بررسی اقتصادسنجی ارتباط میان جریان سرمایه و کسری بودجه The Relationship between Budget Deficits and Capital Inflows: Further Econometric Evidence

This research note examines the impact offederal deficits on U.S. capital inflows. Expanding on the previous work of Bahmani-Oskooee and Payesteh (1994), we employ the relatively new maximum likelihood procedure developed byJ o h ansen (I 988) and Johansen and Juselius (I 990) to do cointegration tests. The results find a long run relationship between blcdget de&its and capital inflow. In addition, findings j?om error-correlation modeling reveal that short&m d&equilibria in financial markets are corrected very rapidly, suggesting that these markets are efficient. I. INTRODUCTION In a recent issue of this journal Bahmani-Oskooee and Payesteh (1994) have shown that in the United States budget deficits led to capital inflows in the 1973- 88 period. Using the Engle-Granger (1987) residual-based two-step cointegration procedure and error-correction model, they found budget deficits and capital inflows to be cointegrated. They attributed the relationship between budget deficits and capital inflows to rising interest rates caused by public borrowing and to improved expectations based on increased economic growth resulting from budget stimulus. The evidence provided by Bahmani-Oskooee and Payesteh (hereafter BP) has made an important contribution to our understanding of the effect of budget deficits on the economy. The purpose of this note is to extend the evidence on the relationship between budget deficits and capital flows by examining the data using the most recent cointegration procedure developed by Johansen (1988) and Johansen and Juselius (1990). Subsequent analysis has suggested that the Engle-Granger cointegration technique suffers from a number of econometric shortcomings. Our results further confirm the existence of a long-run link between budget deficits and capital inflows. Moreover, we find that short-run disequilibria in the relationship of budget deficits 485 486 QUARTERLY REVIEW OF ECONOMICS AND FINANCE and capital inflows are corrected very rapidly, suggesting that global capital markets are very efficient. II. THE METHODOLOGY The Engle-Granger cointegration procedure (hereafter EG) used by\ BP suffers from several econometric shortcomings. First, Banerjee, Dolado, Galbraith and Hendry (1993), Davidson and Mackinnon (1993), and Stock (1987) have shown there is considerable small-sample bias in estimates derived from the EG procedure. In addition, Davidson and Mackinnon contend that “a relatively low value of R* from the cointegration regression should be taken as a warning that the two-step procedure may not work well” (۱۹۹۳, p. 724). It has been demonstrated by Banerjee, Hendry and Smith (1986) and Banerjee et al. (1993) that the size of the small sample bias is inversely related to the magnitude of R* in the EC residual-based cointegrating regression. Since the BP data set consists of only 64 observations, it is vulnerable to this criticism. The cointegration results presented in their Table 2 have low R* values (.54 and .32), and BP warn that their results must therefore be treated cautiously (p. 68). The second weakness of the EC procedure is it ignores the possibility of multiple cointegrating relationships. Economic variables can exhibit more than one long-run relationship in a cointegrated equilibrium space. The third weakness of the EG method is that it relies very heavily on a super-convergence result and employs ordinary leastsquares estimation (hereafter OLS) to derive the parameter estimates of the long-run or cointegrating equation. However, OLS estimates are extremely sensitive to the arbitrary normalization implicit in the selection of the left-hand side variable of the cointegration regression equation (Enders, 1995; Rao, 1994; Banerjee et al., 1993; Muscatelli and Hurn, 19921. This suggests that different arbitrary normalizations can yield different empirical outcomes. A fourth dishculty with the EG procedure is it does not incorporate short-run dynamics in the cointegrating regression. Not accommodating short-run dynamics results in increased bias, loss of information and thus reduced efficiency of the parameters of interest in the cointegrating relationships. Finally, and most importantly, the EG procedure does not enable the researcher to test for various restrictions or exclusions on individual elements of the observed cointegrating vectors. In testing of hypotheses related to long-run economic relationships, this drawback of the EG procedure is a serious shortcoming. The maximum likelihood procedure of Johansen (1988) and Johansen and Juselius (1990)-hereafter J and JJ respectively-is able to overcome the aforesaid shortcomings of the EG procedure. Furthermore, as Gonzalo (1994) has demonstrated in his Monte Carlo study, the Johansen procedure performs better than other estimators of long-run parameters even in the presence of nonnormal errors and unknown dynamics. The system-based procedure of J and JJ THE RELATIONAHIP BETWEEN BUDGET DEFICITS AND CAPITAL INFLOWS 487 provides a natural econometric framework for a combined analysis of the long and short-run behavior of variables of interest. In the JJ cointe~ation procedure, two tests, the maximal eigenvalue and trace tests, are used to determine the number of cointe~ating vectors (see Enders, 1995; Serletis, 1993; Juselius and I-iargreaves, 1992). In the maximal eigenvalue test we test the null hypothesis of r’ cointegrating vectors against the alternative r + 1 cointegrating vectors. In the trace test the null hypothesis is that there are at most r cointegrating vectors against a general alternative (see J, 1988; JJ, 1990). A detailed description and mathematical exposition of the Johansen maximum likelihood procedure can be found in Dickey, Jensen and Thornton (1994), Muscatelli and Hum (1992), J (1988) and JJ (1990). III. THE ESTIMATION AND RESULTS In exploring the relationship between budget deficits and capital flows, BP measure capita1 inflow (CAI) as seasonally adjusted net capital flows and budget deficits (BUS) as the seasonally and cyclically adjusted federal budget deficit.’ They analyze these nominal variables as well as the deflated variables CAIY and BUSY, where CAI and BUS are divided by gross national product. In this study we report the results for CAIY and BUSY. A number of researchers have suggested that the impact of the budget deficit should be judged relative to the size of the economy (Cebula and Belton, 1993; Cebula and Koch, 1989,1994; Cebula, 1990; Darrat, 1988; Evans, 1985). For this reason, and to facilitate comparison with those studies, we report the results from the ratio specification.2