In a recent article, Abrams, Clarke, & Settle (1999, henceforth ACS) examine the relationship between economic growth of U.S. states and changes in banking structure and fiscal policies during the 1950-80 period. They find that the depth of financial assets (particularly commercial bank depth) in a state significantly and positively affects real per capita income growth while restrictions over branch banking and multibank holding companies do not. In addition, ACS find no support for the hypothesis that fiscal policy variables affect income growth.

This paper reexamines the ACS growth model and findings, with emphasis on some specification issues such as multicollinearity and endogeneity problems. A parsimonious model will be tested on the more recent time period 1980-97. Contrary to ACS, we find that changes in banking policies and fiscal spending are important in explaining changes in state per capita income.

ACS use as their general framework a standard neoclassical growth model following Barro (1997) and Barro and Sala-i-Martin (1992). In this model, an area’s per capita real income growth (dy/y) is a function of the initial level of income (y₀) and a longrun "target income" (y*). In turn, y* depends on a vector of explanatory variables including sectoral composition and human capital. Following earlier literature, ACS add banking entry restrictions (branches and multibank holding companies or MBHC), financial depth, types of financial institutions, and fiscal policies resulting in the following augmented neoclassical model (ACS, 1999, p. 373):

dy/y = f(y₀, SECTORAL COMPOSITION, FINANCIAL DEPTH [or COMMERCIAL BANK DEPTH and S&L DEPTH], WELFARE SPENDING, NONWELFARE SPENDING, DEFICIT, BURDEN, STATEWIDE BRANCHING, LIMITED BRANCHING, STATEWIDE MBHC, LIMITED MBHC, HUMAN CAPITAL, STATE DUMMIES) (1)

The basic methodology employed by ACS is similar to that adopted by Jayaratne & Strahan (1996) with two major differences: ACS include financial deepening and mix of financial institutions as determinants of economic growth, and they focus on the 1950-80 period (to delineate effects of commercial banks vs. S&Ls) while Jayaratne & Strahan look at 1972-92. ACS estimated the above specification using a fixed effects model applied to pooled data for 48 states. They subdivided the 1950-80 period into six 5-year intervals for a total sample of 288 observations.

There are several empirical concerns regarding the ACS model and its estimation. First, regression of the full ACS growth model yielded a number of insignificant estimates. Of the 12 main explanatory variables, eight have coefficients that are statistically insignificant. The control variables - initial income, sectoral composition, human capital - are highly correlated with income growth but this result is generally accepted knowledge. The finding that financial depth, and specifically its commercial bank component, is a major determinant of state-level economic growth is consistent with previous studies. We contend that much of ACS’ insignificant results is due to a degrees of freedom problem and to correlation between the independent variables (ex. the fiscal spending and tax variables, and the various banking dummy variables). Second, the ACS results are biased by the endogeneity of an explanatory variable. For example, financial depth can also be affected by economic growth. A third problem is the choice of the 1950-80 time period which ACS defends as "sufficiently long to allow long-run influences" and "avoids any confounding influences of the Depository Institutions Deregulation and Monetary Control Act of 1980, which made S&Ls more competitive with commercial banks." (ACS, 1999, p. 371) By using an outdated period, the authors fail to analyze the effects of banking deregulation implemented in the 1980s. Moreover, changes in bank branching policies affected mainly commercial banks (see Jayaratne & Strahan, 1996, p. 661 and Avery et al., 1997). Finally, definition of some variables may not be appropriate. For example, the sectoral composition variable by construction is highly correlated with the dependent variable.

To address these empirical issues, we adopt a similar formulation as ACS and Jayaratne & Strahan but examine the 1980-97 period. The 1980-97 period is divided into the intervals 1980-90 and 1990-97 to reflect longer time horizons than that of ACS. We also follow the same variable definitions employed by ACS with some exceptions. First, the sectoral composition variable, which is a weighted measure showing the effect on a state’s income growth if all its sectors grew at their respective national sectoral growth rates, is constructed here using employment rather than income data. Moreover, this sectoral composition variable focuses only on the employment changes in the manufacturing sector, since this industry experiences wide structural dissimilarities across states and over time. Another revision relates to financial depth. Unlike ACS who define financial depth using total assets of commercial banks and S&Ls, we utilize the ratio of bank deposits to state income to measure financial depth; moreover, only commercial bank deposits are considered since the branch banking laws studied here primarily affected only commercial banks.

The following estimation and analysis involve panel data for 48 states and two time intervals, for a sample of 96 observations. Given that the fixed effects model uses up a substantial number of degrees of freedom, we attempt to transform the general ACS growth model (equation 1) to a more specific and parsimonious model consisting of only one control variable and a few explanatory variables of interest. This is achieved by first examining the correlation matrix and identifying highly correlated variables. The issue of multicollinearity is thus addressed by removing some of the redundant variables. This results in the following variant of the ACS model:

dy/y = f(FINANCIAL DEPTH, LBRANCH, UNIT, WELFARE SPENDING, BURDEN, SECTORAL COMPOSITION, TIME DUMMY 1990-97, STATE DUMMIES) (2)

where FINANCIAL DEPTH is commercial bank deposits as a percentage of state income; LBRANCH and UNIT are dummy variables for banking restrictions which allow limited or no intrastate branching (BRANCH was highly related to several variables and thus removed); WELFARE SPENDING is the ratio of public welfare spending to income; BURDEN is the ratio of state taxes to state spending; SECTORAL COMPOSITION is measured as the percent of total nonfarm employees in manufacturing multiplied by the national growth rate in manufacturing. DEPTH, WELFARE, and BURDEN are converted to logarithmic form following ACS. Data are taken from various issues of the Statistical Abstract of the United States.

Since ACS found only FINANCIAL DEPTH (after controlling for initial income, sectoral composition, and human capital) to be significant in explaining income growth, we begin our analysis with the simplified equation:

dy/y = f(FINANCIAL DEPTH, SECTORAL COMPOSITION, TIME DUMMY, STATE DUMMIES) (3)

and then sequentially enter various indicators of bank branching and fiscal policies to test for their independent, contributory effects on income growth.

Another empirical concern expressed earlier was that the explanatory variables in equation 3 may not all be strictly exogenous. In particular, FINANCIAL DEPTH could be influenced by economic growth and the model would then be simultaneous. State-level financial depth is postulated here to be a function of banking structure (as measured by the state banking concentration ratio) and per capita income. To test for endogeneity or simultaneity with respect to the DEPTH variable, a Hausman two-step test of endogeneity (see Pindyck & Rubinfeld, 1991, p. 303-305, for discussion) is performed. First, FINANCIAL DEPTH is regressed on the state concentration ratio (based on the domestic banking assets of the five largest firms in a state and taken from Amel, 1989), SECTORAL COMPOSITION, time, and state dummies, and then the estimated residuals are defined to form a "residual variable". In the second step, the residual variable from the first regression is added to the economic growth equation (3). If the coefficient of the residual variable is statistically significant, then simultaneity exists. This "corrected" model (with the residual variable) is the robust, parsimonious specification since it adjusts for the endogeneity of the financial depth variable. The results from this procedure are identical to those estimated using two-stage least squares (see Wooldridge, 2000, p. 484).

The results of applying the fixed effects technique (with cross-section weights to account for heteroscedasticity) on three alternative models are presented in Table 1. Model 1 is the base model (3) described above. Model 2 adds the bank branching variables LBRANCH and UNIT to model 1. Similarly, Model 3 enters the fiscal factors WELFARE SPENDING and BURDEN to model 1. Each model is tested for potential endogeneity of the FINANCIAL DEPTH variable.

The coefficients on the commercial bank depth variable are consistently positive and statistically significant in all three models. The Hausman test also shows that financial depth is endogenous, as evidenced by the highly significant t-statistics of the residual variables. Including the residual variable in the model therefore corrects for or removes the endogeneity feature of the financial depth variable.

Contrary to ACS, the results of Model 2 indicate that limited bank branching and no branching have a negative and significant impact on state economic growth. Removal of these entry restrictions (including BRANCH into the base model; results not shown here), however, positively affects economic growth and is consistent with earlier findings by Jayaratne & Strahan, 1996).

Model 3 reveals that state-level income is positively and significantly related to public welfare spending (in contrast to ACS) but is not influenced by the ratio of taxes to state spending.

Overall, the sectoral composition variable suggests that states heavily dependent on manufacturing have lower growth rates. Finally, the time dummy indicates that state income growth is significantly higher than that in the previous decade.

In their 1999 study, Abrams, Clarke, & Settle concluded that financial depth (specifically commercial bank depth) matters for state-level economic growth while changes in banking laws and fiscal activities do not. This paper evaluated and criticized ACS’ methodology and conclusion on several empirical issues. First, it argued that a more parsimonious specification of the model could be derived by removing redundant explanatory variables to correct for multicollinearity problem. Second, it argued that financial depth is endogenous. A Hausman test confirmed this and several model specifications corrected for simultaneity were estimated. Third, the variables measuring sectoral composition and financial depth were redefined to avoid a tautological correlation with the dependent variable. Moreover, this study covered the 1980-97 period (instead of 1950-80 for ACS) when banking deregulation in most states occurred. Finally, and most important, we find that financial depth, bank branching policies, and state government spending on public welfare are key determinants of state-level economic growth. These results suggest that there is a two-way causation between financial depth and state economic growth and that entry restrictions in the banking sector have a negative impact on economic growth.

Abrams, B. A., Clarke, M. Z., & Settle, R. F.(1999). The impact of banking and fiscal policies on state-level economic growth. Southern Economic Journal, 66,367-378.

Amel, D. F. (1989). Trends in banking structure since the mid-1970s. Federal Reserve Bulletin, 75, 120-135.

Avery, R. B., Bostic, R., Calem, P., & Canner, G. (1997). Changes in the distribution of banking offices. Federal Reserve Bulletin, 83, 707-725.

Barro, R. J. (1997). Determinants of economic growth: A cross-country empirical study. Cambridge, MA: The MIT Press.

Barro, R. J. & Sala-i-Martin, X. (1992). Convergence. Journal of Political Economy, 100, 223-51.

Jayaratne, J. & Strahan, P.E. (1996). The finance-growth nexus: Evidence from bank branch deregulation. Quarterly Journal of Economics, 111, 639-670.

Pindyck, R.S. & Rubinfeld, D.L. (1991). Econometric models and economic forecasts. New York: McGraw-Hill, Inc.

U.S. Bureau of the Census. (1981-97). Statistical abstract of the United States. Washington, DC: U.S. Government Printing Office.

Wooldridge, J. M. (2000). Introductory econometrics: A modern approach. U.S.: South-Western College Publishing.

Note. T-statistics are in parentheses. Hausman-Depth is the Hausman test t-statistic for the financial depth

variable given the null hypothesis that depth is not endogenous.