In a recent issue of this Journal, Professor Cebula [2] empirically investigates the impact of federal deposit insurance on Savings and Loan (S&L) failures. His paper represents a useful scholarly investigation of the causes of the S&L crisis. However, his study contains an econometric flaw that casts doubt as to one of his conclusions, the conclusion that the failure rate of S&Ls is a positive function of the real federal deposit insurance ceiling level. In addition to discussing the econometric flaw, this comment presents some new estimates of the basic model in Cebula which show that the ceiling on federal deposit insurance does not significantly affect the failure rate of S&L's.

Cebula estimates several models in his paper. These models are based extensively on Barth [1]. The most basic model is given by:

[PSL.sub.t] = [a.sub.0] + [a.sub.1][INS.sub.t-2] + [a.sub.2][COST.sub.t-2] + [a.sub.3][RP.sub.t-2] + [a.sub.4][REC.sub.t-1] + [a.sub.5][CAP.sub.t-2] + u (1)

where:

[PSL.sub.t] = the percent of federally insured S&L's that failed in year t.

[a.sub.0] = constant term.

[INS.sub.t-2] = the federal deposit insurance ceiling per account in year t - 2, in thousands of 1982 dollars.

[COST.sub.t-2] = average cost of acquiring funds for the S&L industry in year t - 2, expressed as a percent per annum.

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

[REC.sub.t-1] = a binary dummy variable with a value of 1 for the two years following the 1981-82 recession.

[CAP.sub.t-2] = the required capital/asset ratio for an S&L in year t - 2, expressed as a percent.

u = the stochastic error term.

Using this model and several modifications of it, Cebula shows by estimating this model using OLS that the coefficient on [INS.sub.t-2] is positive and statistically significant, indicating that federal deposit insurance contributed to the S&L failure rate.

The problem with the OLS estimates of the model presented in Cebula [2] is that the cost of funds variable ([COST.sub.t-2]) is, in fact, endogenous. This is indicated by the results of the Hausman specification test [6]. Also, several recent papers on the determinants of nominal interest rates verify the endogeneity of [COST.sub.t-2] [3; 4; 5; 7]. Thus, the conclusion reached in Cebula [2] may be the result of simultaneous equation bias.

To correct for the possible simultaneity bias, equation (1) is re-estimated using an instrument variable (IV)...