An empirical analysis of federal budget deficits and interest rates directly affecting savings and loans.

I. Introductory Remarks

The crisis in the Savings and Loan (S&L) industry in recent years has reached dimensions unmatched since the Depression years prior to 1934. In the period from 1934 through 1989, there were some 14 years when not a single federally insured S&L failed; furthermore, over the same period, there were some eight years when only one federally insured S&L failed. By contrast, during the 1980s alone, some 525 federally insured S&Ls failed.

Studies such as Carron [5] and Barth [1] argue that the rise in the S&L failure rate can to some significant degree be traced to interest rates (especially the cost of funds to S&Ls), which, particularly in the late 1970s and early 1980s, rose to unexpectedly high levels. The basic purpose of this study is to ascertain whether the federal budget deficit played a role in this escalation of interest rates for the S&Ls and thereby may have affected the health of the S&L industry. Although there is a rich literature on the interest-rate impact of federal budget deficits [3; 4; 6; 7; 10; I 1; 13; 15; 16; 17; 18], none of the published research has investigated the impact of the budget deficit on interest rate measures directly faced by S&Ls, such as the cost of funds to the S&Ls, the yield on new home mortgages at S&Ls, or the S&L mortgage portfolio yield. Accordingly, this study investigates empirically the impact of federal budget deficits upon these interest rate measures that directly affect S&Ls.

II. Model

The model adopted in this paper parallels the loanable funds model found in the well-known study by Hoelscher [11, which contains a number of components found in many other loanable funds analyses [3; 4; 7: 10; 17]. The basic Hoelscher model is given by:

R = R (EP, ERSR, RDEF, Y) (1) where:

R = the nominal rate of interest;

EP = expected future inflation;

ERSR = the ex ante real interest rate yield on U.S. Treasury bills;

RDEF= the real federal budget deficit (N.I.P.A.);

Y = the change in per capita real GNP.

Based upon Hoeischer [10; 11], as well as the standard loanable funds model, it is argued that R is an increasing function of both expected inflation and the real budget deficit. Based upon various arguments in Hoelscher [11], it is also expected that R is an increasing function of ERSR, whereas the impact of Y on R is a priori unknown.

The analysis in this paper focuses upon the cost of funds to S&Ls on the one hand (COST) and the yield on new home mortgages at S&Ls (NEW) or the S&L mortgage portfolio yield (MORT) on the other hand. Predicated upon equation (1) above, the quasi-reduced-form equations to be estimated are given by:

[COST.sub.t] = [a.sub.0] + [a.sub.1.EP.t] + [a.sub.2.ERSR.sub.t] + [a.sub.3.RDEF.sub.t] + [a.sub.4.Y.sub.t] + [a.sub.5.sub.TR] + [u.sub.1] (2)

[NEW.sub.t] = [b.sub.0] + [b.sub.1.EP.sub.t] + [b.sub.2.ERSR.sub.t] + [b.sub.3.RDEF.sub.t] + [b.sub.4.Y.sub.t] + [b.sub.5.TR] + [u.sub.2] (3)

[MORT.sub.t] = [c.sub.0] + [c.sub.1.EP.sub.t] + [c.sub.2.ERSR.sub.t] + [c.sub.3.RDEF.sub.t] + [c.sub.4.Y.sub.t] + [c.sub.5.TR ]+ [u.sub.3] (4)

where:

[COST.sub.t] = the average nominal cost of funds to S&Ls in year t, expressed as a percent per annum;

[NEW.sub.t] = the average nominal yield on new home mortgages at S&Ls in year t, expressed as a percent per annum;

[MORT.sub.t]= the average nominal S&L mortgage portfolio yield in year t, expressed as a percent per annum;

[a.sub.0, b.sub.0, c.sub.0] = constants

[EP.sub.t] = the expected rate of inflation of the CPI in year t, expressed as a percent per annum;

[ERSR.sub.t] = the ex ante real interest rate yield in year t on three-month U.S. Treasury bills, expressed as a percent per annum;

[RDEF.sub.t] = the U.S. federal budget deficit N.I.P.A.) in year t, expressed in billions of 1982 dollars;

[Y.sub.t] = the change in year t in the level of per capita real GNP, expressed in 1982 dollars;

[TR] = a simple trend variable;

[u.sub.1, u.sub.2, u.sub.3] = stochastic error terms.

The variables [COST.sub.t] and [MORT.sub.t], were obtained from the Office of Thrift Supervision's publication the 1989 Savings & Home Financing Source Book [14], whereas the variable [NEWT.sub.t] was obtained from The Economic Report of the President, 1991 [8]. The inflationary expectations variable is the Livingston survey data and was obtained from the Federal Reserve Bank of Philadelphia. The variable ERSR, was computed by subtracting the expected inflation rate in year t from the average nominal yield on three-month U.S. Treasury bills in year t. The three-month Treasury bill rate was obtained from The Economic Report of the President, 1991, as were the budget deficit data and the GNP deflator used to convert the budget deficit data into 1982 dollars. Finally, the data used to compute the variable [Y.sub.t] were also obtained from The Ecollomic Report of the President, 1991. The model to be estimated is annual since some of the data are available only in annual form. The time period studied runs from 1965-1989; data on such variables as [COST.sub.t] and [MORT.sub.t] are not available prior to 1965.

The Hausman [9] specification test rejects the null hypothesis of exogeneity for the cases of variables [RDEF.sub.t] and [EP.sub.t]; consequently, both of these variables are treated here as endogenous. The finding that the federal budget deficit is at least partly endogenous comes as no surprise. For instance, as the economy expands, tax collections automatically rise and government transfer payments (especially unemployment benefits) automatically decline. Treating the budget deficit as endogenous is consistent with other studies [6; 7; 15; 18].

Given the endogeneity of variables [RDEF.sub.T] and [EP.sub.t] equations (2), (3), and (4) are each estimated using an instrumental variables (IV) technique, with the instruments being, respectively, the one-year lag of the unemployment rate of the civilian labor force ([U.sub.t] - 1]) and the one-year lag of the actual inflation rate of the CPI ([P.sub.t] - 1]). The choice of instruments ([U.sub.t] - 1) and [P.sub.t] -1 is based upon the findings that [U.sub.t] - 1] systematically explains the budget deficit and [P.sub.t] - 1] systematically explains expected inflation, whereas the contemporaneous error terms in the system are not correlated with the lagged instruments.(1) The instrumental variables [U.sub.t] - 1 and [P.sub.t] - 1] were obtained from The Economic Report of the President, 1991.

III. Findings

Estimating equations (2), (3), and (4) each by IV, adopting in each case the two instruments [U.sub.t] - 1] and [P.sub.t] - 1 to allow for the endogeneity of the variables [RDEF.sub.t] and [EP.sub.t], respectively, yields equations (5), (6), and (7), respectively:

[COST.sub.t] = 1. 95 + 0.51[EP.sub.t] + 0.58[ERSR.sub.t] + 0.29[RDEF.sub.t] - 0.00001[Y.sub.t] (+5.96) (+ 6.04) (+4.59) (-1.187) -0.07TR, DW = 1.97, Rho = -0.01 (5) (-1.34)

[NEW.sub.t] = 3.52 + 0.80[EP.sub.t] + 0.72[ERSR.sub.t] + .037[RDEF.sub.t] - 0.00001[Y.sub.t] (+6.71) (+5.37) (+5.23) (-1.96) - 0. 17TR, DW = 1.92, Rho = 0.03 (6) (-2.45)

[MORT.sub.t] = 4.55 + 0.66[EP.sub.t], + 0.31[ERSR.sub.t] + 0.024[RDEF.sub.t] - 0.00001[Y.sub.t] (+ 3.74) (+4.37) (+5.18) (-0.71) + 0.04TR, DW = 2.00, Rho = -0.02 (7) (+ 1.06)

where terms in parentheses beneath coefficients are t-values.

Of the estimated coefficients shown in equations (5), (6), and (7), nine are statistically significant at the one percent level with the expected sign, and one is significant at the three percent level. In addition, none of the coefficients on the change in per capita real GNP variable ([Y.sub.t]) is statistically significant at even the five percent level. This finding is not at all surprising in view of the argument in the original Hoelscher [11] study that the sign on Y, is a priori unknown; this finding is also consistent with the empirical results obtained in the study by Hoelscher [11]. The simple trend variable (TR) is significant in one of the three cases. On the other hand, in all three estimations, the coefficients on the variables [EP.sub.t] and [ERST.sub.t] are positive (as expected) and statistically significant at the one percent level; thus, it appears that higher expected inflation and a higher ex ante real interest rate yield on three-month U.S. Treasury bills both act to raise the levels of [COST.sub.t] [NEW.sub.t] and [MORT.sub.t]. In addition, and more importantly (at least, from the viewpoint of the purpose of this study), the estimated coefficient on the deficit variable ([RDEF.sub.t]) is positive (as hypothesized) and statistically significant at the one percent level in all three of the estimations; thus, it appears that the real federal budget deficit in the United States acts to elevate each of these three interest rate measures ([COST.sub.t], [NEW.sub.t], and [MORT.sub.t) that affect the profitability and ultimately the solvency of S&Ls in the United States.(2) In principle, this finding of a positive impact of the budget deficit on interest rate measures is consistent with the studies by Barth, Iden, and Russek [3; 4], Hoelscher [11], Cebula [6; 7], Tanzi [16], and Zahid [17].

The evidence in equations (5), (6), and (7) indicates that the following three measures of nominal rates that affect S&Ls, namely,

(a) the average cost of funds to S&Ls (which consists of a variety of instruments and a variety of maturity structures),

(b) the average yield on new home mortgages (some of which are fixed-rate and some of which are variable-rate), and

(c) the average S&L mortgage portfolio yield (which deals with a variety of maturities, rates, and rate-types),

are all increasing functions of the federal government budget deficit in the United States.(3)

Of course, although it appears that the federal budget deficit in the United States does in fact impact upon these three average interest rate measures, it remains to be seen whether the deficit has a net negative adverse impact upon the S&Ls; this is because whereas the deficit apparently, for example, raises [COST.sub.t], and thus S&L costs, it also raises [NEW.sub.t] (and [MORT.sub.t]), and thus S&L revenues. In order to test empirically whether the budget deficit exercises, on the average a net negative or net positive impact upon the S&L cost/revenue structure, we initially examine here, in effect, the term structure of S&L interest rates as one would describe same in terms of the two variables [COST.sub.t] and [NEW.sub.t]. Specifically, we first subtract equation (2) from equation (3); this yields:

[CN.sub.t] = [d.sub.0] + [d.sub.2][EP.sub.t] + [d.sub.3][RDEF.sub.t] + [d.sub.4][Y.sub.t] + [d.sub.5]TR + [u.sup.*] (8)

where:

[CN.sub.t] = [NEW.subt] - [COST.sub.t];

dj = bj - aj, i = 0, ..., 5;

[u.sup.*] = [u.sub.2] - [u.sub.1].

Regression equation (8) in effect models the slope of the S&L yield curve as a function of expected inflation, the ex ante real interest rate yield on three-month U.S. Treasury bills, the real federal budget deficit, the change in per capita real GNP, and a simple trend variable. Once again, we study the period 1965-1989.

Estimating equation (8) by IV, using the same instruments as above, yields:

[CN.sub.t] = 1. 57 + 0.29[EP.sub.t] + 0.14[ERSR.sub.t] + 0.008[RDEF.sub.t] - 0.00001[Y.sub.t]

                       (+4.62)         (+2.50)             (+1.81)             (
-1.57)
                       - 0.11TR,         DW = 1.56, Rho = 0.16.
               (9)
                        (-2.82)

As shown in equation (9), the estimated coefficient on variable [RDEF.sub.t] is positive but statistically significant at only the .085 level.(4) Thus, although the budget deficit appears to raise both [COST.sub.t] and [NEW.sub.t], the evidence that it raises the slope of the S&L yield curve is not especially convincing. Thus, we infer that, despite the fact that the budget deficit apparently contributed to higher S&L costs (by acting to generate a higher level of [COST.sub.t]), there is no convincing evidence of a net negative or net positive impact of the budget deficit on the S&Ls (since the deficit also pushed upwards on the level of [NEW.sub.t]).

The same basic conclusion is reached if we examine the issue at hand in terms of the variables [COST.sub.t] and [MORT.sub.t] rather than [COST.sub.t] and [NEW.sub.t]. In particular, we seek next to test whether the budget deficit exercises a net negative or net positive impact on the "term structure of S&L interest rates" as one would describe such in terms of [COST.sub.t] and [MORT.sub.t]. Following in principle the same basic procedure as above, we first subtract equation (2) from equation (4); this yields:

[CM.sub.t] = [e.sub.0] + [e.sub.1][EP.sub.t] + [e.sub.2][ERSR.sub.t] + [e.sub.3][RDEF.sub.t] + [e.sub.4][Y.sub.t]

+ [e.sub.5]TR + [u.sup.**] (10) where:

[CM.sub.t] = [MORT.sub.t] - [COST.sub.t];

        ej = ci - aj, j = 0, . . ., 5;
        [u.sup.**] = [u.sub.3] - [u.sub.1].

Estimating equation (10) by IV, using the same instruments as above, yields the following:

[CM.sub.t] = 1.34 + 0.17[EP.sub.t] + 0.21[ERSR.sub.t] + 0.007[RDEF.sub.t] - 0.00001[Y.sub.t]

                      (+3.12)          (+2.69)            (+1.70)             (-
1.60)
                      -0.10TR,          DW = 1.67,      Rho = 0. 13.
                (11)
                      (-2.70)

As shown in equation (11), the estimated coefficient on variable [RDEFT.sub.t.] is positive but significant at barely the .09 level. This finding implies that, although the federal budget deficit appears to raise both [COST.sub.t] and [MORT.sub.t], the evidence that it raises the slope of the yield curve is unconvincing. The conclusions derived from equation (11) are entirely consistent with those in equation (9) above.(5) Thus, we again are led to infer that there is no convincing evidence of a net negative or net positive impact of the federal budget deficit upon the S&Ls.

One potentially useful and reasonable "reality check" as to whether this conclusion is valid might consist of a direct test of the impact of the budget deficit on the S&L failure rate. To undertake such a test, we rely on the well-known study by Barth [1] in which principal causes of the S&L crisis are outlined and discussed. Among the central factors Barth [1] identifies as leading to the massive S&L failures of the 1980s are the following: the 1981-82 recession, declining crude petroleum prices, declining capital requirements, and federal deposit insurance. Barth [1] argues that the 1981-82 recession acted to significantly hurt the real estate industry in the United States and thereby to reduce S&L profitability and solvency and to raise S&L failures. The decline in crude oil prices beginning in the early 1980s and the especially sharp decline in those prices between 1985 and 1986 hurt both the economies of the Southwest and the real estate market in the Southwest; the outcrop of these developments is argued by Barth [1] to be a decline in S&L profitability and solvency and a rise in the S&L failure rate, especially in the Southwest. Barth [1] also argues that declining capital requirements for S&Ls resulted in a "thin" capital cushion that exposed the S&Ls to a higher failure risk. Finally, like Kane [11] and others, Barth [1] argues that higher levels of federal deposit insurance acted to encourage risk-taking behavior by the S&Ls and thereby significantly contributed to their incidence of failure.

To empirically examine the impact of the federal budget deficit, within the context of the Barth [1] framework, upon the S&L failure rate, we here simply estimate the following reduced-form equation:(6)

[SLR.sub.t] = [f.sub.0] + [f.sub.1][RDEF.sub.t-1] + [f.sub.2][REC.sub.t-1] + [f.sub.3] [RP.sub.t-2] + [f.sub.4][CAP.sub.t-2]

= [f.sub.t][INS.sub.t-1] + [f.sub.6]TR + [u.sub.***]

(12) where:

[SLR.sub.t] = the percentage of federally insured S&Ls that failed in year t;

[f.sub.0] = constant term;

[RDEF.sub.t-1] = the federal budget deficit in year t - 1, expressed in billions of 1982 dollars;

[REC.sub.t-1] = a binary (dummy) variable indicating whether the U.S. economy was in the 1981-82 recession in year t - 1, if the economy was in the 1981-82 recession in year t - 1, then [REC.sub.t-1] = 1; otherwise, [REC.sub.t-1] = 0;

[RP.sub.t-2] = the price per barrel of imported crude oil in year t = 2, expressed in 1982 dollars;

[CAP.sub.t-2] = the average S&L capital-to-asset ratio in year t - 2; this ratio of required capital to assets is expressed as a percent;

[INS.sub.t-1] = the FSLIC insurance ceiling per deposit in year t - 1, expressed in thousands of 1982 dollars;

TR = as above;

[u.sup.***] = stochastic error term.

The basic data sources for this model are the Statistical Abstract of the United States (various issues), the 1989 Savings & Loan Home Financing Source Book, and Barth [1; 2]. The model expressed in equation (12) is a lagged regression equation, reflecting the fact that these various factors do not impact upon the S&L failure rate instantaneously but rather with a time lag [1]. Paralleling the analysis of equations (2), (3), (4), (8), and (10), we again study the period 1965-1989.

Estimating equation (12) by OLS yields:

[SLR.sub.t] = 16.98 + 0.01[RDEF.sub.t-1] + 3.68[REC.sub.t-1] - 0.32[RP.sub.t-2]

                        (+1.71)              (+4.17)             (-5.66)
                        -2.21[CAP.sub.t-2] + 0.03[INS.sub.t-1] - 0.18TR,
                        (-6.37)              (+2.71)             (-2.84)
                        DW = 1.96, Rho = -0.07, [R.sup.2] = 0.84
            (13)

As shown in equation (13), the coefficients for Barth's four explanatory variables (represented by [REC.sub.t-1], [RP.sub.t-2], and [INS.sub.t-1]) are all statistically significant at the one percent level. However, the coefficient on the deficit variable, although positive, is significant at only the ten percent level. Thus, it appears that the federal budget deficit does not raise the S&L failure rate to any significant degree.(7) This finding is consistent with the conclusion based on equations (9) and (11) above, namely, that on balance the federal budget deficit has not significantly impacted upon the overall economic well-being of the S&Ls per se.(8) Interestingly, Barth [1] did not identify the budget deficit as contributing to the S&L crisis; thus, it appears that our conclusions (results) are also consistent with and indeed even supportive of the latter analysis.

IV. Conclusion

It appears that the federal budget deficit in the United States has acted to raise the levels of various interest-rate yield measures that are directly relevant to S&Ls, namely, the S&L cost of funds ([COST.sub.t]), the rate on new home mortgages issued by S&Ls ([NEW.sub.t]), and the S&L mortgage portfolio yield ([MORT.sub.t). However, the empirical evidence rather clearly indicates that, on balance, these impacts have not affected the S&L failure rate in any perceptible way. Indeed, it appears instead that factors such as those identified in equations (12) and (13), other than the federal budget deficit, may be (are) the real culprits in the S&L crisis in this country.

References

[1.] Barth, James R. The Great Savings and Loan Debacle. Washington, D.C.: The AEI Press, 1991. [2.] _____, Statement before the House Committee on Banking, Finance, and Urban Affairs. 101st Congress. 11 April 1990. [3.] _____, George Iden, and Frank Russek, "Do Deficits Really Matter?" Contemporary Policy Issues, Fall 1984, 79-95. [4.] _____, "Federal Borrowing and Short Term interest Rates: Comment." Southern Economic Journal, October 1985,554-59. [5.] Carron, Andrew S. The Plight of the Thrift Institutions. Washington, D.C.: The Brookings Institution, 1982. [6.] Cebula, Richard J., "Federal Government Budget Deficits and Interest Rates: A Brief Empirical Note." Southern Economic Journal, July 1988, 206-10. [7.] _____, "Federa; Government Fiscal Actions and the Cost of Municipal Borrowing." Quarterly Review of [8.] The Economic Report of the President, 1991. Washington, D.C.: U.S. Government Printing Office, 1991. [9.] Hausman, Jerry, A., "Specification Tests in Econometrics." Econometrica, October 1978, 125-51. [10.] Hoelscher, Gregory, "Federal Borrowing and Short Term Interest Rates." Southern Economic Journal. October 1983.319-33. [11.] _____, "New Evidence on Crowding Out." Journal of Money, Credit, and Banking, February 1986, 1-17. [12.] Kane, Edward, "S&Ls and Interest-Rate Reregulation: The FSLIC as an in-place Bailout Program." Housing Finance Review, July 1982, 219-43, [13.] Makin, John, "Real Interest, Money Surprises, Anticipated Inflation and Fiscal Deficits." Review of Economics and Statistics, May 1983, 374-84, [14.] Office of Thrift Supervision. 1989 Savings & Home Financing Book. Washington, D.C.: U.S. Government Printing Office, 1989. [15.] Ostrosky, Anthony, "Federal Budget Deficits and Interest Rates: Comment." Southern Economic Journal, January 1990, 802-801 [16.] Tanzi, Vito, "Fiscal Deficits and Interest Rates in the United States: An Empirical Analysis. 1960-1984." IMF Staff Papers, December 1985, 551-76. [17.] Zahid, Khan, "Government Budget Deficits and Interest Rates: The Evidencel Since 1971, Using Alternative Deficit Measures." Southern Economic Journal, April 1988, 725-31 [18.] Zelhorst, Dick and Jacob de Haan, "Federal Government Budget Deficits and Interest Rates: Comment." Public Finance/Finances Publiques, No. 2 1991, 324-30.

(1.)Using as the instruments [U.sub.t-2] in lieu of [U.sub.t] - 1] and [P.sub.t] -1 leaves our basic results essentially unchanged. (2.) OLS estimates of equations 2), (3), and (4) yield these same basic conclusions. (3.) It should be noted that expressly allowing in the [COST.sub.t] regression for the potential impact of Regulation Q leaves the conclusions unchanged. in this situation, we have allowed for Regulation Q using a dummy variable on the one hand or actual Regulation Q ceiling values for passbook savings on the other hand. (4.) Lagging the deficit variable yields even smaller t-values for the estimated coefficient on variable RDEF in equation (9). (5.) First-differences estimates of equations (8) and (10) also indicate that the federal budget deficit does not significantly affect the S&L term structure [as defined in (8) and (10)]. (6.) In a separate study, one using cointegration and a larger and somewhat different data set, we have found that Barth's [1] arguments as to the causal factors in the S&L crisis are valid. Likewise, as in the OLS estimation provided here, cointegration reveals no link between the budget deficit and S&L failures. (7.) The same conclusions are obtained if we regress the number of S&L failures against these same variables. (8.) It should be noted that we have also empirically investigated the impact of the federal budget deficit upon the volatility of [COST.sub.t], [NEW.sub.t], and [MORT.sub.t]; in all instances, the deficit did not exercise a statistically significant impact.