional empirical saving model tends to underpredict the average private saving rate for countries without borrowing subsidies and overpredict the saving rate for countries with borrowing subsidies. For some of the models, I found regression errors were uniformly positive for the no-subsidy countries and uniformly negative for the countries with subsidies. However, the results were so sensitive to sample size, choice of regressors, and sample period that it was impossible to make a convincing case one way or the other. The general nonrobustness of Feldstein-like empirical saving models is also reported by Slemrod (1990) and Bosworth (1990). (1980). The expected sign of a l is positive if the type of measurement problem Jum p iden tifies is the primary channel through which infla tion rates help to explain aggregate savings. The unemployment variable is a proxy for differences in cyclical conditions across the two countries. Assuming that changes in unemploy ment primarily reflect deviations from the equi librium rate of unemployment, an appeal to the reasoning underlying the permanent-income hypothesis implies that a 2 < 0. In other words, we expect higher unemployment and more dis saving when income is temporarily low. The coefficient a 3 measures the relationship between public saving and private saving. In the simplest scenario, we expect to find a 3 = -1 if the conditions necessary for Ricardian equiva lence are true and a 3 > -1 if those conditions are not true.15 However, unambiguous predic tions for the value of a 3 are complicated by the fact that equation (8) does not control for inde pendent effects associated with government ex penditures (see Aschauer [1985D. The significance of the sheltered saving vari able is the key finding of Carroll and Summers. SHELTt specifically measures the U.S.-Canadian differential in total personal saving in tax-sheltered forms (as a percentage of disposable per sonal income). In the United States, sheltered saving is represented by contributions to indi vidual retirement accounts (IRAs). The Canadian equivalent of IRAs are registered retirement savings plans. Carroll and Summers estimate values of a 4 that range between 1 and 2, implying that in creases in the amount of saving in tax-sheltered assets are associated with greater than one-toone increases in total private saving. Although this impact seems large, it is qualitatively consis tent with microdata evidence presented by Venti and Wise (1987), who estimate that 80 to 90 percent of IRA contributions represent net in creases in personal saving. The final two variables, NWt and Rtat, are expected to enter equation (8) with negative and positive coefficients, respectively. The networth variable is included to capture the possi bility that private saving, as measured on a national income accounts basis, changes as households seek to maintain target wealth-toincome ratios. Thus, as net worth rises relative to GNP, private saving tends to fall. ■ 15 The literature on Ricardian equivalence is massive. Good general discussions can be found in Bernheim (1987,1989) and Barro (1989a, 1989b). TABLE 3 Regression Results M odel C oefficient V alues CONST INFL 1 4 -.027 .004 .004 .016 (.45) (.43) (2.3)b .231 .197 -.017 .156 (.04) (.43) .430 .404 .290 (.62) SURP 3 (3.3)a (.95) UN 2 -.837 (3.4)a SHELT (1.1) .506 (1.4) -.267 (1.1) -.281 (1.3) -.365 (1.2) (1.2) (1.8)c 1.74 1.98 -.665 (4.2)a (3.2)a (.64) -.254 .228 (.55) (.54) R at NW .179 (2.9)a Adj. R 2 .559 .760 .751 .824 P .585 .408 .407 .226 from regression analysis on equation (8) cannot be viewed as decisive indicators of the structural relationships between U.S.-Canadian saving dif ferentials and the explanatory variables.16 The appropriate interpretation of the approach taken here is that of an investigation into whether par tial correlations of saving differentials and in cluded regressors are consistent with structuraltheoretical predictions. Table 3 presents the results of several regres sions based on equation (8). The data are annual and, with a few exceptions, are from Carroll and Summers (1987).17 Model 1 in table 3 includes inflation, unemployment, and gov ernment surplus differentials as regressors. The coefficients on the inflation and government surplus variables have the anticipated sign, but only the government surplus variable is statisti cally significant.18 The coefficient on the unem ployment differential has the “wrong” sign, but is not statistically different from zero. Models 2-4 in table 3 all include the differen tial in sheltered saving as a regressor. Models 2 and 3 essentially replicate the crucial Carroll and Summers result — the coefficient on SHELT is positive, large, and statistically significant. The coefficient on SHELT does become statistically in significant when the U.S.-Canadian net wealth differential is added to the basic regression model. a. The null hypothesis that the corresponding coefficient is zero can be rejected at the 99 percent confidence level. b. The null hypothesis that the corresponding coefficient is zero can be rejected at the 95 percent confidence level. c. The null hypothesis that the corresponding coefficient is zero can be rejected at the 90 percent confidence level. NOTE: The dependent variable is the U.S.-Canadian differential in private saving relative to disposable income. All other variables are as defined in equation (8). The variable p is the first-order autocorrelation coefficient of the residual series. The numbers in parentheses are the absolute value of the t statistics for the corresponding coefficient estimate. SOURCE: Author’s calculations. The after-tax real interest rate is included to capture the effects of changes in the return to saving. The expectation that a 6 > 0 assumes that preferences cause substitution effects to domi nate income effects and that ex post real rates are reasonable proxies for ex ante real rates. An important consideration in discussing the expected signs of the coefficients in equation (8) is that I have described the relationships that would arise in an explicitly structural saving func tion. Equation (8) is, of course, decidedly nonstructural. Thus, coefficient estimates derived ■ 16 The problems in interpreting coefficient estimates from equa tion (8) are twofold. First, the coefficients in equation (8) are almost cer tainly “mongrel parameters,” that is, unspecified functions of the under lying structural parameters. Second, no attempt is made to control for biases that may arise if the regressors are correlated with the error term n i, a situation that seems likely. With respect to this latter problem, I did some limited experimentation with instrumental variables (IV) estimation. Unfortunately, the standard errors of the IV estimates were so large that no interesting inferences were possible. ■ 17 Unemployment rates are taken from the OECD Labor Force Statistics. The SHELT variable was constructed from data graciously provided by Chris Carroll (for Canada) and from data reported in Carroll and Summers (for the United States). ■ 18 The tables indicate coefficients that are statistically nonzero at the 90 percent, 95 percent, and 99 percent confidence levels. In the Car roll and Summers paper, reported t statistics are corrected for serial cor relation. Although such corrections were made for all of the models reported in this paper, I have chosen not to report corrected t statistics for two reasons. First, almost all of the models estimated yield DurbinWatson statistics that fall within the “ inconclusive" range. Furthermore, although many of the models estimated yield “ large” values of the firstorder correlation coefficient of the residual series (p), the null hypothesis p = 0 is rarely rejected at the 95 percent confidence level. Second, work by Mishkin (1990) indicates that the type of correction employed by Car roll and Summers has undesirable properties in small samples. In most cases, the basic message is independent of whether t statistics are cor rected or uncorrected. 29 TABLE 4 Regression Results Including Subsidy Variable M odel C o efficien t V alu es CONST 6 7 8 .100 .092 .093 .064 (4.7)a (1.8)b (6.1)a INFL UN -.018 .004 .060 .108 (.14) (.03) (.21) (.36) -.077 .010 .025 .081 (.32) (.04) (.09) (.29) -.260 SURP 00 5 (1.8)b SHELT -.210 (1.4) -.206 (1.3) -.247 (1.5) .353 .273 -.372 (.89) (.50) (.44) .069 (.21) .185 (.54) R at .062 NW (1.0) SUB Adj. R 2 P -4.25 -3.76 -3.80 -3.19 (8.0)a (4.9)a (4.7)a (3.2)a .893 .891 .885 .885 -.128 -.150 -.170 -.232 a. The null hypothesis that the corresponding coefficient is zero can be rejected at the 99 percent confidence level. b. The null hypothesis that the corresponding coefficient is zero can be rejected at the 90 percent confidence level. NOTE: SUB is measured as the ratio o f nonhousing personal interest deductions to adjusted gross income reported on itemized returns. See table 3 for other definitions. SOURCE: Author’s calculations. Note also that the sign on the net wealth coeffi cient is positive and statistically significant.19 Table 4 presents results of regressions that add to models 1-4 a variable measuring the average borrowing subsidy. The subsidy variable is con structed as the ratio of total nonhousing interest deductions on personal tax returns to the adjusted gross income of all taxpayers with itemized deductions. This series on average subsidy rates is constructed from various issues of the Statistics ■ 19 Carroll and Summers do not find the same sensitivity of the SHELT coefficient in their empirical analysis. The differences between their results and mine apparently result from the data. As subsequent results make clear, I find that no stable inference can be made about the relationship between U.S.-Canadian private saving differentials and dif ferences in the amount of sheltered saving in the two countries. http://fraser.stlouisfed.org/ Federal Reserve Bank of St. Louis o fIncom e fo r Individuals (published by the In ternal Revenue Service).20 The results in table 4 are striking. In every case, the null hypothesis that the subsidy vari able has zero effect on private saving is easily rejected at the 99 percent confidence level. As would be expected, the explanatory power of the saving models also increases when the sub sidy variable is included— in some cases, sub stantially. It is necessary to bear in mind, however, that the average subsidy variable included in these regressions is at best a crude proxy for the vari able that is theoretically important — namely, the m arginal subsidy rate on consumption loans. In fact, it is difficult to distinguish move ments in the subsidy variable that result from changes in tax incentives for borrowing from movements that result from shifts in the demand for consumption loans that are not associated with tax distortions. For example, suppose that an individual, facing no change in borrowing subsidies, simply decides to borrow an extra $10 at the margin. Suppose further that the rate of interest on this loan is 10 percent. Then the individual’s saving falls by $10 while his or her interest expense rises by $1. This single episode would suggest that the coefficient on the subsidy variable constructed from reported interest expense is -10, even though the borrowing behavior had nothing to do with tax-related borrowing subsidies.21 The regressions reported in table 5 replicate the regressions reported in table 4, with the sub sidy variable calculated as 20 percent of the average nominal annualized return on threemonth Treasury bills. Because personal interest deductibility provisions did not change during the sample period, exogenous changes in bor rowing subsidies arose through two channels— changes in structural marginal tax rates and changes associated with variation in the rate of inflation in the context of a tax code that allowed for the deductibility of nom inal interest expense. The subsidy variable used for the regressions in table 5 is designed to capture the effects of the lat ter channel.22 ■ 20 Values for personal interest deductions are interpolated for the odd years from 1961-1971 and for 1974 by assuming that total interest deductions and mortgage-related interest deductions increase from the previous tax years at the same rate as total itemized deductions. ■ 21 I am grateful to Chris Carroll for suggesting this example, as well as the alternative subsidy variable discussed in the subsequent para graphs. ■ 22 Twenty percent is chosen as a rough approximation to the average marginal subsidy rate on borrowing in accordance with the num bers reported in table 1 of Altig and Davis (1989). TABLE 5 Regression Results with Alternative Subsidy Variable M odel C oe fficie n t V alues CONST INFL 9 10 11 .016 .017 .017 -.012 (2.1)a (2.2)a (2.1)a (.72) .227 (1.7) UN SURP .221 (1.6) 12 .142 .214 (.43) (.70) .072 .138 .113 .167 (.28) (.49) (.37) (.59) -.506 -.438 -.441 -.458 (3.5)b (2.5)a (2.4)c (2.7)a .326 .428 -.873 SHELT (.69) R at (.69) (1.0) -.094 .171 (.26) (.48) NW .111 (2.0)c SUB Adj. R 2 P -2.94 -2.60 -2.58 -2.00 (6.9)b (3.9)b (3.8)b (2.9)a .867 .864 .856 .877 -.074 -.068 -.064 -.232 a. The null hypothesis that the corresponding coefficient is zero can be rejected at the 95 percent confidence level. b. The null hypothesis that the corresponding coefficient is zero can be rejected at the 99 percent confidence level. c. The null hypothesis that the corresponding coefficient is zero can be rejected at the 90 percent confidence level. NOTE: SUB is measured as 20 percent of the average annualized return on three-month Treasury bills. See table 3 for other definitions. SOURCE: Author’s calculations. The results in table 5 do not differ appreciably from those reported in table 4. Although the coef ficients on the subsidy variables decrease in mag nitude, they remain large in absolute value and are always statistically different from zero. Fur thermore, as in the regressions reported in table 4, inclusion of the subsidy variable renders the SHELT variable insignificant in all cases.23 Table 6 presents the results of regressions based on other variations of the model given in equation (8) for each of the two subsidy variables used in tables 4 and 5. Models 13-16 report the results of estimated models in which demographic and income-growth variables are included as explana tory variables, extensions suggested by the theoreti- cal model in section I. Models 13 and 14 include the U.S.-Canadian differential in the percentage of the population aged 15-65. Models 15 and 16 report results in which the real GNP growth-rate differential is included as a regressor. 24 Models 17 and 18 of table 6 report results with personal saving taken as the dependent variable and corporate saving introduced separately as a regressor. Analogous to the ob servations made about the government surplus variable in equation (8), corporate saving, after controlling for total wealth, should have a onefor-one negative effect on personal saving if in dividuals “pierce the corporate veil.” 25 In every case, including numerous regres sions not reported in the tables, the result is the same. With the arguable exception of the government surplus variable, the borrowing subsidy, however measured, is the only explan atory variable that consistently shows up with a statistically significant effect on the U.S.Canadian saving differential. Furthermore, the effect is always negative, and strongly so. One further set of tests is reported in table 7. Because borrowing subsidies are zero for Canada, all variation in the subsidy variable arises from the U.S. data. The regressions in table 7 are therefore based on U.S. data alone.26 Although the models with the subsidy variable constructed from Treasury bill rates yield results that are consistent with regressions based on U.S.-Canadian saving differentials, it is apparent ■ 23 The subsidy proxy included in the table 5 regressions is, of course, subject to some of the same potential endogeneity problems as the subsidy variable employed in the table 4 regressions. For example, suppose that individuals in the economy anticipate better times ahead (and that these expectations are not closely related to effects that are con trolled for by the inclusion of unemployment or GNP growth differen tials). Permanent-income theory then tells us that the response w ill be an average increase in the desire to borrow. The resulting shift in the ag gregate saving curve will drive up both real and nominal interest rates (holding expected inflation fixed). ■ 24 If faster GNP growth means steeper life-cycle productivity profiles, the results of the simulations in section I suggest that coeffi cients on the GNP growth differential should be negative. However, the growth-rate differential may also pick up changes in cyclical conditions not captured by the unemployment-rate differential. This latter interpreta tion seems more likely in light of the significant positive coefficient es timates reported in table 6. ■ 25 The necessity of controlling for total wealth is emphasized in the empirical studies by Auerbach and Hassett (1989) and Poterba (1989). The results in these papers suggest to me that individuals do in deed internalize corporate saving when making personal consumption decisions. However, the evidence is, as usual, ambiguous. ■ 26 I am grateful to Randall Eberts for suggesting these regressions. 31 TABLE 6 Regression Results with Alternative Models M odel C oefficient Values 14 15 16 I7 a 18a .065 -.056 .082 -.057 .071 .015 (1.8)b (2.0)b (2.3)c (.30) (3.9)d (2.5)c .293 (.88) .271 .289 .263 .380 (.90) (.87) (.99) .294 .285 .102 (.98) (.84) (.45) 13 CONST INFL .111 (.36) UN SURP .078 -.822 (.27) (1.4) -.242 -1.26 (1.4) (2.9)c -.306 -.481 (1.9)b (2.7)c CORP SHELT -.193 (.74) -.439 (2.6)c -.075 (.28) .64 .442 (1.3) (.70) -2.4 .144 -.617 (.42) (2.0)b (.16) (.64) .186 .401 .283 .222 (.52) (.98) (.85) (.60) (.14) .131 (.47) .059 .104 -.005 .084 -.014 .002 (.34) (.05) NW (.87) POPRAT (1.6) .031 -.011 (.13) (.05) YGROW (.07) .204 (1.6) -3.19 -3.24 (3.1)d (3.4)d SUB 2 -1.88 (1.2) .034 .097 (.65) -2.65 (3.4)d -1.82 -1.89 (2.6)c (2.8)c (2.6)c .878 .871 .924 -.221 -.383 -.276 Adj. R 2 P .193 (.80) -.364 R al SUB I -.192 (1.4) (1.3) .873 .910 .896 -.193 -.122 -.009 a. The dependent variable is the personal saving rate differential. b. The null hypothesis that the corresponding coefficient is zero can be rejected at the 90 percent confidence level. c. The null hypothesis that the corresponding coefficient is zero can be rejected at the 95 percent confidence level. d. The null hypothesis that the corresponding coefficient is zero can be rejected at the 99 percent confidence level. NOTE: POPRAT is the differential in the percentage o f the population aged 15-65; YGROW'is the differential in real GNP growth rates; CORP is the differential in private minus personal saving rates; SUB 1 is the subsidy variable as defined in table 4; and SUB 2 is the subsidy variable as defined in table 5. See previous tables for other definitions. SOURCES: Author’s calculations and OECD National Accounts, various issues. TABLE 7 Regression Results for U.S. Personal Saving M odel C oefficient V alues 19 CONST UN SIM P SHELT Rdt 22 .107 .085 .066 23 (3.5)c (2.8)a .499 (2.4)a (.36) .103 (2.1)b -.489 (2.2)a -.515 .086 -.071 .007 -.472 -.024 (1.8)h (.52) (.24) (.04) (1.9)b (.13) -.271 -.643 (3.2)c -.106 -.373 -.444 -.636 (1.3) (.42) (2.3)a (2.3)a (3.2)c -1.21 -1.19 -.194 -1.52 (1.5) (2.3)a (.21) -1.31 (2.6)a -.675 .028 00 (5.9)c -.466 (1.2) (2.1)a -.771 .463 -.492 (3.0)c (2.0)b (2.0)b .002 .009 (3.2)c NW (.54) .006 (1.4) .008 (2.7)a YGROW .175 .090 (1.8)b (1.2) (1.2) Adj. R 2 P .655 .791 -.195 -.012 .310 .756 -0.18 (1.2) -2.60 -1.91 (4.6)c .490 .322 (1.2) (.25) 1.68 (4.0)c (1.3) .007 -1.18 -3.44 .325 (2.1)a 2.80 SUB 2 .079 (8.2)c .001 (2.6)a SUB I 24 .079 (8.1)c (2.8)a INFL 21 o .060 20 (2.4)a .745 .797 .040 .052 a. The null hypothesis that the corresponding coefficient is zero can be rejected at the 95 percent confidence level. b. The null hypothesis that the corresponding coefficient is zero can be rejected at the 90 percent confidence level. c. The null hypothesis that the corresponding coefficient is zero can be rejected at the 99 percent confidence level. NOTE: All variables refer to the U.S. values o f the variables defined in earlier tables. The dependent variable is U.S. personal saving as a percentage of disposable income. See table 6 for other definitions. SOURCE: Author’s calculations. 33 that the effects of borrowing subsidies are far less consistent when included as regressors in the U.S. private saving-rate models. Note also that the sheltered saving variables are in some cases negative, large, and statistically significant. Explaining these anomalies is an important topic for future investigations. References Aghevli, Bijan, et al. The Role o f N ational Saving in the W orld Economy: Recent Trends a n d Prospects. Washington, D.C.: Interna tional Monetary Fund, 1990. Altig, David, and Steve J. Davis. “The Timing IV. Concluding Remarks The United States is not alone in recent attempts to mitigate the attractiveness of consumption loans through less-favorable tax treatment of personal interest expense. Recent tax reforms in Denmark and Sweden, for instance, have in cluded provisions that effectively restrict the value of personal interest-expense deductions. Informative discussions of these changes and others can be found in Tanzi (1987) and Pechman (1988). The evidence presented in this paper, though cursory by design, does indeed point toward important effects on aggregate saving behavior as a result of changes in the tax treat ment of personal interest expense. In addition, as noted in section I, quite disparate models of intertemporal consumption behavior predict that changes in the degree to which consump tion loans are subsidized through the tax system can have substantial effects on aggregate saving. The combination of these observations suggests that no assessment of U.S., or world, tax reform is complete without careful scrutiny of the treatment of personal interest expense. of Intergenerational Transfers, Tax Policy, and Aggregate Savings,” Federal Reserve Bank of Cleveland, Working Paper 8917, December 1989- Aschauer, David A. “Fiscal Policy and Aggre gate Demand,” Am erican Econom ic Review, vol. 75 (March 1985), pp. 117-27. Auerbach, Alan J., and Kevin Hassett. “Corpo rate Savings and Shareholder Consumption,” National Bureau of Economic Research, Working Paper No. 2994, June 1989. Barro, Robert J. “The Neoclassical Approach to Fiscal Policy,” in R. Barro, ed., M odem B usi ness Cycle Theory. Cambridge, Mass.: Har vard University Press, 1989a. ________ . “The Ricardian Approach to Budget Deficits,” Jo u rn a l o f Econom ic Perspectives, vol. 3, no. 2 (Spring 1989b), pp. 37-54. ________ . “Are Government Bonds Net Wealth?” Jo u rn a l o f P olitical Economy, vol. 82 (November/December 1974), pp. 1095-17. Bernheim, B. Douglas. “A Neoclassical Per spective on Budget Deficits,” Jo u rn a l o f Econom ic Perspectives, vol. 3, no. 2 (Spring 1989), pp. 55-72. ________ . “Ricardian Equivalence: An Evalua tion of Theory and Evidence,” in S. Fischer, ed., NBERM acroeconom ics A n n u a l 1987. Cambridge, Mass.: MIT Press, 1987. Bosworth, Barry P. “International Differences in Saving,” Am erican Econom ic Review, vol. 80, no. 2 (May 1990), pp. 377-81. Bryan, Michael F., and Susan M. Byrne. “D on’t Worry, W e’ll Grow Out of It: An Anal ysis of Demographics, Consumer Spending, and Foreign Debt,” Federal Reserve Bank of Cleveland, Econom ic Com m entary, October 1, 1990. Carroll, Chris, and Lawrence H. Summers. “W hy Have Private Savings Rates in the United States and Canada Diverged?”Jo u rn a l o f M onetary Economics, vol. 20, no. 2 (Sep tember 1987), pp. 249-79. Diamond, Peter A. “National Debt in a Neoclas sical Growth Model,” A m erican Econom ic Review, vol. 55, no. 5 (December 1965), pp. 1126- 50. Feldstein, Martin. “International Differences in Social Security and Saving/'Jo u rn a l o f P ublic Economics, vol. 14, no. 2 (October 1980), pp. 225-44. Jump, Gregory V. “Interest Rates, Inflation Ex pectations, and Spurious Elements in Meas ured Real Income and Saving,” A m erican Econom ic Review, vol. 70, no. 5 (December 1980), pp. 990-1004. Mishkin, Frederic S. “Does Correcting for Heteroscedasticity Help?” National Bureau of Economic Research, Technical Working Paper No. 88, May 1990. Pechman, Joseph, ed. W orld Tax Reform: A Progress Report, Report of a Conference Held in Washington, D.C., on November 1213, 1987. Washington, D.C.: The Brookings Institution, 1988. Poterba, James M. “Dividends, Capital Gains, and the Corporate Veil: Evidence from Britain, Canada, and the United States,” Na tional Bureau of Economic Research, W ork ing Paper No. 2975, May 1989- Sheshinski, Eytan. “Treatment of Capital In come in Recent Tax Reforms and the Cost of Capital in Industrialized Countries,” National Bureau of Economic Research, Tax Policy a n d the Economy, vol. 4 (1990), pp. 25-42. Slemrod, Joel. “Fear of Nuclear War and Inter country Differences in the Rate of Saving,” Econom ic Inquiry, vol. 28, no. 4 (October 1990), pp. 647-57. Summers, Robert, and Alan Heston. “A New Set of International Comparisons of Real Product and Price Levels: Estimates for 130 Countries, 1950-1985,” Review o f Incom e a n d W ealth, vol. 34, no. 1 (March 1988), pp. 1-25. Tanzi, Vito. “The Response of Other Industrial ized Countries to the U.S. Tax Reform Act,” N ational TaxJo u rn al, vol. 40, no. 3 (Sep tember 1987), pp. 339-55. ________ . “Interest Rates and Tax Treatment of Interest Income and Expense,” in V. Tanzi, ed., Taxation, In flatio n , a n d Interest Rates. Washington D.C.: International Monetary Fund, 1984. Venti, Steven F., and David A. Wise. “IRAs and Saving,” in M. Feldstein, ed., The Effects o f Taxation on C apital A ccum ulation. Chi cago: University of Chicago Press, 1987, pp. 7-51. Economic Review ■ 1989 Q uarte r 4 ■ 1990 Q uarte r 2 Deposit-Institution Failures: A Review of Empirical Literature School Reform, School Size, and Student Achievement by Ash Demirgiic-Kunt by Randall W. Eberts, Ellen Kehoe Schwartz, and Joe A. Stone Settlement Delays and Stock Prices by Ramon P. DeGennaro The Effect of Bank Structure and Profitability on Firm Openings by Paul W. Bauer and Brian A. Cromwell ■ 1990 Q uarte r 1 A Hitchhiker’s Guide to International Macroeconomic Policy Coordination In Search of the Elusive Credit View: Testing for a Credit Channel in Modern Great Britain by Katherine A. Samolyk Seasonal Borrowing and Open Market Operations by E J. Stevens ■ 1990 Q uarte r 3 by Ow en F. Humpage Public Infrastructure and Regional Economic Development by Randall W. Eberts Using Market Incentives to Reform Bank Regulation and Federal Deposit Insurance A Reexamination of the Relationship between Capacity Utilization and Inflation by Paul W. Bauer Financial Restructuring and Regional Economic Activity by Brian A. Cromwell by James B. Thomson The Short-Run Dynamics of Long-Run Inflation Policy by John B. Carlson, William T. Gavin, and Katherine A. Samolyk Eil Fourth Quarter Working Papers Current Working Papers of the Cleveland Federal Reserve Bank are listed in each quarterly issue of the Economic Review. Copies of specific papers may be re quested by completing and mail ing the attached form below. Single copies of individual papers will be sent free of charge to those who request them. A mailing list service for personal subscribers, however, is not available. Institutional subscribers, such as libraries and other organiza tions, will be placed on a mail ing list upon request and will automatically receive Working Papers as they are published. ■ ■ ■ 9011 9013 Information and Voting Power in the Proxy Process by Sanjai Bhagat and Richard H. Jefferis W hat Does the Capital In come Tax Distort? by Jinyong Cai and Jagadeesh Gokhale ■ ■ 9012 The Effect of Subordinated Debt and Surety Bonds on Banks’ Cost of Capital and on the Value of Federal Deposit Insurance by William P. Osterberg and James B. Thomson 9014 ■ Cointegration and Transformed Series by Jeffrey J. Hallman ■ 9016 Some Problems of Infinite Regress in Social-Choice Models: A Category Theory Solution by Fadi Alameddine 9017 An Insider’s View of the Political Economy of the Too Big to Fail Doctrine by Walker F. Todd and James B. Thomson 9015 Loan Commitments and Bank Risk Exposure by Robert B. Avery and Allen N. Berger ■ 9018 Pro-Integrative Subsidies and Housing Markets: Do Race-Based Loans Work? by Brian A. Cromwell Please complete and detach the form below and mail to Research Department Federal Reserve Bank of Cleveland P.O. Box 6387 Cleveland, Ohio 44101 Check item (s) requested Please send the following Working Paper(s): □ □ □ 9011 9012 9013 □ □ □ 9014 □ 9015 □ 9016 9017 9018 Send to: Please print _______ Name Address City State Zip