The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Review is published 10 times per year by the Research and Public Information Department o f the Federal Reserve Bank o f St. Louis. Single-copy subscriptions are available to the public f r e e o f charge. Mail requests f o r subscriptions, back issues, or address changes to: Research and Public Information Department, Federal Reserve Bank o f St. Louis, P.O. Box 442, St. Louis, Missouri 63166. Articles herein may be reprinted provided the source is credited. Please provide the Bank’s Research and Public Information Department with a copy o f reprinted material. Federal Reserve Bank of St. Louis Review January 1983 In This Issue . . . This Review contains two articles that represent variations on a common theme. The theme is that, because “ money matters, ” its influence should be incorporated explicitly in macroeconomic analysis if better policy decisions are to be made. Both articles use comparative economic studies to demonstrate the validity o f this approach. In the first article, Dallas S. Batten and B. W. Hafer use a modified form o f the “ St. Louis equation” to assess the relative importance o f monetary and fiscal actions on econom ic activity in several developed countries. The purpose o f this cross-country comparison is to determine whether the well-known results for the United States— that monetary actions have a permanent influence on GNP growth while fiscal actions have no lasting influence whatsoever— are unique to the United States, or whether they apply to other economies as well. To adjust for the importance o f international trade (for the respective economies), Batten and Hafer modify the St. Louis equation, which relates growth in GNP to monetary and fiscal influences, by adding a third influence on GNP growth, the growth o f merchan dise exports. W hen Batten and Hafer examine the effects o f monetary and fiscal actions on GNP growth for Canada, France, Germany, Japan, the United Kingdom and the United States, they find that the U.S. results are generally supported across the other nations. Monetary actions have a significant and lasting effect on GNP growth in all six countries. Fiscal actions, on the other hand, are important only in two countries: France and the United Kingdom. Further, Batten and Hafer examine the stability o f the money-GNP growth relationship in each country over the change from fixed to floating exchange rates in the early 1970s. They find that this international policy shift had no effect on the significant money-GNP growth relationship in any o f the six countries examined. The stability o f this relationship in the face o f “ one o f the most significant interna tional policy shifts in the past two decades” provides evidence o f the robustness of the money-GNP relationship and demonstrates that the economic relationships summarized in the “ St. Louis equation” generally describe the economies of developed nations. In the second article, Keith M. Carlson and Scott E. Hein compare the long-run characteristics o f three well-known econometric models — Chase, DBI and Whar ton — to those o f the St. Louis model. The purpose o f this comparison is to determine whether the St. Louis model — an explicitly monetarist model — provides different implications about the long-run effects o f monetary policy from those generated by non-monetarist econometric models. The comparisons were based on the impacts o f four alternative monetary policy scenarios on simulations o f a number o f economic variables — including inflation rates, real GNP growth and nominal GNP growth. The simulations from the four models cover the period from 1987 to 1991, which represents the last five years o f a ten-year simulation for the 1982-1991 period. These simulations were chosen to 3 In This Issue . . . 4 represent the “ long-run” policy simulations; differences in the simulations across the models represent differences in the long-run characteristics o f the models. These long-run simulations also were compared to the actual relationships b e tween money growth and the same economic variables over the period from 1955 to 1981 to determine their consistency with the historical record. Significant deviations from the historical pattern is taken as evidence that there is some “ problem ” with the m odel’s ability to capture the long-run consequences o f monetary policy actions. Carlson and Hein find that none o f the four models consistently matches the historical record for all o f the relationships assessed. For example, none o f the large scale models’ simulations o f nominal income growth is generally consistent with the historical record; only the St. Louis model provides simulations that fit with past experience. On the other hand, the St. Louis model alone shows significant long-run variation in real output growth related to different monetary scenarios. This result runs counter to the historical evidence that long-run money growth has no lasting impact on real output. Despite the variation among the models’ long-run characteristics, one particu larly useful policy conclusion emerges from the comparative evaluation: there are no long-run econom ic gains from faster money growth. This conclusion holds, in general, whether the structure o f the model is explicitly “ monetarist” or not. The Relative Impact of Monetary and Fiscal Actions on Economic Activity: A Cross-Country Comparison DALLAS S. BATTEN and R. W. HAFER C o n s i d e r a b l e research has been devoted to assessing the empirical relationship between both monetary and fiscal actions and econom ic activity in the United States. Much o f this research was sparked by the controversial results obtained from investigat ing the impact o f monetary and fiscal actions on GNP using the “ St. Louis equation. ” 1 The St. Louis results can be summarized neatly: monetary actions have a significant, permanent effect on nominal GNP growth, while fiscal actions exert no statistically significant, lasting influence. This paper is a shortened version o f an earlier study presented in seminars at D e Nederlandsche Bank N. V ., Erasmus University and at the 1982 Southern E con om ic Association meetings. W e wish to express our thanks to all the participants at these sessions. 'T h e original articles presen tin g the controversial results are Leonall C . Andersen and Jerry L. Jordan, “ Monetary and Fiscal Actions: A Test o f Their Relative Importance In E con om ic Stabi lization,” this Review (N ovem ber 1968), pp. 11-24; and Leonall C. Andersen and Keith M. Carlson, “ A Monetarist M odel for E co nom ic Stabilization,” this R eview (April 1970), pp. 7-25. Substantially less work has been conducted within this framework for countries other than the United States.2 Consequently, it is uncertain whether the St. Louis approach can be used universally in evaluating the economic impact o f monetary and fiscal actions on income growth. This study investigates the generality o f the St. Louis approach by applying it to other countries. Based on evidence generated from the study o f six developed countries — Canada, France, Germany, Japan, the United Kingdom and the United States — we conclude that money growth is more important than fiscal actions in determ ining GN P growth. Moreover, our results are robust across the “ fixed” and “ flexible” exchange rate regimes that characterized the past two decades. ESTIMATING THE ST. LOUIS EQUATION ACROSS COUNTRIES The St. Louis equation typically estimated for the United States consists o f only three variables: nominal Early critics include Frank D e Leeuw and John Kalchbrenner, “ Monetary and Fiscal Actions: A Test o f Their Relative Importance in E conom ic Stabilization — C om m en t,” this R eview (April 1969), pp. 6-1 1 ; Richard G. Davis, “ H ow M uch D oes M oney Matter? A Look at Some Recent E vid en ce,” Federal Reserve Bank o f New York Monthly Review (June 1969), pp. 119-31; and Edward M. Gramlich, “The Usefulness o f M onetary and Fiscal Policy as D is cretionary Stabilization T ools,” Journal o f M oney, Credit, and Banking (May 1971), pp. 506-32. ^Two exceptions are Michael W . Keran, “ Monetary and Fiscal Influences on E con om ic Activity: The Foreign E xperience,” this R eview (February 1970), pp. 16-28; and W illiam G. Dew ald and Maurice N. Marchon, “ A M odified Federal Reserve Bank o f St. Louis Spending Equation for Canada, France, Germany, Italy, the United Kingdom and the United States,” Kredit und Kapital (Heft 2 1978), pp. 194-212. M ore recent sparring over the same issues is reported in Ben jamin M. Friedman, “ Even the St. Louis M odel N ow Believes in Fiscal Policy, ’’ Journal o f M oney, Credit, and Banking (May 1977), pp. 365-67; Keith M. Carlson, “ D oes the St. Louis Equation Now Believe in Fiscal Policy?” this Review (February 1978), pp. 13-19; and R. W . Hafer, “ The Role o f Fiscal Policy in the St. Louis Equation,” this Review (January 1982), pp. 17-22. Our approach differs from these and other works in that a) we focus solely on the growth-rate version o f the St. Louis equation (see footnote 6); b) w e jettison the com m only used polynomial estimation technique for unconstrained ordinary least squares (see footnote 8); c) w e explicitly examine the stability o f the underlying relationships from each country over time; and d) w e extend the sample period studied. 5 FEDERAL RESERVE BANK OF ST. LOUIS GNP, a variable summarizing monetary actions and one summarizing fiscal actions. Because the equation is formulated solely to test the relative efficacy of monetary and fiscal actions, it is not intended to in corporate all o f the exogenous forces that affect nomi nal GNP. Conceptually, therefore, the equation is misspecified. This conceptual misspecification poses a statistical problem, however, only if the omitted ex ogenous variables are correlated with the policy mea sures used in the equation.3 If, as assumed generally, the “ missing” exogenous variables are neither policy variables nor closely correlated with the variables rep resenting monetary and fiscal actions, their omission does not pose a serious statistical problem .4 This discussion implicitly assumes that the domestic economy being analyzed is relatively “ closed” to the rest o f the world. While this may adequately charac terize the United States, it is not true for countries whose exports account for a large proportion o f their GNP. In addition, because monetary and fiscal actions obviously affect the foreign sector, the correlation be tween external and domestic influences on GNP rises as the economy becomes more open. Consequently, these external influences should be included in analyz ing the relative impacts o f monetary and fiscal actions on GNP in such “ open” economies. In response both to past criticism o f the St. Louis equation and the likely interrelation o f domestic and 3For examples o f the specification error argument, see Franco Modigliani and Albert Ando, “ Impacts o f Fiscal Actions on Aggre gate Incom e and the Monetarist Controversy: Theory and Evi d e n ce,” in Jerome L. Stein, e d ., Monetarism , vol. 1, Studies in M onetary E con om ics (N orth-H olland, 1976), pp. 17-42; and Robert J. G ordon, “ Com m ents on Modigliani and Ando, ” in M one tarism, pp. 52-66. To understand the necessary condition for bias due to misspec ification, consider the following equation: (1') Y, = a,i + ai X , + et. N ow if equation 1' is not the “ true” model, but some other exoge nous variable, Z, has been om itted, the true m odel is: (2') Y, = b 0 + b] X, + b 2 Z, + T)t. Estimating equation 1' instead o f 2' yields an estimate o f a, with an expected value o f a] + b 2 where Xj is obtained by estimating (3') Z, = \() + X] X, + 4),. Obviously, the estimate o f a, is biased only if X, 4- 0, but X[ equals r „ (|^ where rv/ — the sim ple correlation coefficient betw een X and Z , and S, — the standard deviation o f i. Consequently, Xj ^ 0 only if rX7 i= 0; that is, X and Z must be correlated before the omission o f Z results in a specification error. 4This point also is made in Andersen and Jordan, “ Monetary and Fiscal A ctions,” p. 24. 6 JANUARY 1983 external influences on GNP in other countries, the following modified version o f the St. Louis equation is used: (1) Yt = «o J + 2 lrij M t_j K + i= 0 2 . g, Gt_j i= 0 L + 2 es E X t_ j + Et, i= 0 where Y, M, G and EX represent GNP, narrow money (M l), federal government expenditures and merchan dise exports, respectively.5 The dots above each vari able indicate that the equation is estimated in growth rate form.6 The appropriate lag lengths (J, K and L) are determined using an orthogonal regression procedure with sequential hypothesis testing.' Finally, one additional modification is made in esti mating the equation. The St. Louis equation typically is estimated with each distributed lag’s coefficients restricted to lie on a fourth-degree polynomial with endpoints constrained to equal zero. Because these constraints may not be valid across countries, we esti- T ’ ven though many countries included in this study do not explicit ly target the narrow (M l) definition o f m oney, this definition provides a consistent and com parable set o f explanatory variables across countries. Also, to rem ove the impact o f cyclical changes, h igh-em ploym ent governm ent expenditures is the measure o f fiscal policy action typically included in the estimation for the United States. Because com parable measures o f governm ent ex penditures do not exist for the other countries in the sample, federal governm ent expenditures that are not adjusted for cyclical changes are used for each country. It should be noted, however, that using either measure for the United States did not alter the conclusions reached in this paper. Furthermore, a criticism frequently leveled at using OLS to estimate equation 1 is that the right-hand-side variables are not exogenous with respect to GNP, resulting in simultaneous equa tion bias. This issue is addressed in an earlier, expanded version o f this paper through the use o f Granger-type causality tests. These tests did not indicate any causal relationship from incom e growth to m oney growth or governm ent expenditure growth in any o f the countries analyzed. A lternatively, in com e growth appears to “ cause” export growth in France and the United States, but not in the remaining countries. Statistically speaking, then, the esti mated parameters o f equation 1, as specified for the United States and France, may be biased. This does not appear to b e the case for the rest o f the sample. fiCarlson, “ D oes the St. Louis Equation N ow Believe in Fiscal Policy?” demonstrates that the original first-difference form o f the m odel, when updated through the 1970s, is plagued by heteroscedasticitv. This problem is not evident in the growth-rate version, however. T h is procedure involves a Gram-Schmidt orthogonalization o f the data and the use o f a testing procedure introduced by Marcello Pagano and Michael J. Hartley, “ On Fitting D istributed Lag M od els Subject to Polynomial Restrictions,” Journal o f Econom etrics (June 1981), pp. 171—98; and extended by Dallas S. Batten and D aniel L. Th orn ton , “ Polynom ial D istribu ted Lags and the Estimation o f the St. Louis Equation,” this R eview (forthcoming). FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 Table 1 Summary of Estimation Results1 Country and Sample Period Canada II/66—IV/81 France 11/65-111/81 Germany II/63—1/82 Japan II/60—II/80 United Kingdom 11/66-1/81 United States II/62-I/82 Constant —.006 (0.90) .007 (1.43) .007 (131) .010 (1.65) .001 (0.21) .007 (194) SM ,7262 (3.41) ,2892 (1.75) .5182 (3.50) .5522 (3.76) .4192 (2.50) 1,0942 (4.29) 2G -.011 (0.09) .1922 (1.90) -.2 2 5 (1.44) .006 (0.87) 3452 (2.90) -.1 9 9 (1.21) SEX .5432 (3.04) .246z (3.21) .2762 (2.48) .067 (1.65) 2092 (3.02) .114 (1.64) .41 Coefficient R2 .49 .82 .29 .19 .59 SE .006 .008 .011 .016 .013 DW 1.92 (.30)3 2.09 1.91 1.79 2.04 .008 2.24 1Absolute values of t-statistics in parentheses. R2 is the coefficient of determination adjusted for degrees of freedom; SE is the standard error of the regression; and DW is the Durbin-Watson test statistic. Statistically significant at the 5 percent level using a one-tailed test. 3Estimate of rho, the first-order serial correlation coefficient. mate equation 1 using unconstrained ordinary least squares (OLS) instead o f subjecting the data to poten tially invalid polynomial restrictions.8 Equation 1 is estimated using quarterly data from Canada, France, Germany, Japan, the United King dom and the United States.9 A summary o f the OLS regression results is reported in table 1. (The detailed results can be found in the appendix.) The sample periods differ due to differences in data availability. The regressions exhibit a relatively wide range o f ex planatory power in describing GNP growth in the sFor a discussion o f the possible effects o f using polynomial and endpoint restrictions, see Peter Schmidt and Roger N. Waud, “ The Alm on Lag T echnique and the Monetary Versus Fiscal Policy Debate , ” Journal o f the A m erican Statistical Association (March 1973), pp. 11-19. The imposition o f polynom ial and endpoint constraints is moti vated primarily b y the desire to estimate m ore precisely coef ficients o f highly colinear variables (a com m on characteristic o f distributed lag models). Our concern, in contrast, is the total or cumulative impact o f monetary and fiscal actions on G N P growth. Consequently, OLS will yield estimates o f linear combinations o f coefficients that are as precise as those obtained by im posing polynomial and endpoint restrictions. See Henri Theil, Principles o f Econom etrics (John W iley and Sons, Inc., 1971), pp. 147-52. W hen estimated for France, equation 1 also contains a dummy variable representing the student riots and subsequent nationwide strikes that occurred in II/1968. different economies; the R2 varies from a high o f .82 in France to a low o f .19 in Japan. The Durbin-Watson statistics indicate that the estimates generally are not plagued by first-order serial correlation problems. In only one instance, that o f Canada, is a first-order serial correlation correction technique necessary. As shown in table 1, this correction (rho is estimated to be .30) adequately removes the problem. The United States The “ standard” results appear to hold for the United States; that is, they are not affected significantly by our modifications. The summed impact o f money growth is significantly positive (t = 4.29) and does not differ from unity (t = 0.37). This means that a 1 percentage-point increase in money growth leads to a permanent 1 per centage-point rise in GNP growth. M oreover, the esti mated coefficients for the individual lag terms (see appendix) suggest a large effect o f money on income during the first three quarters, with a varying impact throughout the remaining lag terms. The estimated coefficients for the fiscal measure are interesting because they indicate only a minor initial effect on income growth with a mostly negative impact thereafter. This is supported by the cumulative effect 7 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 o f fiscal policy being negative and negligible (2g = —0.199), and statistically insignificant (t = —1.21). actions is smaller than that o f a change in money growth in each country. The results obtained for exports are similar to those for fiscal actions: none o f the individual coefficients are large in absolute magnitude compared with those o f M or G, and most are statistically insignificant. M ore over, the cumulative effect o f export growth on GNP growth is not statistically significant at any conven tional level. The Impact o f Exports Thus, the standard St. Louis equation results con tinue to hold for the United States even with the changes in the specification: money growth exerts a significant, lasting impact on income growth; govern ment expenditure growth and export growth have only transitory influences at best. With these results form ing the basis for comparison, we will now examine what the application of this framework produced in the other countries. The Impact o f Money Looking first at the effects o f changes in money growth, we observe that the qualitative results for each country are quite similar to those for the United States. Specifically, changes in money growth have a statisti cally significant, permanent impact on nominal income growth in each country.10 The quantitative results, however, exhibit some differences. The cumulative impact o f money growth for each country except Cana da is noticeably smaller than it is for the United States. For Canada, the cumulative impact is not statistically different from one (t = 1.29). Thus, while changes in money growth exert a positive, statistically significant influence on the growth o f income across all the econo mies studied, a 1 percentage-point increase in money growth results in a less than 1 percentage-point rise in income growth for all o f the countries except the United States and Canada. The Impact o f Fiscal Actions The results of changes in fiscal actions are interesting because they tend to confirm the U.S. findings. The cumulative impact of a change in the growth o f govern ment expenditures on income growth is statistically significant for the United Kingdom and France. For the remaining countries, however, the cumulative im pact is negligible and, for Canada and Germany the variable takes on an u n expected negative sign. Moreover, the cumulative impact o f a change in fiscal 10Because the expected cumulative impact o f each variable in equa tion 1 is positive, one-tailed hypothesis tests are em ployed. 8 Not surprisingly, export growth is an important fac tor in explaining GNP growth for the countries in our sample other than the United States and Japan.11 The cumulative impact is statistically significant and ranges in magnitude from 0.54 in Canada to 0.21 in the United Kingdom. Consequently, it appears that the inclusion o f export growth is an important modification o f the St. Louis equation for explaining econom ic activity in open econom ies.12 IT WORKS, BUT IS IT STABLE? The comparison o f the empirical results from a vari ety of countries indicates that the St. Louis equation is useful in assessing the relative impact o f monetary and fiscal actions, and that its explanatory power can be increased with the addition o f export growth as an explanatory variable. Furthermore, the evidence here suggests that changes in money growth have a perma nent and significant influence on GNP growth. The evidence does not provide a similar conclusion for fiscal actions, except for the United Kingdom and France. The usefulness o f any equation that purports to ex plain macroeconomic phenomena depends crucially on the stability o f the estimated relationship. This issue is even more significant if some o f the right-hand-side variables in the estim ated equation are p olicy determined. 13 Consequently, it is always important to u The export results for Japan are not surprising, even though the general perception o f Japan is that o f a large exporter. Japan’ s export sector as a percent o f nominal G N P is actually quite low relative to other countries in our sample. For example, in 1980 Japan’s exports accounted for only 12 percent o f GNP. In com pari son, the figures for the other countries are; United States (8 percent); Canada (27 percent); United Kingdom (21 percent); France (18 percent); and Germ any (24 percent). 12W hen equation 1 is estimated excluding the distributed lag o f export growth, the qualitative results for France are the only ones affected. In that case, the cumulative impact o f a change in m oney growth is no longer statistically significant (even at the 10 percent level). Furtherm ore, there is little change in the quantitative results concerning the cumulative impacts o f either monetary or fiscal actions. This finding is com forting given the discussion in footnote 5. I3The argument is that if estimated parameters change with policy changes, then there is no stable foundation upon which policy makers may project the outcom e o f today’s actions into the future. This argument is presented in Robert E. Lucas, Jr., “ Econom etric Policy Evaluation: A C ritique,” in Karl Brunner and Allan A. Meltzer, eds., The Phillips C u rve and L a b or M arkets, vol. 1 (1976), The Carnegie-Rochester C onference Series on Public Policy, Supplement to the Journal o f M onetary Econom ics, pp. 19-46. FEDERAL RESERVE BANK OF ST. LOUIS examine the statistical stability o f the estimated param eters across alternative policy rules if the equation is being used in policy analysis. Although the determination o f each policy shift in each country is a task well beyond the scope o f this paper, there is a single event common to all o f the countries that can be used to assess the stability o f the estimated relationships. That event, which occurred during the early 1970s, is the collapse o f the Bretton W oods system. In general, the period before the second quarter o f 1973 is viewed as a fixed exchange rate regime while the period since then usually is characterized as a floating exchange rate period.14 While one may quibble about this characterization, the early 1970s would seem to mark a significant turning point in the implementation o f domestic monetary and fiscal policies for the open economies in our sample. Consequently, this apparent policy shift provides a useful point to test the stability properties o f the esti mated income relationships.15 It is essential to understand that we are investigating the stability o f the relationship that explains the trans mission o f changes in money growth and government expenditure growth, however determined, to changes in GNP growth. W e are not concerned with how or why a change in money growth or government expen diture growth occurs; we simply wish to determine the extent to which these variables affect the growth of nominal GNP. Consequently, the use o f the exchange rate regime change does not require monetary or fiscal actions to have any greater or lesser effect on GNP growth after the break than before. The change in 14The break points for the United Kingdom and Canada tested are slightly different from the L/1973 point. See text. 15It is typically thought that during a fixed exchange rate regim e the reserve currency country determ ines monetary policy for the rest o f the world. I f this w ere the case, the measured influence o f monetary actions on econ om ic activity during the Bretton W oods period actually w ould indicate actions motivated by the reserve currency country, not by the dom estic monetary authorities. To test this proposition, w e perform ed G ranger-type causality tests to see if changes in U.S. m oney growth “ caused” changes in foreign m oney growth during the fixed exchange rate period. These tests results did not indicate any systematic relationship betw een U.S. m oney growth and m oney growth in any o f the countries included in our sample. O ur results support those o f Edgar L. Feige and James M. Johannes, “ Was the United States Responsible for W orldw ide Inflation U nder the Regim e o f Fixed Exchange Rates?” Kyklos (Fasc. 21982), pp. 263-77; and Edgar L. Feige and Kenneth J. Singleton, “ Multinational Inflation Under Fixed Exchange Rates: Som e Empirical E vidence From Latent Variable M o d e ls ,” The R eview o f E con om ics and Statistics (February 1981), pp. 11-19. Consequently, connecting observed m oney growth with monetary policy decisions in these countries, even during the fixed exchange rate period, appears to have some empirical support. JANUARY 1983 Table 2 Stability Test Results Absolute values of t-statistics Country M G Canada 0.26 0.40 France 1.44 1.08 Germany 1.65 0.68 Japan 0.70 1.55 United Kingdom 0.07 2.241 United States 0.61 1.35 'Statistically significant at 5 percent level. exchange rate regimes is chosen as a likely break in the income equations primarily because o f its universality. To examine the stability o f the estimated income relationships, (0,1) dummy variables are used to form multiplicative slope-dum m y terms for the money growth and government expenditure growth variables. Stability is investigated by testing the hypothesis that the cumulative impact o f each dummied variable’s dis tributed lag is significantly different from zero.16 If the resulting t-statistic is less than a predetermined critical value, the null hypothesis that these coefficients are stable across exchange rate regimes cannot be re jected. The calculated t-statistics for each variable’s stability test are reported in table 2. The break point for the United States, France, Germany and Japan is 1/1973, the widely accepted timing o f the breakdown o f the 16This approach is suggested by Dam odar Gujarati, “ Use o f D um m y Variables in Testing for Equality Between Sets o f Coefficients in Linear Regressions: A Generalization,” The A m erican Statistician (D ecem ber 1970), pp. 18-22. W e em ploy this m ethod by con structing a slope-dummy; term for each variable in the distributed la g o fM a n d o fG (e .g ., D M , = D -M ,w h e re D = 0 in the fixed-rate period and 1 in the floating-rate period). The hypothesis that the cumulative impact o f M has changed with the m ovem ent o f float- K ing exchange rates is then investigated by testing 2 . D M t_, = 0. i= 0 A similar procedure is used for G. This approach is chosen over the m ore com m only used C how test, because the C how test examines the stability o f the entire relationship. Thus, the coefficients o f one variable may change dramatically over tim e, while the C how test will not reject the hypothesis o f stability if that variable’s explanatory pow er is weak relative to that o f other variables w hose coefficients are relatively stable. The dum m y variable approach circum vents this potential problem . 9 JANUARY 1983 FEDERAL RESERVE BANK OF ST. LOUIS Smithsonian extension o f the Bretton W oods system. Because the United Kingdom and Canada had refused earlier to peg the value o f their currencies to the U.S. dollar, the break points tested are 11/1972 and 11/1970 for the United Kingdom and Canada, respectively. The results reported in table 2 support the hypothesis that in each country the cumulative impact o f a change in money growth is stable across the break in exchange rate regimes. The cumulative impact o f a change in the growth of government expenditures exhibits instabil ity only for the United Kingdom. The results for the United Kingdom indicate that the estimated equation does not reliably capture the rela tionship between changes in the growth o f government expenditures and GNP growth. Furthermore, a shift in the trend rate o f velocity growth (captured by the constant term) is detected. To correct for both o f these deficiencies, equation 1 is re-estimated for the fullsample period with the coefficients o f government ex penditure growth and the constant term allowed to take on different values during the two exchange rate periods. The re-estimated United Kingdom equation is (absolute value o f t-statistics in parentheses): 11 0.024 D1 + 0.679 X M, j (1.90) (2.12) i = 0 Y = 0.008 (1.05) 2 2 - 0.043 2 C L , + 0.530 2 GfL j (0.19) i = 0 (3.39) i = 0 2 + 0.200 2 EXt_ (2.95) i = 0 R2 = 0.66 SE = 0.012 DW = 2.15 These results indicate that, after separating the in- fluence of government expenditures and the constant term into the two periods, the cumulative effect of changes in British money growth increases in magni tude and remains positive and significant and now is not statistically different from one (t = 1.00). Digitized for 10 FRASER This suggests that the failure to incorporate the secular decline in the trend rate o f velocity growth since 1973 seriously understated the initially estimated impact o f changes in money growth. Export growth continues to influence GNP growth significantly, although the sum med coefficient indicates a slight decline. The United Kingdom estimates indicate that the government expenditure results in table 1 are captur ing the post-II/1972 effects. For the period 11/1966 to 11/1972, fiscal actions have no significant lasting effect on income growth. The post-II/1972 results, on the other hand, point to a significant and fairly substantial fiscal effect. The post-II/1972 results indicate that in creasing the growth o f government expenditures by 1 percentage point will permanently increase income growth by about one-half as much. Thus, in contrast to the evidence presented for the other countries ex amined, the cumulative impact o f fiscal actions is high ly significant only in the United Kingdom, and then only after 11/1972. SUMMARY The results in this paper demonstrate that the St. Louis equation can be applied to a variety o f other countries and that monetary actions dominate fiscal actions in determining the pace o f economic activity in these countries. Estimating a modified St. Louis equa tion for six different countries, our results indicate that changes in money growth have a significant and lasting impact on nominal income growth in all six cases. O f equal importance, the money-GNP link was stable in each country across one o f the most significant interna tional policy shifts o f the last two decades — the move from fixed to floating exchange rates. In contrast, fiscal actions are significant only in the United Kingdom and France. Moreover, this effect does not appear to be stably related to income in the United Kingdom where fiscal actions have exerted a lasting impact on income growth only during the re cent floating exchange rate period. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 Appendix Detailed Estimation Results1 Constant -.0 0 6 (0.90) cn "tf; CVJ CO DW = 2.09 .246 (3.21 )3 .202 .045 - .0 1 7 -.0 1 8 .064 (4.71 )2 (1.16) (0.45) (0.49) (1.45) R2 = .29 SE = .011 DW = 1.91 .276 (2.48)3 .013 .161 .289 -.1 2 0 .209 .552 (0.13) (1.44) (2.54)2 (1.03) (1.83) (3.76)3 .006 (0.87) .006 (0.87) .067 (1.65) - .007 .335 .069 -.2 7 4 .190 -.1 1 3 -.0 4 1 .108 .160 -.1 6 9 -.0 1 6 .177 .419 (0.07) (3.31 )2 (0.64) (2.76)2 (1.81) (1.07) (0.38) (1.04) (1.45) (1.60) (0.15) (1.75) (2.50)3 .274 (3.11 )2 —.094 (1.24) .165 (2.19)2 .156 (4.86)2 .117 (3.42)2 - .0 6 4 (1.81) .067 (1.65) R2 = .19 SE = .016 DW = 1.79 CD II LO .001 (0.21) SE = .008 cvj United Kingdom Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Lag 6 Lag 7 Lag 8 Lag 9 Lag 10 Lag 11 Sums .518 (3.50)3 (0.53) (0.66) (1.46) (0.56) (1.50) (2.09)2 (1.44) (1.64) (1.16) (1.83) (0.03) (0.83) (0.89) (1.53) DW = 1.92 (,30)4 ICC .010 (1.65) .022 -.0 2 8 -.0 6 4 - .0 2 4 -.0 6 2 -.0 6 9 -.2 2 5 .192 (1.90)3 .066 .035 .052 .001 .024 .025 .043 SE = .006 II Japan Lag 1 Lag 2 Lag 3 Lag 4 Sums .087 (0.67) .024 (0.16) .407 (3.06)2 (1.47) (0.08) (0.10) (1.55) (1.78) (2.48)2 (2.00)2 (1.43) (2.73)2 (4.36)2 (0.94) (0.00) (1.03) (2.68)2 (0.92) (1.31) (1.88) (0.12) (0.85) (2.64)2 (1.00) (3.04)3 CVJ .007 (1.31) -.0 3 9 -.0 0 2 .003 .042 .047 .065 .049 .027 .063 .124 .025 .000 .029 .076 .025 .036 .051 -.0 0 3 .021 .073 .023 .543 ICC Germany Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Sums -.0 1 1 (0.09) (0.57) (2.01 )2 (0.98) (0.43) (0.37) (1.17) (0.19) (0.77) (1.08) (1.75)3 -.0 3 6 .132 .075 -.0 3 4 .030 .088 .015 -.0 5 8 .077 .289 (0.31) (1.48) (0.25) (0.39) (0.16) (1.92) Summary statistics II .007 (1.43) .006 -.0 3 5 -.0 0 7 -.0 1 1 -.0 0 5 .041 EX CM (2.51 )2 (3.15)2 (2.03)2 (0.51) (3.05)2 (2.25)2 (2.00) (1.00) (1.73) .726 (3.41 )3 France Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Lag 6 Lag 7 Lag 8 Sums .142 .205 .131 -.0 3 3 .215 .154 -.1 3 4 -.0 6 9 .115 G ICC Canada Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Lag 6 Lag 7 Lag 8 Lag 9 Lag 10 Lag 11 Lag 12 Sums M SE = .013 DW = 2.04 .345 (2.90)3 .209 (3.02)3 (continued on next page) 11 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 Appendix Continued1 Constant United States Lag 1 Lag 2 Lag 3 Lag 4 Lag 5 Lag 6 Lag 7 Lag 8 Lag 9 Lag 10 Lag 11 Sums .007 (1.94) M .709 .425 .380 -.2 5 5 .204 -.3 6 9 (4.14)2 (2.70)2 (2.64)* (1.74) (1.29) (2.08)2 1.094 (4.29)3 1See notes accompanying table 1. Statistically significant at the 5 percent level using a two-tailed test. Statistically significant at the 5 percent level using a one-tailed test. 4Estimate of rho, the first-order serial correlation coefficient. Digitized for 12 FRASER G .065 .055 -.1 2 1 .030 - .0 6 3 - .0 4 9 .079 -.0 2 8 -.0 7 6 -.0 9 1 (1.22) (1.04) (2.28)2 (0.56) (1.17) (0.93) (1.49) (0.49) (1.28) (1.61) EX .022 .015 .005 .000 -.0 0 9 .006 - .009 .015 -.0 1 6 .015 .030 .040 (2.41 )2 -.1 9 9 (1.21) .114 (1.28) (0.86) (0.31) (0.01) (0.54) (0.36) (0.51) (0.87) (0.91) (0.84) (1.73) (1.64) Summary statistics R2 = .41 SE = .008 DW = 2.24 Four Econometric Models and Monetary Policy: The Longer-Run View KEITH M. CARLSON and SCOTT E. HEIN O NE key element in the making o f an informed economic policy decision is the accuracy with which policymakers can gauge the longer-run consequences o f their policy actions and strategies. Crucial to such attempts to grasp these policy consequences is the use o f econometric models. Whether current econometric models are useful in this respect depends upon their “long-run” characteristics; unfortunately, until recent ly, there had been virtually no study o f the compara tive long-run properties o f the major econom etric models currently in use. Most analyses instead have dealt with how well these models forecast a few quar ters ahead. This situation changed with the publication o f a re cent study by the Joint Economic Committee (JEC) o f Congress that focused explicitly on the econom ic im pact o f alternative long-run monetary strategies using three well-known econometric m odels.1 Missing from the JEC study, however, was an econometric assess ment using an explicit monetarist model. The purpose of this paper is to extend the JEC study by comparing their results with those obtained for the St. Louis model. Analysis o f the St. Louis model according to criteria used in the JEC study is informative for two reasons. First, it indicates whether a monetarist framework provides additional insight into the longrun effects o f monetary policy. Second, it provides policymakers the opportunity to compare the long-run properties o f a monetarist model with those o f the major nonmonetarist models. FEATURES OF THE JEC STUDY The JEC study examined the simulated perfor mance o f certain key macroeconomic variables under four different long-run monetary strategies. Three large-scale econometric models were analyzed: those o f Chase Econometrics, Data Resources Incorporated 1Robert E. Weintraub, T hree Large Scale M odel Simulations o f F ou r M oney G row th Scenarios, a staff study prepared for the use o f the Subcomm ittee on Monetary and Fiscal Policy o f the Joint E conom ic Com m ittee o f Congress (G overnm ent Printing Office, 1982). (DRI) and Wharton, the best-known and most widely used models. Four separate monetary strategies were considered over a 10-year simulation period (1982 through 1991), using the fourth quarter o f 1981 and an M l growth rate o f 5.8 percent as points o f departure; (1) a sudden deceleration of M1 growth to zero percent in one year, and then held at zero; (2) gradual deceleration of M 1 growth to zero percent over a five-year period, and then held at zero; (3) sudden deceleration o f M l growth to 3 percent in one year, and then held at 3; (4) gradual acceleration of M l growth to 10 percent over a five-year period, and then held at 10. In addition, each m odel’s proprietor was asked to run a baseline projection with freedom to choose the mone tary strategy.2 2The baseline simulations thus represented each m odel’s assump tion about the future course o f monetary policy as o f March 1982. These assumptions w ere as follows: M l Growth Rate: 1982-91 Chase DRI Wharton 6.3% 4.5 5.2 The m odel proprietors w ere further instructed to simulate each o f the four monetary strategies twice: first, without making any judgmental adjustments, and second, making any adjustments deem ed necessary to ensure consistency and generate results that were considered sensible. These adjustments w ere at the discre tion o f the individual m odel proprietor and involved no contact with the JEC staff. The JEC labeled these two sets o f simulations “ pure” and “ managed. ” The JEC study concluded, on the basis o f the pure simulations, that none o f the models can b e used b y themselves to decide among the monetary strategies. The results o f these pure simulations w ere term ed “ puzzling,” because the links betw een the m oney growth and the key m acroeconom ic variables ran counter to historical experience. The JEC conclusions about the managed simulations w ere m ore positive. W hile there still remained som e inconsistencies with historical relationships, the managed simulations w ere ju dged to provide a better basis for considering the longer-run policy im plications o f alternative monetary aggregate growth strategies. Thus, in the discussion to follow, only the managed simulation results from the large scale models are considered. 13 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 The St. Louis Model The basic structure o f the St. Louis model, de veloped in the late 1960s, has remained essentially unchanged since then.1 The model consists of five equations and two identities (see appendix). The foundation o f the model and the basis for its mone tarist label is the GNP equation. The growth rate o f GNP is specified as a function o f current and lagged values o f M l growth and current and lagged values o f the growth o f high-employment federal expendi tures. The monetarist label stems primarily from the estimated coefficients: the sum o f the coef ficients on money growth is about unity and the sum o f the coefficients on high-employment expenditure growth is about zero. The price equation relates the rate-of-change o f prices to current and lagged values o f demand pres sure, current and lagged values o f changes in the relative price o f energy, and a measure o f antici pated price change. Demand pressure is defined as the growth o f output relative to the growth o f highemployment output. Anticipated price change is a weighted sum o f past price changes with the weights obtained by estimating the corporate Aaa rate as a function o f past inflation. Output growth (real GNP) is determined residually via the GNP identity; nominal GNP growth is the sum o f real GNP growth and the rate o f change o f prices. 'Leonall C. Andersen and Keith M. Carlson, “ A Monetarist M odel for E conom ic Stabilization,” this Review (April 1970), pp. 7-25. M inor changes which have been made include: (1) a respecification in rate-of-change form from the original first difference form; (2) the addition o f energy prices as an exoge nous variable; and (3) a change in estimation procedure from ordinary least squares to generalized least squares for those equations in which serial correlation is evident. See Keith M. Carlson and Scott E. H ein, “ An Analysis o fa M odified St. Louis M od el,” a paper prepared for the Spring C onference on C om paring the Predictive Performance o f M acroeconom ic M odels at Washington University in St. Louis (April 20, 1982). The model’s remaining equations provide esti mates o f three other macroeconomic variables. Un employment is estimated as a function o f the gap between actual output and high-employment out put. The Aaa bond rate is a function o f past inflation. The 4-month commercial paper rate is a function of contemporaneous M l growth and current and lagged values o f changes in the relative price o f energy, output and prices. Though the instructions were specified in terms o f M l, none o f the models permitted direct control o f this monetary aggregate. Both Chase and DRI specify the control o f money growth through nonborrowed re serves. Thus, nonborrowed reserves were manipu lated to achieve the desired M l growth.3 For the Wharton model, the target variable is M2 instead of M l. The Wharton simulations were conducted using M2 target rates o f 4 percent, 7 percent and 14 percent, respectively, whereas the JEC specified M l targets of zero percent, 3 percent and 10 percent.4 3D RI has an iterative procedure that allowed them to hit M 1 targets exactly as specified by the JEC. Chase, on the other hand, used a trial and error procedure, and was unable to achieve M l targets precisely. Simulation o f the St. Louis model for the long-run monetary strategies outlined in the JEC study re quired assumptions about other exogenous variables: potential output was assumed to grow 2.5 percent per year, high-employment expenditures to increase at a steady 8 percent rate, and the change in the relative price o f energy to be zero. To determine a baseline strategy, an average o f the baseline strategies for the large-scale m odels was con stru cted . W hat this amounted to was a gradual reduction in M l growth from a 5.8 percent rate in fourth quarter 1981 to 5.0 percent in 1991. Since the simulations w ere run in March 1982, Chase E con ometrics has revised their m odel to incorporate a new monetary sector to reflect changes in Federal Reserve policy procedures in O ctober 1979. At the time the simulations w ere run, the Chase m odel used an index o f credit rationing as the primary channel o f monetary influence. 4M 1 is an endogenous variable in the m odel, however, so there is a basis for com paring the W harton m odel with the other m odels. The resulting M l growth rates w ere generally, but not precisely, con sistent with the JEC s instructions. 14 FEDERAL RESERVE BANK OF ST. LOUIS PROPERTIES OF THE MODELS AS REVEALED BY THE SIMULATION RESULTS This study follows the general format o f the JEC study, using the U.S. econom ic experience from 1956 through 1981 as a guide in comparing the models. If certain systematic relationships among key variables have held over the past 26 years, the simulation results for the next 10 years should be roughly consistent with that experience if one is to place much faith in the model. Deviations from historical experience place the burden o f explanation on the individual model pro prietor. Simulation results relating money growth to (1) nominal GNP growth, (2) inflation and (3) real output growth are considered first. Then, the relationships between real output growth and unemployment, and between nominal interest rates and inflation are evalu ated. Since the longer-run relationships are o f primary interest and since short-run adjustments make the re sults difficult to interpret, the results for the last five years o f the simulations, 1987-91, are investigated. GNP, Money and Velocity With simulations o f the four long-run monetary strategies and a baseline simulation, five observations characterizing the 1987—91 period were generated for each model, providing a basis for examining the rela tionship between money growth and nominal GNP implicit in each. This relationship is referred to con ventionally as the velocity o f money. The well-known equation o f exchange portrays this as M V = Y, o r V = where M is money stock, Y is nominal GNP, and V is the velocity o f money. In its growth rate form, M + V = Y. Although velocity growth is influenced by many variables, it has shown considerable stability during the 1956-81 period. The implication o f this stability is that, in the long run, nominal GNP growth is related closely to the growth o f M l. The stability o f velocity growth further suggests that a 1 percent change in rate of growth o f money should coincide generally with a 1 percent change in the rate o f growth o f nominal GNP. The large-scale econometric models do not specify GNP as a direct function o f money. In these models, JANUARY 1983 money affects GNP indirectly via interest rates and wealth or real balance effects. Despite this, the large models still yield systematic relationships between money and GNP. Chart 1 summarizes the money-GNP simulation re sults. Each model is summarized by plotting the aver age growth o f simulated nominal GNP for the 1987-91 period against the average growth rate o f M 1 for the same period. Each point represents model results for a particular long-run monetary strategy.5 As noted above, these strategies are stated in terms o f M l growth, and include (1) a sudden deceleration to zero percent, (2) a gradual deceleration to zero percent, (3) a sudden deceleration to 3 percent, (4) a gradual accel eration to 10 percent, and (5) a baseline strategy chosen by the model proprietor. The historical line is derived by regressing the fiveyear average growth rate o f nominal GNP on the fiveyear average growth rate o f M l. The parallel lines depict the regression estimate plus or minus one stan dard error o f the equation. If velocity growth is totally independent o f money growth, then the slope o f the historical line would be 45 degrees. The estimated slope, in fact, is not significantly different from 45 degrees. Comparing the different models with historical ex perience suggests that none o f the large-scale models is generally consistent with the actual past. Only four of the 15 simulated cases for these models fall within the historical band. The DRI and Chase simulations indi cate that velocity growth is related negatively to money growth, so that higher rates o f money growth do not yield proportionally higher nominal GNP growth. On the other hand, simulation results for the Wharton model indicate that higher money growth results in more than a proportional increase in GNP growth. This result, however, follows from the nature o f the finan cial sector in the Wharton model. On the basis o f M2, which is Wharton’s actual monetary target variable, velocity growth is related negatively to money growth as in the Chase and DRI models. Not surprisingly, the St. Louis model falls clearly within the historical band; after all, the GNP equation 5For the M l growth rate associated with each strategy, refer to the accompanying table. The points on the chart are connected for each m odel in ascending order o f M l growth. Consequently, the results for the Chase and W harton m odels are not charted with the JEC’ s slowest growth strategy farthest to the left. See also footnote 3. 15 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 C h a rt 1 Table 1 M oney and GNP Y Y (Percent) GNP, Money and Velocity (1987-91) (Percent) Average Annual Results Model and Strategy M (Percent) is constructed to be consistent with this historical experience.6 The proprietors o f the other models offer no explanation as to why their models predict that velocity behavior in the future will be different from the past.1 Inflation and Money Economists generally agree that, over the long run, fT h e St. Louis m odel simulations do show a weak positive rela tionship betw een velocity growth and m oney growth. This result occurs because the estimated sum o f the coefficients on M in the G NP equation is slightly greater than unity. 7The JEC study suggests that the reason the large-scale m odels run contrary to historical velocity experience is that they are built to short-run specifications, that is, their focus is on forecasting for short periods into the future. Such an explanation might be appropriate for the Chase m odel, but the D R I and W harton models are annual models. The results suggest that something more fun damental is awry. In addition, the St. Louis m odel, which is a quarterly m odel, does not exhibit any departure from historical long-run velocity behavior. Digitized for 16 FRASER M Y V Chase 1 2 3 4 Baseline 3.0% 1.1 2.8 10.6 6.4 8.1% 8.0 8.3 11.5 9.4 5.1% 6.9 5.5 1.0 3.0 DRI 1 2 3 4 Baseline 0.0 0.0 3.0 10.0 4.1 7.2 7.3 9.1 13.5 9.6 7.2 7.3 6.0 3.2 5.3 Wharton 1 2 3 4 Baseline 3.0 1.5 3.2 6.5 4.9 5.8 6.4 7.5 12.5 9.8 2.8 4.8 4.2 5.7 4.6 St. Louis 1 2 3 4 Baseline 0.0 0.1 3.0 9.9 5.2 2.8 2.9 6.2 14.0 8.7 2.8 2.8 3.1 3.7 3.3 inflation is related directly to money growth.8 In terms o f the equation o f exchange, with rates o f change o f prices and output ( P + X ) substituted for Y , M + V = P + X. A justification o f the money-inflation relationship is that V and X are not related systematically to M over the long run. Consequently, variations in M eventually are reflected in P . To evaluate the money-inflation relationship for the different models, the simulation results are summa- 8For example, Barro and Fischer introduced their 1976 survey o f monetary theory with the follow ing statement: “ Perhaps the most striking contrast between current views o f money and those o f 30 years ago is the rediscovery o f the endogeneity o f the price level and inflation and their relation to the behavior o f m oney.” Robert ]. Barro and Stanley Fischer, “ Recent D evelopm ents in Monetary Theory, ” Journal o f M onetary Econom ics (April 1976), p . 133. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 C h o rt 2 M o n e y a n d In fla tio n Table 2 P Inflation and Money (1987-91) ,- (P e r c e n t) 14 Average Annual Results 13 Model and Strategy 12 11 - 10 9- M is average a nnual rate fo r 1987-91; P is fo r 1991. M Final Year P P 4.8% 4.9 5.2 7.8 6.3 4.9% 4.4 4.8 7.9 5.9 Chase 1 2 3 4 Baseline 3.0% 1.1 2.8 10.6 6.4 DRI 1 2 3 4 Baseline 0.0 0.0 3.0 10.0 4.1 4.0 4.0 6.1 10.4 6.5 3.6 3.6 5.8 9.8 6.2 Wharton 1 2 3 4 Baseline 3.0 1.5 3.2 6.5 4.9 2.9 3.4 4.2 9.8 6.6 2.1 2.3 3.8 10.3 6.2 St. Louis 1 2 3 4 Baseline 0.0 0.1 3.0 9.9 5.2 -2 .7 -0 .9 1.6 10.0 5.2 -1 .7 - 1 .9 2.8 12.8 6.0 H isto rica l re la tio n sh ip : P = - 1.46 + 1.36 M (2.05) (9.70) R2 = 0.82 SE = 1.22 M (Percent) rized in chart 2. Without exception, all four models show a direct relationship between monetary growth and inflation. There is substantial variation, however, in the degree o f sensitivity among the models. The Chase model shows a difference in inflation forecasts of only 3.5 percent betw een the slowest and fastest monetary growth strategies. DRI shows a 6.2 percent age point differential and Wharton a differential o f 8.2 percentage points. The St. Louis model shows the largest differential o f 14.5. To provide a basis for historical comparison, the inflation rate was regressed on the average o f money growth over the previous five years for the 1956-81 period. Comparing the simulation results o f the four models with this historical line suggests that there is some bias in each. The Chase and DRI models exhibit a sensitivity o f inflation to money growth that appears too low, while the Wharton and St. Louis models show a sensitivity that appears too high. W hile the models generally are inside the historical band for money growth rates in the neighborhood o f the 1956-81 aver age o f 4.7 percent, a wide range o f results occurs for monetary strategies that lie at the extremes o f his torical experience.9 9An explanation o f these diverse results w ould require a detailed analysis o f the inner workings o f each m odel. For the most part, the large-scale models estimate the price level primarily by marking up som e measure o f labor costs. Consequently, the insensitivity o f inflation to m oney growth developm ents in the Chase and D RI m odels might b e related to the stickiness o f wages. This explana tion does not seem to explain the W harton results, however. The Wharton m odel shows considerable sensitivity in the 3 percent to 7 percent range for m oney growth, yet the price determination process apparently is similar to that for Chase and D RI. The St. Louis m odel differs from the large-scale m odels in that prices are determ ined directly b y dem and pressure and past prices. The influence o f past prices tends to capture effects operating through wages, yet inflation remains sensitive to m oney growth throughout the full range. 17 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 C h a rt 3 Table 3 M on ey and Real GNP Real GNP Growth and Money (1987-91) X -- (Percent) (Percent) 71--------- Average Annual Results 7 Model and Strategy H istorical relationsh ip ( 1956- 81 ) M and X are average annual rates fo r 1987-91. Historical relationship: X = 3.87 - 0.07M R2 = 0.02 (7.42) (0.72) SE = 0.89 _L_ _L_ _L_ 5 6 _L_ _L_ _l_ 10 11 M (P e rce n t) Real GNP and Money A corollary to the long-run, money-inflation rela tionship is the hypothesis that the trend growth o f real GNP is not systematically related to long-term money growth. Money may affect the growth o f real GNP in the short run, but if inflation rises one-for-one with accelerated money growth, as the equation o f ex change indicates, there are no cumulative effects on real GNP. Chart 3 summarizes the money-real GNP rela tionship from the simulations o f the four models. The three large-scale models all show real GNP growth rates in the neighborhood o f 3 percent, regardless o f which monetary strategy is considered. The St. Louis model, on the other hand, shows greater variation of real GNP growth among the strategies. This is because the dynamic lag structure o f the St. Louis model is such that, after 10 years, the model is still a considerable time away from steady-state equilibrium in growth Digitized for 18 FRASER M X Chase 1 2 3 4 Baseline 3.0% 1.1 2.8 10.6 6.4 3.1% 3.0 2.9 3.4 3.0 DRI 1 2 3 4 Baseline 0.0 0.0 3.0 10.0 4.1 3.0 3.2 2.9 2.8 2.9 Wharton 1 2 3 4 Baseline 3.0 1.5 3.2 6.5 4.9 2.8 2.9 3.2 2.5 3.0 St. Louis 1 2 3 4 Baseline 0.0 0.1 3.0 9.9 5.2 5.6 3.9 4.5 3.7 3.3 terms. Given more time to adjust, the St. Louis model tends to approach about 3 percent real growth, regard less o f money growth. The historical line in chart 3 is based on five-year growth rates o f both money and real GNP. The slope of the line is not significantly different from zero, and the standard error is quite large relative to the mean. The results for the three large-scale models are virtually identical. Relative to the large-scale models, the St. Louis model is the outlier, though four o f the five simulated observations are well within the historical band; only the strategy o f sudden deceleration o f M l to zero yields real output growth that is outside the his torical band. Again, this makes sense because o f the long adjustment process in the St. Louis model; very weak output growth in the early years under the zero money growth strategy is offset by very strong output growth in the 1987—91 period. In general, the simulation results suggest that money has a neutral effect on real output growth in the FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 Table 4 C h a rt 4 R eal G N P a n d U n e m p lo y m e n t R ate AU Real GNP Growth and Unemployment AU (1987-91) Model and Strategy (P o r te n t) long run. A sustained change in the money growth rate has little or no effect on the long-run growth rate o f real GNP. Real GNP and Unemployment Another relationship o f interest in macroeconomics is the one between real GNP growth and the unem ployment rate. All three o f the large-scale models show essentially the same rates o f real growth for each o f the monetary strategies. Thus, Okun’s law, which relates unemployment to deviations o f actual from potential output, suggests that the change in the unemployment rate would be approximately equal for all strategies.10 Such is not the case. Each o f the large-scale models shows considerable variation in the change in the un 10Arthur M. Okun, “ Potential GNP: Its M easurem ent and Signifi can ce,” 1962 Proceedings o f the Business and E conom ic Statistics Section o f the Am erican Statistical Association, pp. 98-104. Average Annual Rate Change in U: X 1986-91 Chase 1 2 3 4 Baseline 3.1% 3.0 2.9 3.4 3.0 0.0% 0.5 0.1 - 2 .3 - 1 .3 DRI 1 2 3 4 Baseline 3.0 3.2 2.9 2.8 2.9 -1 .5 -1 .9 -0 .9 0.0 - 0 .6 Wharton 1 2 3 4 Baseline 2.8 2.9 3.2 2.5 3.0 -1 .9 -1 .4 - 3 .2 1.4 - 1 .0 St. Louis 1 2 3 4 Baseline 5.6 3.9 4.5 3.7 3.3 - 6 .6 - 2 .9 - 4 .5 - 3 .0 - 2 .0 employment rate despite near-equal rates o f real GNP growth. What is not known, o f course, are the assumed growth rates for the labor force and other determinants o f potential output in these models. Nevertheless, when the strategies are compared across models, the results o f a sudden deceleration o f M l growth to zero range from no change in the unemployment rate for the Chase model to a 1.9 percentage point decline for the Wharton model. The results for the opposite ex treme, gradual acceleration o f M l to 10 percent, show even greater variation— from a 2.3 percentage drop in the unemployment rate for the Chase model to a 1.4 point increase for the Wharton model. The St. Louis model also shows considerable varia tion in the change in unemployment across monetary strategies; however, this is due to substantial variation in the growth rate o f real GNP. All the unemployment changes are negative, because the simulated real growth rates exceed the assumed growth rate o f 2.5 percent for potential GNP. Moreover, because the St. 19 JANUARY 1983 FEDERAL RESERVE BANK OF ST. LOUIS mm mmmmmmmm^mmm C h a rt 5 Table 5 In fla tio n a n d Long-Term Interest Rate (P e rc e n t) Percent) Inflation and Interest Rates (1987-91) Model and Strategy St. Louis Historical relationship ( 1956 - 8 1 ) is average annual rate fo r 1987-91; RL is MA corporate bond rate fo r 1991. H istorical rela tion sh ip: RL = 2.68 + 1.03 P (5.28) (10.04) 1 R2 = 0.83 SE = 1.09 i____I____I____I____I____ I___ (P e r c e n t) Louis model simulates very strong 1987-91 real output growth in conjunction with the sudden deceleration of money growth to zero, sizable reductions in unem ployment go hand in hand with such a policy. The historical line in chart 4 is estimated by regress ing the change in the unemployment rate over fiveyear periods on the five-year growth rate o f real GNP. The historical band encompasses only one observation from the 20 that are charted. The models’ failure to replicate history may not be as bad as appears in the chart, however. Potential output supposedly grew faster in the 1956-81 period than it is assumed to be growing in 1987-91. The simulation results suggest an implied growth rate o f potential output o f 2.5 percent to 3.0 percent for 1987—91, instead o f the 3.6 percent rate calculated for 1956-81. Nevertheless, the largescale models show the inverse relationship between real growth and unemployment suggested by Okun’s law. In contrast to the St. Louis model, however, the degree o f sensitivity is not well defined. Digitized for 20 FRASER Average Annual Results M Chase 1 2 3 4 Baseline 3.0% 1.1 2.8 10.6 6.4 DRI 1 2 3 4 Baseline 0.0 0.0 3.0 10.0 4.1 Wharton 1 2 3 4 Baseline St. Louis 1 2 3 4 Baseline P RL Final Year RS RL 11.5% 11.5 9.5 11.6 10.0 16.3% 15.8 10.6 9.0 8.4 10.7% 11.6 8.7 12.7 9.6 4.0 4.0 6.1 10.4 6.5 10.1 10.1 11.1 14.8 11.4 8.0 7.8 9.5 12.4 10.0 9.5 9.5 10.7 14.1 10.9 3.0 1.5 3.2 6.5 4.9 2.9 3.4 4.2 9.8 6.6 8.8 8.1 10.7 16.0 12.3 6.5 6.2 8.6 13.8 9.4 6.9 7.4 9.2 16.5 11.7 0.0 0.1 3.0 9.9 5.2 - 2 .7 - 0 .9 1.6 10.0 5.2 5.6 7.6 8.4 14.0 11.3 1.9 2.8 4.9 11.3 7.3 4.6 5.8 8.5 16.1 11.5 4.8% 4.9 5.2 7.8 6.3 Inflation and Interest Rates The relationship between inflation and nominal in terest rates is the final relationship considered. The inflationary experience o f the last 15 years provides an ample basis for examining the nature o f this rela tionship. Monetary theory suggests that nominal interest rates reflect inflationary expectations. These expecta tions can be modeled as a function o f past inflationary experience. The question examined here is whether the econ om etric m odels incorporate such a rela tionship. Chart 5 summarizes graphically the simulation re sults for inflation and long-term interest rates. The Chase model does not appear to show any consistent relationship between inflation and long-term interest rates. The Wharton model displays a peculiar kink at relatively low rates o f inflation, while the DRI and St. Louis models display a strong positive relationship. JANUARY 1983 FEDERAL RESERVE BANK OF ST. LOUIS C h a rt 6 C h a rt 7 Inflation an d Short-Term Interest Rate M is e ry In d e x p+u P +U (P e r c e n t) (P e r c e n t) 10 11 10 11 M ( P e r c e n t) What is most obvious from the chart is the incon sistency with historical experience. The slopes o f the simulation results are roughly consistent, but the general level is vastly different. For the St. Louis model, the inconsistency arises because o f the use of the serial correlation adjustment in the simulations. With long-term rates in late 1981 well above the infla tion rate, this differential only gradually disappears during the simulation period. It appears that the largescale models are following a similar procedure. In this regard, it seems that most o f the models would do much better at predicting the change in long-term rates, rather than the level itself. Chart 6 plots the simulation results for inflation and short-term interest rates. Again, with the exception of the Chase model, the models demonstrate substantial similarities. The St. Louis model tends to simulate the lowest level o f short-term rates for a given rate of inflation. The historical line, as in the case o f long-term rates, is below all the model results, but the discrepan cy is not as great as that for long-term rates. All the models, with the exception o f the Chase model, in corporate an inflation premium into short-term rates, suggesting that the lower the inflation rate, the lower short-term interest rates will be. THE POLICY IMPLICATIONS OF THESE SIMULATION RESULTS The discussion above emphasized the long-run properties o f econometric models as revealed by the simulation results. What remains to be determined are the implications o f these results for long-run monetary policy. From this longer-run perspective, do the mod els’ simulation results favor a strategy o f slow M l growth, fast M l growth or something in between? To aid in this assessment, a crude index, called a “ misery index,” is constructed to summarize the re sults. The index is simply the sum o f the inflation rate 21 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1983 Table 6 Misery Index (1987-91) Average Annual Result Model and Strategy M Misery Index Final Year P U P + U Chase 1 2 3 4 Baseline 3.0% 1.1 2.8 10.6 6.4 4.9% 4.4 4.8 7.9 5.9 DRI 1 2 3 4 Baseline 0.0 0.0 3.0 10.0 4.1 3.6 3.6 5.8 9.8 6.2 6.5 6.4 6.3 6.6 6.5 10.1 10.0 12.1 16.4 12.7 Wharton 1 2 3 4 Baseline 3.0 1.5 3.2 6.5 4.9 2.1 2.3 3.8 10.3 6.2 7.6 7.6 5.5 9.1 6.1 9.7 9.9 9.3 19.4 12.3 St. Louis 1 2 3 4 Baseline 0.0 0.1 3.0 9.9 5.2 - 1 .7 -1 .9 2.8 12.8 6.0 3.9 6.2 3.9 1.7 4.9 2.2 4.3 6.7 14.5 10.9 and the unemployment rate at some point in tim e.11 Construction o f such an index is, o f course, simplistic, yet it provides general information for evaluating the effect o f the alternative monetary strategies. Chart 7 summarizes this misery index for the 198791 period for the four econometric models. In general, the simulation results indicate that there is a long-run payoff from following a slow M l growth strategy; the results from the Chase model provide the only excep tion. There seems to be little basis for choosing be tween sudden and gradual deceleration to zero money growth, however, because the misery index differs little when these strategies are compared. An evalua tion o f these strategies would involve a more detailed 10.5% 9.8 8.8 2.3 5.0 15.4% 14.2 13.6 10.2 10.9 analysis o f the adjustment path o f inflation and unem ployment. The general levels o f the misery index for the four models indicate substantial variation in the predicted effects o f alternative monetary strategies. For the slow M l growth scenarios, the St. Louis model is by far the most optimistic, and the Chase model is the most pessimistic. For the fast M l growth strategy, Chase is most optimistic and Wharton is most pessimistic. Thus, using this set o f results, a policymaker is con fronted with a disturbing diversity o f opinion. Yet, three o f the four models show a definite payoff from following a strategy o f slow to moderate growth o f M l. SUMMARY AND CONCLUSIONS n This simple index originated with the late Arthur Okun, although he called it a “ discom fort index. ” The term “ misery index” is used b y Jerome L. Stein, M onetarist, Keynesian and N ew Classical Econom ics (N ew York University Press, 1982), p. 159. 22 This article, extending recent work by Robert Weintraub at the Joint Economic Committee, has compared simulation results from various econometric models to FEDERAL RESERVE BANK OF ST. LOUIS the historical record o f the last 26 years. The emphasis is on the longer-run econom ic impact o f alternative money growth scenarios. No single model was found to be consistent with the historical record on all counts. The simulation results generally show, however, the positive consequences o f following a slow M 1 growth strategy. Higher rates o f money growth are associated with higher rates o f spending growth, which eventual ly are reflected in higher inflation rates. Using a simple social loss function called the misery index, three o f the four models indicate that, over the long run, unem JANUARY 1983 ployment gains, if any, are insufficient to offset the increase in inflation. Consequently, this article — like the JEC study before it — concludes that there are no long-run eco nomic gains from higher rates o f money growth. This is true even though the models run counter to historical experience in some important aspects. Moreover, the results indicate that higher inflation rates are associ ated with higher levels o f both short- and long-term interest rates, so that interest rates tend to be higher when the faster monetary strategies are followed. 23 Appendix Revised Form of St. Louis Model The version o f the St. Louis model used for the simulations in this article is summarized in table 1, with the coefficients given in table 2. Equations 1-4 are estimated with Almon constraints on the coefficients. Equation 5 is estimated with ordinary least squares. Three characteristics differentiate this model from the original version published in 1970: (1) most variables are entered in rate-of-change form rather than firstdifference form; (2) the demand slack variable is en tered in real rather than nominal terms; and (3) where relevant, the m odel’s equations have been corrected for serial correlation problems. Table 1 The Model (1) Y, = C1 + 4 4 2 CM| (Mt _i) + 2 CEi(E,_i) + e1, i= 0 i= 0 4 5 (2) P, = C2 + 2 CPE, (PE, ,) + 2 CD,(Xt l - XF,*_,) i= 0 i= 0 Table 2 In-Sample Estimation: 1/1955—IV/1981 (absolute value of t-statistic in parentheses) + CPA (PA,) + CDUM1 (DUM1) + CDUM2 (DUM2) + e2, (1) Y, = 2.81 + 1 . 1 3 (3.11) (6.91) 20 (3) RL, = 2 CPRLi (Pt . ,) + e3t i= 0 R2 = 0.40 2 CPERSi (PE,_,) + CMRS (M,) i= 1 SE = 1.28 (9) X, = ((X,/X,_,)4 - 1) 100 (10) P, = ((P ./P ,-,)4 - 1) 100 (11) GAP, = ((XF, - X,)/XF,) 100 (12) XF,* = ((XF,/X,_1)4 - 1) 100 Y M E P PE X XF RL RS U UF = = = = = = = = = = = nominal GNP money stock (M1) high employment expenditures GNP deflator (1972 = 100) relative price of energy output in 1972 dollars potential output (Rasche/Tatom) corporate bond rate commercial paper rate unemployment rate unemployment rate at full employment 24 i= 0 DW = 2.13 DW = 2.00 p = 0.16 (3) RL, = 0.87 2 P,_i (3.50) i = 0 e5, R2 = 0.12 SE = 0.32 DW = 1.76 2 p = 1.00 16 (4) RS, = 0.05 2 PE,_i -0 .0 8 M, + 0.77 i= 1 (2.84) (3.22) 2 i= 0 16 (7) Y, = (P./100) (X.) (8) Y, = ((Y,A,,_ 1)4 - 1) 100 (0.06) 20 2 CPRSi (P,_i) + e4, 21 2 CPRLi (P,_i) i= 1 SE = 3.72 R2 = 0.76 i= 0 (6) PA, = 0 + 1.13 PA, - 0.80 DUM1, + 1.72 DUM2, (11.44) (1.33) (2.79) 16 (5) U, - UF, = CG (GAP,) + CG1 (GAP,_,) + i= 5 (2) P, = 1.12 + 0.06 2 PE, , +0.08 2 (X, , - XFt*_,) (3.27) i= 1 (5.00) i = 0 16 + 2 CXRSi(X,_i) i= 0 + 4 2 E,_, 4 2 (4) RS, = 4 2 M,_| - 0.01 + 0.97 (5.62) R2 = 0.32 2 P,_, i =0 SE = 0.90 DW = 1.83 p = 0.89 (5) U, - UF, = 0.29 GAP, + 0.14 GAP,_, (14.85) (6.84) R2 = 0.70 SE = 0.19 p, = 1.33 p2 = -0.44