The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
____________ Review____________ Vol. 68, No. 1 January 1986 5 E s tim a tin g E x c h a n g e R ate E ffe cts on E x p o r ts : A C a u tio n a ry N ote 17 R e c e n t R evisions o f GNP D ata T h e Review is p u b lish ed 10 tim es p e r y ea r by th e R esea rch a n d Public In form ation D ep a rtm en t o f th e F ed era l R eserv e Rank o f St. Louis. Single-copv su bscrip tio ns a re available to th e p ublic f r e e o f ch a rg e. Mail re q u ests f o r su bscrip tio ns, back issues, o r a d d res s ch a n g e s to: R esea rch a n d Public Inform ation D ep a rtm en t, F ed era l R eserv e Rank o f St. Louis, P.O. Rope 442, St. Louis, M issouri 63166. T h e views e x p re s s e d a re th ose o f th e individual a uthors a n d d o not n ecessa rily re flec t official positions o f th e F ed era l R eserve Rank o f St. Louis o r th e F ed era l R eserve System . A rticles h e re in may b e re p rin te d p ro v id ed th e s o u rc e is cred ited . Please p rov id e th e Rank's R esea rch a n d Public Inform ation D ep a rtm en t with a copy o f re p rin ted material. Federal Reserve Bank of St. Louis Review January 1986 In This Issue . . . H om er Jones 1 9 0 6 -1 9 8 6 It is w ith great sorrow that w e m ark the passing of o u r friend and m entor, H om er Jon es. As d irector of re se a rch at the Federal Reserve Bank of St. Louis from 1958 to 1971, H om er p resid ed over a staff that established new stan d ard s for the gathering and publication of m on etary statistics and m on etary research . His role in p rom oting the then -ob scu re notion that m on ey grow th affects eco n om ic activity w as fundam ental, and he w as responsible, in large part, for w hatever reputation this Bank h as today for rigorous, scientific research . H om er cam e to the Federal Reserve Bank of St. Louis in 1958 after working for the Federal Deposit In su ran ce C orporation an d th e Federal Reserve Board. W hen he retired from the St. Louis Bank in 1971, he w as senior vice p resid en t and d irector of research . He earn ed his BA . an d M A. at the University of Iowa in Iowa City, an d his Ph.D. from the University of Chicago. * * * The dollar’s exch an ge rate, w hich has been associated w ith m an y eco n om ic developm ents in recen t years, con tin u es to o ccu p y a prom inent place in policy discussions. In the first article in this Review, “Estim ating E xch an g e Rate Effects on E xp orts: A Cautionary N ote,” M ichael T. Belongia show s th at alternative m easures of the dollar’s value offer quite different p ictu res of its behavior in recen t years. Since 1980, for exam ple, the real value of the dollar rose anyw here betw een 32 p ercen t and 57 p ercen t, depending on the exch an ge rate m easure u sed. Belongia goes on to d em onstrate that qualitative judgm ents about the effects of exch an ge rates on exp orts o r the effects of interest rates an d oth er variables on the exch an ge rate vary substantially, depending on th e p articu lar m easure of the dollar’s value used in the analysis. M oreover, b ecau se there is no generally a cce p te d w ay to d eterm ine w hich exch an ge rate m easu re is “b est,” the a u th o r w arns that con clu sion s b ased on analysis of only one exch an ge rate index m ust be viewed as tentative. In the seco n d article of this issue, "R ecent Revisions of GNP D ata,” Keith M. Carlson d iscu sses the natu re an d extent of the C om m erce D ep artm en t’s recently released revision of the n atio n ’s incom e and p ro d u ct acco u n ts. The article focuses on the effect of the D ep artm en t’s revision on nom inal GNP, outp u t and p rices, and finds that, over the 1 9 4 8 -8 4 period, the revision h ad large effects on the levels of these variables, but very little effect on their rates of change. In addition, Carlson found th at the revision h ad only a m arginal effect on e co n o m etric estim ates of certain key m acro eco n o m ic relationships. 3 Estimating Exchange Rate Effects on Exports: A Cautionary Note Michael T. Belongia L - 7 INCE the ab and on m en t of fixed exch ange rates in the early 1970s, the value of the U.S. dollar has gained increasing p rom in en ce in d om estic and international eco n om ic policy discussions. The dollar’s value gen erally fell against o th er cu rren cies betw een 1973 and 1979; its declining value red u ced U.S. co n su m ers' p u r chasing p ow er as p rices of im ported goods rose rela tive to dom estically p ro d u ced item s. At the sam e time, U.S. industries that relied heavily on foreign sales, su ch as agriculture and m anufacturing, benefitted as prices of U.S. goods fell relative to prices offered bv com peting exporters. This situation was reversed from 1979 to early 1985, w hen the dollar m ad e its persistent rise. Analysts now cite the dollar’s historically high and rising value d u r ing this period as a fundam ental, if not the primary, cau se of declining p ro d u cer incom es and loss of jobs in the U.S. agricultural and m anufacturing industries in recent years. While analysts generally agree on the qualitative asp ects of the exch an ge ra te ’s effect on U.S. exports, the actual m agnitude and p ersisten ce of these effects are subject to considerable controversy. This article dem onstrates that one so u rce of this disagreem ent reflects differences arising from the use of various exch ange rate indexes. Using U.S. agricultural exports as an exam ple, this article show s that an analysis based on different exch an ge rate m easures can ren d er substantially different co n clusions about the U.S. com petitive position in w orld m arkets, the estim ated effects of changes in the dollar’s value on exports and the relationship betw een the exchan ge rate and oth er eco n om ic variables. MEASURING TH E EXCHANGE RATE: AN OVERVIEW In exam ining the effect of exch ange rate m ovem ents on exports, it is tem pting to co n sid er the exports of specific com m od ities to specific cou n tries on a case by-case basis. F o r exam ple, if the U.S. exp orted corn only to Fran ce, Germany and Japan, it might seem reasonable to assu m e that only ch an ges in bilateral exchange rates — that is, changes in the dollar’s value against the franc, d eu tsch e mark (dm) and yen individ ually — affect exp orts to these countries. Yet, this ap p roach w ould be misleading. Aside from p ractical difficulties inherent in h an dling large num bers of bilateral rates sim ultaneously, changes in relative prices, including the relative prices of cu rrencies, ind u ce m any form s of substitution am ong p rodu cers, co n su m ers and nations. For exam ple, a change in the value of the dollar that raised the price of U.S. relative to foreign coi n would cau se im porters of U.S. co rn to import co rn from an oth er c o u n try o r to substitute o th er grains in place of corn in p rodu ction and con su m p tion . This relative price change also w ould give foreign co rn p rod u cers an incentive to increase corn p rodu ction . U.S. p rod u cers receiving a higher d ollar-denom inated price for their corn w ould face a sim ilar incentive — at least in the sh ort run — to shift resou rces from o th e r crop s into corn produ ction . Simplv looking at a variety of bilat eral exchange rate m ovem ents will not cap tu re fully these m any and diverse substitution possibilities; to accom plish this, one needs a single m easure of changes in the dollar's value relative to multiple cu rren cies.' In the sam e w ay that the co n su m e r price index rep resents a w eighted sum of a specific sam ple of m any individual retail prices, an exch an ge rate index is a w eighted sum of the dollar's price in term s of a specific sam ple of foreign cu rren cies. The weights used typically are the p ercen t of total U.S. trade con Michael T. Belongia is a senior economist at the Federal Reserve Bank of St. Louis. David J. Flanagan provided research assistance. 'This judgment, of course, abstracts from the many well-known problems with index numbers, including the use of fixed weights, and choice of base period, sample of countries and mathematical formula. 5 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 1 Percentage Weights Assigned to Major Currencies in Five U.S. Dollar Exchange Rate Indexes Exchange Rate Index Country Germany Japan France United Kingdom Canada Italy Netherlands Belgium Sweden Switzerland Australia Mexico Spain South Korea Denmark All Other TOTAL FRB 20.8 13.6 13.1 11.9 9.1 9.0 8.3 6.4 4.2 3.6 — MERM 13.02 21.25 10.11 5.06 20.28 7.47 3.24 2.44 2.73 1.69 4.86 — — — 2.44 — — 32.8 22.4 22.4 22.4 — — — — — — — MG 10.9 23.2 5.9 9.2 30.3 4.1 3.0 3.5 1.7 2.8 2.4 — 1.17 — — — 100.00 — 1.4 — — 100.0 100.00 100.0 100.0 A variety of alternative trade-w eighted exchange rate indexes have been co n stru cted and used. Among the best-know n are those p ro d u ced by the Federal Reseive Board (FRB), M organ Guaranty (MG), the U.S. D epartm ent of Agriculture IUSDA), the International M onetaiy Fund (MERM) and one co n stru cted from International M onetaiy Fund data on Special Drawing Rights (SDR). Table 1 indicates the w eights that each of these indexes assigns to different foreign cu rrencies. The m ost narrow index is the SDR index, w hich a s 8.99 21.05 2.65 4.63 8.31 4.78 11.26 2.59 — — 0.6 1.0 CHOICES O F EXCHANGE RATE M EASURES USDA 3.37 3.67 4.65 0.95 21.93 — 1.40 4.01 — d u cted with the individual cou n tries selected . C ur rencies ch o sen for the sam ple usually are those of the cou n tries that make up the five or ten largest shares of total U.S. foreign trade. F o r exam ple, excluding im p orts from consideration, if the United States exp orted only co rn and F ran ce bought half, while G erm any and Jap an each bought 25 percen t, an index of the dollar's value could be co n stru cted bv multiplying the franc/ dollar, dm /dollar and ven/dollar bilateral exchange rates by 1/2, 1/4 and 1/4, respectively, and adding up the resulting figures. The sum w ould be an export trade-w eighted index of the dollar’s value against the cu rren cies of these three countries. 6 SDR signs w eights based on the four o th e r cu rren cies (be sides the U.S. dollar) that make up SDRs.The FRB, MERM and MG indexes base their w eights primarily on trade with the G-10 cou n tries an d Switz erland.3 These indexes reflect trad e am on g developed, industrialized econ om ies but do not include lessdeveloped cou n tries' (LDC) cu rre n cy values.4 The MERM and MG indexes, however, are som ew hat m ore broadly based than the FRB index in that they include Australia, Spain and several o th er cou n tries. The USDA index has the broadest coverage, with m ore than 35 p ercen t of its weight given to non-G-10 cou n tries. This index, based only on trad e in agricultural p ro d u cts, is designed specifically to assess ch an ges in the co m p e t itiveness of U.S. agricultural p ro d u cts as the dollar rises o r falls. Especially notable in the USDA index are 2SDRs are the International Monetary Fund’s official unit of account and sen/e as an international reserve asset often used in place of gold for making international payments. Since SDRs are denomi nated in terms of only the U.S. and four other nations’ currencies, however, a dollar exchange rate based on SDR weights reflects changes in the dollar against a very small range of currencies. 3The Group of Ten, or G-10, countries include Belgium, Canada, France, West Germany, Italy, Japan, the Netherlands, Sweden, the United Kingdom and the United States. 4A less-developed country typically is defined as one in which per capita income is less than one-fifth of U.S. per capita income. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 D ifferen ces B etw een A rithm etic an d G eom etric M eans T he Federal Reserve Board ch an ged the co m p u tation of its exch ange rate index from one using arithm etic m ean s to one using geom etric m eans. The Board dropp ed the p ractice of using arithm etic m ean s because, “as cu rren cies diverged from each o th er over time, ch an ges in cu rren cies that rose against the dollar had a red u ced im p act on the index while changes in cu rren cies that fell against the dollar h ad an increased im p act on the index. As a result, arithm etic averaging im parted a system atic upw ard bias to the m easurem en t of chan ges in the dollar’s average exch an ge value.”1 The two form ulas for calculating the index value at time t can be w ritten as: Arithmetic m ean: w, E,. Geom etric m ean: 100 exp 2 w, log E„. i= l As m entioned in the text, the arithm etic m ean is a simple sum of n cu rren cy values (E,) w eighted by 'See Federal Reserve Bulletin lAugust 19781, p. 700. the relatively large w eights given to the N etherlands and su ch LDCs as M exico and South Korea. Problems in Index Construction C onstructing a m ultilateral exch an ge rate index is a difficult m arriage of theory an d p ra ctice .’ For exam ple, choosing a base y ear for an index is difficult because, in theory, this base should be one in w hich absolute pu rch asin g p ow er parity holds and the countries used to co n stru ct the exch an ge rate index con su m e identical com m od ity bundles.15It generally is not p os sible, however, to find a y e a r in w hich absolute p u r chasing p ow er parity held o r actu al consum ption bundles acro ss cou n tries w ere identical. Other p ractical problem s associated with co n structing an exch an ge rate index include the ch o ice of 5See Dutton and Grennes (1985) for a detailed discussion of theoreti cal and statistical issues concerning the construction of exchange rate indexes. A similar discussion focusing on agricultural tradeweighted indexes is in Goolsby and Roberson (1985). 6Absolute purchasing power maintains that the exchange rate will be at a value that equates the price levels between nations. each cu rren cy's weight (w,) in the index. The geo m etric m ean, in co n trast, averages the percentage changes in the individual exch an g e rates to d eter mine the p ercen tage ch an ge in the index. The difference b etw een the form ulas can be illus trated by a simple exam ple. C onsider observations for five exch an ge rates, ea ch with an index weight equal to 0.2, at two points in time. Time 1 Time 2 E, = 100 E, = 110 E3 = 90 E, = 75 E3 = 125 Arithmetic m ean Geometric m ean E, = 100 E, = 110 E3 = 90 E4 = 150 E, = 125 Arithm etic m ean = 115.0 G eom etric m ean = 113.17 100.0 98.52 Using the sam e values, the two techniques produ ce different index values and, thus, grow th rates. For exam ple, changing only E4 betw een time 1 and time 2 p rodu ces a 15 p ercen t change in the arithm etic index and a 14.87 p ercen t ch an ge in the geom etric index. Although this difference in the ch an ges may seem small, sim ilar ch an ges will cau se the gap betw een the two indexes to grow larger and larger over time. weighting sch em es and the m athem atical differences am ong alternative index form ulas.7 One particularly im portant distinction arises betw een indexes that are co n stru cted using arithm etic m eans (Laspevres and Paasche indexes) vs. geom etric m eans. Indexes co n stru cted using arith m etic m ean s give larger w eights to those cu rren cies that ch an ge m ore than o th er cu rre n cies in the index. In co n trast, indexes created by geom etric m ean s resp o n d to proportional exchange rate m ovem ents. F o r exam ple, an exch an ge rate index based on an arith m etic m ean of 10 cou n tries' ex change rates will ch an ge by m ore than an index based on the geom etric m ean of the sam e co u n tries’ cu rre n cies, if som e co u n tries’ cu rren cy values change by m u ch larger am ou nts than the oth ers. Thus, even if two indexes are co n stru cted from the sam e cu rrencies an d the sam e trade weights, the m eth od used to calculate the index ca n p ro d u ce different m easures of ch an ges in the dollar’s value (see sh ad ed box above for one exam ple). 7See Dutton and Grennes, pp. 20-27. 7 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Chart 1 Selected Real Effective Exchange Rates Expressed as V a lu e of Dollar The difficulty of choosing tin exch an ge rate m easure for eco n om ic analysis is p erhaps best illustrated bv the relationships in ch art f and table 2. Using m ea sures of the real exch an ge rate, w hich are the nom inal exchange rate indexes adjusted for differences in p rice levels betw een th e United States an d foreign cou n tries, the ch art show s that, betw een 1973 and 1980, th e real value of the dollar fell as little as 3 p ercen t based on the MG m easure, o r by as m u ch as 14 p ercen t b ased on the FRB m easure. Similarly, the ch art indicates that the real value of the dollar rose by as m u ch as 57 p ercen t (FRB) o r as little as 32 p ercen t (MG) betw een 1980 and 1984. The divergent behavior of these indexes also is evi dent in table 2. The top portion of the table indicates that the USDA index has the low est average quarterly Digitized for 8 FRASER change, sm allest stan d ard deviation an d sm allest val ues for m inim um an d m axim um changes. The SDR index, at the o th e r en d of the sp ectru m , has the largest values for three of these statistics; only the FRB index has a larger value for the m ean quarterly change. The bottom portion of the table, w hich rep o rts simple correlation coefficients, however, show s that changes in each index are co rrelated significantly. Overall, the data in ch art 1 and table 2 indicate that, although m ovem ents in the indexes are positively correlated, there are substantial quantitative differences in their m ovem ents over time. The problem of assessing the im p act of exch an ge rate m ovem ents on exp orts m ight be som ew hat am e liorated if there w ere a clear guide to ch oosin g the best index. But, theoretical and statistical criteria that e s JANUARY 1986 FEDERAL RESERVE BANK OF ST. LOUIS Table 2 Summary Statistics for Changes in Alternative Real Exchange Rate Measures, 1/1973-1/1985 Index Mean Standard deviation Minimum Maximum FRB 0.670 3.741 -5 .8 5 8 8.292 MERM 0.470 3.181 -5 .7 1 2 7.160 SDR 0.594 4.011 -7 .6 4 4 8.747 MG 0.500 3.048 -6 .1 2 2 7.143 USDA 0.260 2.538 -4 .7 8 6 5.725 Correlation Coefficients and Significance Levels Index FRB MERM SDR MG USDA 0.983 (0.0001) 0.919 (0.0001) 0.854 (0.0001) 0.908 (0.0001) 0.976 (0.0001) 0.864 (0.0001) 0.921 (0.0001) 0.853 (0.0001) 0.909 (0.0001) MERM SDR MG 0.835 (0.0001) NOTE: Significance levels in parentheses tablish m inim um stan d ard s of perform ance for an index do not offer clear guidelines for discrim inating am ong alternative indexes that m eet these basic standards.8W ithout guidelines, two questions em erge: Does the ch o ice of an index make a substantial differ en ce in em pirical work? If so, w hat o th er grounds might be used to ch o ose the ap p rop riate index? These questions are investigated below. EM PIRICAL D IFFER EN C ES CAUSED BY ALTERNATIVE EXCHANGE RATE M EASURES: TH E CASE O F FARM EX PO R T S The real issue in estim ating em pirical relation ships betw een exch ange rates an d exp orts depends not so m u ch on the levels of the alternative exchange rate series, but on th eir sp ecific ch a n ges o v er time. That is, if the various index levels differ by con stan t absolute am ou nts (or co n stan t proportions in loga rithms), the m agnitudes of the exch an ge rate coef ficient in an export equation will vary but the m od el’s explanatory pow er will be the sam e across all m ea sures. In con trast, if the exch an ge rate indexes are of sim ilar m agnitude but follow different p ath s arou nd the sam e m ean, both a m odel's exch an ge rate coef ficient and its explanatory p ow er will vaiy. The latter p ro sp ect is particularly relevant if an exp ort equation derived from theory p ro d u ces substantially different estim ates of an exch an ge rate elasticity since there are no clear grounds, a priori, for preferring one single exchange rate index to another. We can illustrate this problem by considering the case of farm exports. A general expression of the ex port dem an d for U.S. farm p ro d u cts can be w ritten as: m (1) lnX, = a + P + (3, In FGNP,_, 2 7, In (USAGP/USCPI)h i= i q + 8See Dutton and Grennes, pp. 8-11, for a discussion of these criteria. 2 i= 0 2 8k In RER,_k + e,, k= l 9 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 3 Estimates of Equation 1 Over a 1/1973—IV/1981 Sample Exchange rate Intercept 2 In FGNP 2 In (USAGP/USCPI) 2 In RER R* DW FRB 4.686 (1.45) 0.825 (2.37) 2 -0 .6 2 0 (5.25) 2 -0.671 (2.49) 3 0.93 1.60 MERM 14.396 (21.22) -0 .9 6 4 (11.31) 8 -1 .5 9 7 (12.59) 7 0.95 1.51 — 4.316 (1.51) 0.809 (2.47) 2 -0 .5 9 4 (5.60) 2 -0 .6 0 3 (2.74) 3 0.93 1.61 -4 .1 0 8 (4.19) 1.779 (12.48) 1 -0 .3 6 5 (4.16) 2 — 0.91 1.16 -1 .2 2 9 (0.24) 1.611 (3.15) 2 -0 .5 9 4 (4.07) 8 0.94 1.66 SDR MG USDA — — -0 .2 2 6 (0.45) 5 NOTE: Absolute values of t-statistics are in parentheses. Lag lengths for right-hand-side variables, chosen in pretest estimation by a final prediction error (FPE) criterion, are shown below the t-statistics. w here: X FGNP USAGP use PI RER e = = = = = real exp orts of all U.S. farm com m odities; foreign real GNP; index of U.S. farm prices; index of U.S. co n su m er prices; real, trade-w eighted exch an g e rate, ex p ressed in foreign cu rren cy units p er dol lar; and = a random erro r term .9 The m odel w as estim ated over several sam ple periods using quarterly d ata.10 The only difference am ong m odels w as the ch oice of an index for the real exch an g e rate from the five series d escribed in table 1. E ach index w as rebased to have a co m m o n value of 100 in 1/1973. Tables 3 an d 4 report these results. Results show n in table 3 apply to the first sam ple period, w hich en ds in the fourth q u arter of 1981 w hen real U.S. farm exp orts peaked; the secon d period results, rep o rted in table 4, cover the entire p eriod of flexible exch an ge rates up to the 9This export equation is derived and discussed in Batten and Belongia (1984). This article also contains more detailed discussion on the distinction between real and nominal exchange rates. 10Lag lengths for right-hand-side variables were chosen by an FPE criterion following procedures outlined in Batten and Thornton (1984). 10 first q u arter of 1985. The critical results are those showing the estim ated elasticities of farm exp orts w ith resp ect to the real exch an g e rate, w hich are show n in the fifth colum n of th ese tables. These values indicate the p ercen tage ch an ge in real farm exp orts that will result from a 1 p ercen t change in the real value of the dollar, as m easured by the various indexes. Although the general statistical ch aracteristics and eco n om ic im plications of the alternative m odels are broadly similar, there is considerable variation am ong the estim ated elasticities, both a cro ss sam ple periods and across exch an g e rate m easures. In table 3, the estim ated exch an ge rate elasticity varies from zero (no effect) for the MG index and —0.23 for the USDA m easure to —1.60 for th e MERM index. Table 4 show s the estim ated exch an ge rate elasticity varies from —0.80 (SDR) to —1.42 (MG). It also is interesting to note that extending the sam ple period raises the exch an ge rate elasticities for the MG an d USDA indexes from zero and —0.23, respectively, to —1.42 an d —1.23 in con trast to o th er indexes, w hich do not exhibit the sam e sensitivity to ch o ice of an estim ation interval. Thus, using th e sam e m odel, it is possible to show that the dem an d for U.S. farm exp orts is eith er elastic or inelastic m erely by changing th e m easure of the dol lar’s value u sed in the analysis. Clearly, the estim ated response of farm exp orts to ch an ges in th e d ollar’s real value is sensitive both to th e ch o ice of sam ple period and the specific exch an g e rate m easure used. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 4 Estimates of Equation 1 Over a 1/1973-1/1985 Sample Exchange rate R2 DW -0 .8 7 8 (9.50) 5 0.93 1.83 -0 .7 5 9 (7.65) 8 -1 .3 8 0 (9-10) 7 0.93 1.78 0.772 (5.20) 0 -0 .5 9 0 (6.67) 7 -0 .7 9 6 (9.38) 5 0.93 1.75 8.096 (4.61) 1.016 (6.81) 0 -0 .8 1 0 (7.19) 5 -1 .4 2 3 (8.50) 5 0.92 1.61 8.451 (4.81) 0.630 (3.68) 0 -0 .6 9 8 (6.48) 7 -1 .2 2 9 (8.58) 5 0.92 1.49 Intercept X In FGNP FRB 5.724 (4.22) 0.819 (5.65) 0 -0 .6 4 3 (7.01) 7 MERM 9.851 (5.90) 0.540 (3.36) 2 SDR 5.305 (3.96) MG USDA 2 In (USAGP/USCPI) 2 In RER NOTE: Absolute values of t-statistics are in parentheses. The number of lags for each right-hand-side variable, chosen in pretest estimation by a final production error (FPE) criterion, are shown below the t-statistics; zero lags indicate a contemporaneous value only. CRITERIA FO R CHOOSING AMONG ALTERNATIVE INDEXES The previous d iscussion d em o n strated that altern a tive exchange rate m easu res diverge widely over time and have different estim ated effects on farm exports. Unfortunately, n eith er eco n om ic theory n or index theory provides a clear criterion for preferring one exchange rate m easu re to another. T here are, h ow ever, two ap p roach es that can be used to indicate w hich index is potentially m ore useful: its out-ofsam ple forecasting perform an ce and its relationship to variables that are thought to affect its value. Out-of-Sample Forecasting Performance The descriptive statistics for the in-sam ple estim a tions of equation 1 do not provide clear grou nd s for preferring a p articu lar exch an ge rate index. This in conclusiveness, as w e noted, leaves open the question of the tru e m agnitude of the exchange rate elasticity. The ch oice of an index, how ever, can be based on how well it p red icts the future path of exports; thus, its outof-sam ple p erform an ce in predicting ch an ges in farm exports is crucial. This criterion is exam ined in table 5 an d ch art 2. The statistics in table 5 are derived from the esti Table 5 Out-of-Sample Error Statistics for Projected Farm Exports Using Alternative Real Exchange Rate Series (1/1982-1/1985) _____________ Mean absolute error RMSE -0 .0 3 2 0.062 0.074 0.021 0.073 0.099 SDR -0 .0 4 4 0.066 0.080 MG -0 .1 8 8 0.190 0.229 USDA -0 .2 2 4 0.224 0.271 Exchange rate series Mean error FRB MERM m ated export equation coefficients rep orted in table 3. The estim ated coefficients an d the actu al values for the equation's right-hand-side variables w ere used to sim ulate paths for farm exp orts over the period 1/1982 to 1/1985. The only difference am on g these alternative p aths is the exch an ge rate m easu re used. Com pari sons of actu al farm exp orts over this interval with each of th e sim ulated p ath s p ro d u ce the erro r sum m ary statistics rep orted in table 5. 11 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Chart 2 Errors in Projected Farm Exports Re a l f a r m e x p o r t s Re a l f a r m e x p o r t s L a t e s t d a t a p l o t t e d : 1st q u a r t e r On the basis of these m easures, th e FRB, MERM and SDR series perform substantially b etter than the oth er two. Ironically, the USDA index, w hich is designed specifically for em pirical work on farm exports, p er forms m u ch w orse than the o th er m easures. M ore over, it is clear from ch art 2, w hich plots the out-ofsam ple (actual m inus predicted) errors m ade in predicting farm exports, that the USDA index co n sis tently overpredicts farm exp ort volum e bv a su bstan tial am ount. The line den oted MG, w hich also indi cates persisten t overpredictions of exports, applies to the m odel that show ed no significant exchange rate effect based on the MG index. These data point out Digitized for 12FRASER why care m u st be taken in ch oosin g a p articu lar e x change rate m easu re for use in em pirical work and farm policy analyses that co n sid er the exp ected future path of farm exp orts. Specifically, the data in table 5 and ch art 2 indicate that, based on equation 1 and estim ates of the MG o r USDA index's future value, future farm exp orts w ould have been consistently overpredicted by large am ounts, even if the exchange rate m ovem ent had been p red icted perfectly." "It should be noted that, as In the previous analysis, these error statistics could vary over sample periods and specifications of export demand equations. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Exchange Rate Indexes and Other Variables w here AlnRER A secon d possible criterion for preferring one index to an oth er is the in d ex’s relationship with variables thought to affect the dollar’s value. This criterion is im portant b ecause projections of future exports n e c essarily involve som e pred iction of the dollar’s future value. Faced with a ch o ice betw een an exch an ge rate index that apparently sh ares no significant relation ship with variables that, theoretically, should in fluence it an d one that is related system atically to, say, ch an ges in interest rates, one w ould prefer the latter index, all o th er things equal. ARID There currently is w idespread debate am ong e co n om ists over w hat factors affect the exch an ge rate. A fairly general theoretical m odel of international cu r ren cy values, however, suggests four variables as the main influences. These include: differences in in flation rates betw een countries, differences in real rates of interest betw een countries, differences in real eco n om ic conditions that affect trad e flows an d differ en ces in political o r o th er risks associated with invest m ents in different cou ntries. We return to this issue by investigating how each of the alternative exch an ge rate indexes respon d s to changes in variables that are proxies for the theoretical factors listed above.'3 The d ependent variable in our investigation is the chan ge in the various m easures of the real exchange rate. To the extent possible, weights and countries used to co m p u te each equ ation ’s righth and-side variables are the sam e as those u sed to calculate the real exch an ge rate m easu re.14 A2CAB e, = the change in the log level of the real exch an ge rate; = the change in the epc ante real interest rate differential betw een the U.S. and foreign cou n tries; = the change in the U.S. cum ulative cu rrent a cco u n t b alance; and = a ran d om erro r term . More detailed variable definitions and m ethods of con stru ction a p p ear in the app en dix to this article. The subscript “t ” indicates quarterly tim e periods. E ach equation w as estim ated over the III/1974-III/1984 time period; the estim ation period is sh orter b ecause of the availability of OECD inflation forecasts needed to co n stru ct the RID variable. The results rep orted in table 6 again reveal som e differences am ong th e alternative exch an ge rate m ea sures. In general, the signs and m agnitudes of individ ual coefficients are sim ilar a cro ss equations. For ex am ple, the co n tem p o ran eo u s and lagged term s for the cu rren t a cco u n t balan ce are significant in ea ch equa tion. In con trast, the lagged real interest differential is significant only in the equations that use the FRB, MERM and USDA indexes. Overall, the MERM index d em onstrates a slightly b etter fit than the oth er m easures. A nother specification of ch an ges in the real ex change rate m aintains the argum ents of the previous model and adds the effects of ch an ges in the grow th rates of the m on ey stock both in the U.S. (AAlnM) and abroad (AAlnM*). This expression can be w ritten as: The first m odel used can be w ritten as: (3) A in RER, = oc + + p 4A 2 CAB,_, + e„ 3 + 2 k= 0 "Derivations of these specifications are based on analyses in Hooper and Morton (1982), Shafer and Loopesko (1983), and Isard. Esti mates for a broader range of specifications for the FRB index only are reported in Batten and Belongia (1986). '“•Construction of the ex ante real interest differential, ARID, de pended on the availability of inflation forecasts for countries in the index. In those cases in which a country was not included in the OECD forecast survey, it was dropped from the analysis and all weights used to construct the index were expanded by a common proportion so the adjusted weights still summed to one. 3 p, A A ln M,_, + i= 0 (21 Ain RKR = a + P, ARID, + p, ARID,., + p., ASCAB, '2These influences are derived from the general framework devel oped by Isard (1983). On the other hand, some economists who have investigated these relationships empirically have found changes in the exchange rate to behave as a random walk. See, for example, Meese and Rogoff (1983) and Hakkio (1985). 3 2 2 7 , A A ln M,* j —0 3 S JS C A B U + 2 t , ARID,.,, p= 0 Although the su m m ary statistics show n in table 7 indicate som e difference in goodness-of-fit across equations, the divergence of the results' qualitative interpretations is m ore interesting. For exam ple, changes in the grow th rate of the U.S. m oney stock have significant effects on the SDR index, but not on the oth er four. Similarly, ch an ges in the real interest differential exhibit significant effects on the FRB, SDR an d MERM indexes, but not on the others. Finally, only the cum ulative cu rren t a cco u n t balance and intercept have a significant effect on the MG and USDA indexes. If w e are looking for an exch an ge rate index that is related significantly to variables that eco n om ic theory 13 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 6 Applications of a Common Exchange Rate Equation to Alternative Exchange Rate Indexes Exchange rate ASCAB,., R2 DW -0.007 (3.59) 0.008 (3.42) 0.28 1.57 0.004 (2.02) -0.006 (3.80) 0.007 (3.60) 0.30 1.66 0.001 (0.46) 0.004 (1.70) - 0.007 (3.54) 0.009 (3.38) 0.26 1.65 0.008 (1.89) 0.001 (0.33) 0.003 (1.70) -0.005 (3.27) 0.006 (3.36) 0.23 1.90 0.005 (1.27) 0.001 (0.78) 0.003 (2.18) -0.005 (3.47) 0.005 (3.26) 0.29 1.66 Intercept ARID, ARID,., A2CAB, FRB 0.005 (1.05) 0.001 (0.71) 0.004 (2.04) MERM 0.004 (0.94) 0.001 (0.63) SDR 0.004 (0.79) MG USDA NOTE: Absolute values of t-statistics in parentheses. suggests should d eterm ine cu rren cy values, the MG and USDA series are th e w eakest can d id ates.15 Choices am ong th e o th er three, however, rem ain problem atical. SUMMARY Changes in th e exch an ge value of the dollar over the past six y ears have been attributed to a wide variety of eco n om ic developm ents. This article h as show n, how ever, that determ ining how m u ch the dollar has ch an ged an d w hat effect it has had on o th er variables can depend on the specific exch an ge rate index ch o sen for the analysis. Both the set of cou n tries included in the index and the w eighting sch em e used to aggre gate m ovem ents in foreign cu rren cy values will affect the interpretation. Using farm exp orts as one exam ple, the analysis show ed that different exch an ge rate indexes p rodu ce large differences in the estim ated effects of exchange rates on exports. M oreover, fu rther analysis show ed that different indexes exhibit substantial differences in their ability to p red ict future ch an ges in the volum e of exports. Finally, if one is interested in the effects of changes in m oney grow th, interest rates, the cu rren t accou n t balance or o th er variables on the exchange ,5Estimates of other equations showed a similar diversity of results in which no right-hand-side variable was significant in all equations and different combinations of variables were significant across exchange rate measures. 14 rate, one m u st realize that th e significance an d m agni tude of su ch effects vary widely a cro ss exch an ge rate m easures. B ecau se n eith er eco n o m ic n o r statistical theory gives a clear indication of w hich exch an ge rate index is the "b est” m easure, th ese broad differences in results suggest that considerable cau tion be used in relying on a single exch an ge rate m easu re to indicate the effects of ch an ges in the dollar’s value on exports. R EFER E N C ES Batten, Dallas S., and Michael T. Belongia. “The Recent Decline in Agricultural Exports: Is the Exchange Rate the Culprit?" this Re view (October 1984), pp. 5-14. _________ "Monetary Policy, the Real Exchange Rate and U.S. Agricultural Exports,” American Journal of Agricultural Economics (May 1986), forthcoming. Batten, Dallas S., and Daniel L. Thornton. “How Robust Are the Policy Conclusions of the St. Louis Equation?: Some Further Evidence," this Review (June/July 1984), pp. 26-32. Dutton, John, and Thomas Grennes. “The Measurement of Effec tive Exchange Rates Appropriate for Agricultural Trade,” Depart ment of Economics and Business (November 1985), North Caro lina State University. Goolsby, O. Halbert, and Ronald R. Roberson. “Exchange Rate Developments and Their Impact on U.S. Agricultural Exports: 1970-84," U.S. Department of Agriculture, FAS Staff Report No. 5 (May 1985). Hakkio, Craig. “Does the Exchange Rate Follow a Random Walk? A Monte Carlo Study of Four Tests for a Random Walk,” Research Working Paper 85-02, Federal Reserve Bank of Kansas City (June 1985). FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 7 Applications of a Common Exchange Rate Equation to Alternative Exchange Rate Indexes Intercept 3 2 AAlnM,., i= 0 3 2 AAlnM'•H j= 0 FRB 0.013 (2.99) 3.728 (1.81) -0 .9 5 9 (0.63) -0.001 (4.31) MERM 0.011 (2.64) 2.360 (1.15) -0 .4 0 5 (0.27) SDR 0.012 (2.74) 5.202 (2.46) MG 0.013 (3.33) USDA 0.011 (3.23) Exchange rate 3 2 (SCAB),.* k= 0 3 2 ARID,., p= 0 R2 DW 0.020 (2.72) 0.60 2.34 -0.001 (3.87) 0.015 (2.23) 0.52 2.24 -1 .7 1 5 (1.20) -0 .0 0 2 (4.69) 0.026 (3.29) 0.64 2.41 1.300 (0.61) 0.596 (0.42) -0.00 1 (3.13) 0.011 (1.69) 0.46 2.71 1.836 (1.22) -0 .1 4 5 (0.12) -0.001 (4.32) 0.009 (1.75) 0.52 2.20 NOTE: Absolute values of t-statistics in parentheses. Hooper, Peter, and John Morton. “Fluctuations in the Dollar: A Model of Nominal and Real Exchange Rate Determination,” Jour nal of International Money and Finance (April 1982), pp. 39-56. Meese, Richard, and Kenneth Rogoff. “Empirical Exchange Rate Methods of the Seventies: Do They Fit Out of Sample?" Journal of International Economics (February 1983), pp. 3-24. Isard, Peter. “An Accounting Framework and Some Issues For Modeling How Exchange Rates Respond to News,” in Jacob A. Frenkel, ed., Exchange Rates and International Macroeconomics (University of Chicago Press, 1983), pp. 19-56. Shafer, Jeffrey, and Bonnie E. Loopesko. “Floating Exchange Rates After Ten Years,” Brookings Papers on Economic Activity (1:1983), pp. 1-86. (See appendix on next pagel 15 JANUARY 1986 FEDERAL RESERVE BANK OF ST. LOUIS APPENDIX D efinitions of V ariables Used in Equ ations 1—3 1 R e a l I n te r e s t D ifferen tial (RID) OECD forecasts of the CPI for individual countries for July are applied to quarters 1 and 2; forecasts for D ecem ber are u sed for quarters 3 and 4. These tradew eighted e* ante inflation differentials are then sub tracted from a trade-w eighted nom inal interest differ ential u sin g M organ G u aran ty T ru st th re e - to four-m onth com p arable m oney m arket rates. C u rre n t A c co u n t B a la n c e s (SCAB) U.S. cu rren t acco u n t balan ce accu m u lated since 1970; billions of dollars. T ra d e -W e ig h te d R e s t o f W o rld M oney (M‘ ) M oney stock for various cou n tries indexed to 1/1973 and w eighted by sam e trad e w eights u sed in co n stru c tion of the respective exch an ge rate indexes. 'Trade-weights for each variable are those applied to the respective exchange rate indexes. All exchange rates are real and indexed to 1/1973 = 100. 16 U.S. M oney S to ck (M l) M l indexed to 1/1973. F o re ig n GNP (FGNP) Foreign real GNP o r GDP m easu res indexed to 1/1973 and trade-w eighted. U.S. GNP U.S. real GNP indexed to 1/1973. USAGP Unit value of agricultural exp orts index; 1/1973 = 100. USCPI U.S. co n su m e r p rice index; 1/1973 = 100. Recent Revisions of GNP Data Keith M. Carlson I n D ecem ber 1985, the U.S. D epartm ent of C om m erce an n ou n ced a m ajor revision of the n atio n ’s incom e and p rod u ct acco u n ts.' This revision, w hich is d one about every five years, w as the eighth of its kind. The purp ose of this com prehensive revision w as to upd ate the gross national p ro d u ct (GNP) accou n ts, reflecting any new inform ation, new p ro ced u res, and changes in the eco n om ic stru ctu re. The U.S. incom e an d p ro d u ct acco u n ts w ere cre ated in the 1930s, though they w ere not published on a regular basis until after W orld W ar II * Their p u rp ose is to provide a m easure and u nderstanding of the e c o n om ic health of the nation. (For a brief su m m aiy of national incom e accoun ting, see the sh ad ed box on p. 18.) This article d iscusses the natu re and extent of the m ost recen t revision, along w ith som e background information to aid the n on tech n ical reader. The article focuses on the effect of the revision on GNP, output and prices. The effect of the revision on the interp reta tion of post-W orld W ar II eco n o m ic fluctuations and on certain key historical relationships also receives consideration. TH E MAGNITUDE O F T H E REVISION The sh ad ed box on page 20 describes the m ajor so u rces of the revision. Although GNP d ata for earlier y ears w ere also affected som ew hat, the revision pri m arily affected GNP data from 1970 to 1984. Nominal GNP Table 1 sum m arizes the effect of the revision on nom inal GNP for alternate y ears from 1948 to 1984. The revision has in creased th e level of GNP in ea ch y e a r show n; the largest changes, however, have o ccu rre d since 1970. The revision h ad little im p act on the a n nual grow th rates of nom inal GNP; it raised the grow th rate from 1 9 4 8 -8 4 from 7.6 to 7.7 p ercen t. Real GNP Growth Keith M. Carlson is an assistant vice president at the Federal Reserve Bank of St. Louis. Sandra Graham and Thomas A. Pollmann provided research assistance. 'A detailed discussion of the revision can be found in various articles in the Survey of Current Business. See U.S. Department of Com merce (1985b, 1985c). 2For a discussion of the historical development of the U.S. income and product accounts, see U.S. Department of Commerce (1985a). Nominal GNP revisions can be co m p ared directly in term s of dollar am ou n ts; constant-dollar, o r real, GNP estim ates can n o t be as easily co m p ared b ecau se the base period h as been shifted. Consequently, to co m p are the effect of the revision on real GNP estim ates, one m u st exam ine its im p act on the grow th rates of the old and revised real GNP estim ates. 17 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 T h e E ssen tials of National In co m e and P ro d u ct A ccounting The national incom e an d p ro d u ct a cco u n ts p ro vide a statistical sum m ary of th e econom y, showing the volum e, com position an d uses of the national outp u t. The total p rod u ction of the nation is m ea su red in two w ays: in term s of p ro d u cts, that is, the value of goods and services, an d in term s of the incom es generated in p rodu ction . The acco m p a n y ing table sum m arizes the national incom e and p ro d u ct a cco u n t for 1984. The left side of the table, the incom e side of the acco u n t, show s w ages and salaries an d o th er forms of incom e, indirect business taxes an d capital co n sum ption allow ances (and o th er small items) gen erated in the p roduction p ro cess. The total of these item s is labeled "charges against gross national p roduct." (Because the two sides of the incom e and p ro d u ct acco u n t are estim ated independently, given im perfections in the so u rce data, they are not necessarily equal. The erro r is called statistical dis crep an cy; it has no eco n om ic significance.) The right side of the table, the p ro d u ct side, is divided into the m ajor m arkets for the e co n o m y ’s output: personal con su m p tion , business invest m ent, governm ent p u rch ases an d n et exp orts. The su m of the expen d itu res is th e gross national p ro d u ct (GNP). The table is only one of m any in the a cco u n ts, but it is the m ost fundam ental one. Am ong the m ost im portant of the rem aining a cco u n ts are those that show the receip ts an d exp en d itu res of the m ajor eco n om ic groups in the econ om y. The personal incom e an d outlay a cco u n t show s the in com e re ceipts and expen d itu res of p ersons. The govern m en t receip ts and expen d itu res a cco u n t su m m a rizes the activities of federal, state an d local governm ents. The foreign tran saction s a cco u n t sum m arizes international tran saction s that im pinge on U.S. incom e an d p ro d u ct. Finally, the gross saving and investm ent a cco u n t cu ts a cro ss eco n om ic groups, show ing th eir saving an d invest m en t tran saction s in su m m ary form. National Income and Product Account, 1984 (billions of dollars) Compensation of employees Proprietors’ income Rental income Corporate profits and inventory valuation adjustment Net interest National income Business transfer payments Indirect business tax and nontax liability Less: Subsidies less current surplus of government enterprises Capital consumption allowances Statistical discrepancy $2,221.3 233.7 10.8 273.3 300.2 SOURCE: Council of Economic Advisers. 18 $2,423.0 674.0 GROSS NATIONAL PRODUCT $3,774.7 736.8 -5 9 .2 3,039.3 17.3 310.6 10.1 418.9 - 1 .5 CHARGES AGAINST GROSS NATIONAL PRODUCT $3,774.7 NOTE: Numbers may not add due to rounding. Personal consumption expenditures Gross private domestic investment Government purchases of goods and services Net exports of goods and services JANUARY 1986 FEDERAL RESERVE BANK OF ST. LOUIS Table 1 Table 2 A Comparison of Old and Revised Nominal GNP: 1948-84 (dollar amounts in billions) The Growth of Real GNP: Old and Revised Series (compounded annual rates of change) 1948 1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 Old Revised Percent change $ 259.5 286.5 348.0 366.8 421.7 449.7 506.5 565.0 637.7 756.0 873.4 992.7 1,185.9 1,434.2 1,718.0 2,163.9 2,631.7 3,069.3 3,662.8 $ 261.6 288.3 351.6 372.5 428.2 456.8 515.3 574.6 649.8 772.0 892.7 1,015.5 1,212.8 1,472.8 1,782.8 2,249.7 2,732.0 3,166.0 3,774.7 .81% .63 1.03 1.55 1.54 1.58 1.74 1.70 1.90 2.12 2.21 2.30 2.27 2.69 3.77 3.97 3.81 3.15 3.06 Table 2 sum m arizes, on a peak-to-peak basis, the grow th of the old an d revised estim ates of real GNP from 1948 to 1985. The grow th of real GNP w as higher only for the earliest period, w hich includes the d e fense buildup for the Korean W ar. All o th er revised peak-to-peak grow th rates w ere low er; as a result, real GNP grow th for the entire IV/1948-III/1985 period w as revised dow nw ard about 0.2 p ercen t, from a 3.4 p er ce n t annual grow th rate using the old estim ates to a 3.2 p ercen t rate with the revised data. GNP Deflator Changes in the GNP deflator reflect ch an ges in both p rices and the com p osition of spending. C on se quently, revision of the GNP a cco u n ts affects estim ates of th e deflator via several ch an n els. Table 3 su m m a rizes rates of change in the GNP deflator for peak-topeak periods from 1948 to 1985. With only two excep tion s, IV /1948-II/1953 and 1/1980—III/1981, the ch an g e in the deflator w as revised upw ard. In con ju n ction w ith the virtually identical sized revisions in the grow th of real GNP su m m arized in table 2, it is clear that the revision prim arily red is tributed a given change in nom inal GNP from real output to higher p rices. F or the period as a w hole, the Peak-to-Peak Previous Revised IV/1948 - 11/1953 11/1953 —111/1957 111/1957-1/1960 1/1960-111/1969 111/1969-IV/1973 IV/1973 -1/1980 1/1980-111/1981 111/1981 -111/1985' 5.3% 2.2 3.0 4.2 3.5 2.7 1.1 2.6 5.7% 1.8 2.8 4.0 3.0 2.5 0.6 2.4 IV/1948-111/1985' 3.4 3.2 Direction of revision + - - 'Data calculated by the previous method are not available after 111/1985. Table 3 Changes in the GNP Deflator: Old and Revised Series (compounded annual rates of change) Peak-to-Peak Previous Revised IV/1948-11/1953 11/1953-111/1957 111/1957-1/1960 1/1960-111/1969 111/1969-IV/1973 IV/1973-1/1980 1/1980-111/1981 111/1981 -111/1985' 2.2% 2.5 1.9 2.6 5.2 7.6 9.8 4.1 1.9% 2.9 2.3 2.8 5.9 8.0 9.6 4.3 IV /1 9 4 8 -111/1985’ 4.1 4.3 Direction of revision _ + + + + + - + + 'Data calculated by the previous method are not available after 111/1985. revised deflator in creased at a 4.3 p ercen t annual rate, up slightly from the previously estim ated 4.1 p ercen t rate. TH E E F F E C T O F TH E REVISION ON BUSINESS CYCLES As pointed out above, the revision had only a m inor effect on the grow th of nom inal GNP: the grow th of real GNP w as revised dow nw ard slightly an d th e in- 19 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 T h e S o u rce s of National In co m e an d P ro d u ct A ccou n ts Revision The C om m erce D epartm ent divides the so u rces of revision into two m ajor categories: (1) definitional an d classificatory, an d (2) statistical.1 Definitional and classificatoiy ch an ges upd ate the a cco u n ts to reflect the changing stru ctu re of the U.S. econ om y. Statistical ch anges in corp orate newly available and revised so u rce data, im proved estim ating p ro ce du res and a shift in the base period for calculating co n stan t-d ollar estim ates an d the associated price indexes. Definitional and Classificatory Changes This category includes: (1) reclassification of c e r tain bu siness exp en d itu res as investm ent, (2) ch an ged treatm en t of federal em ploym ent benefit program s, (3) changed treatm ent of certain foreign tran sactions, and (41 reclassification of certain gov ern m en t assistance program s. Despite n u m erou s definitional an d classificatory changes, the revision of nom inal GNP arising from this so u rce w as pri marily attributable to (1) the capitalization of m ajor replacem en ts to residential stru ctu res and (2) the im putation of a social insu ran ce fund for military retirem ent. The net effect of o th e r ch an ges on GNP w as slightly negative. Exp enditures for the rep lacem en t of m ajor items (like a roof o r a heating system) in a h ou se w ere reclassified as investm ent in residential stru ctu res. Previously, su ch replacem en ts w ere ch arged off to cu rren t exp en se. This ch ange increased nom inal GNP by $14.1 billion in 1984. The Defense Authorization Act of fiscal 1984 es tablished a militaiy retirem ent trust fund in w hich contributions by the governm ent are equal to benefits paid. These exp en ditu res are now treated as national defense p u rch ases; previously, su ch benefits h ad been included in governm ent transfer paym ents. This ch an ge increased nom inal GNP in 1984 by $16.7 billion. Statistical Changes T hese ch an ges include the shift of the base p e riod from 1972 to 1982, the incorporation of new and revised d ata from regularly u sed so u rces avail able annually o r on a "benchm ark basis," the u se of new so u rce data, an d new estim ation p ro ced u res. The statistical ch an ges w ith the largest im p act w ere as follows: (1) Im pro v ed adjustm ents fo r m isrep o rtin g on ta,x return s. Although these ad ju stm ents are re lated to "underground" activities, the ad ju st m ent itself is not a m easure of the s iz e of the u nderground econom y.- T h ese ad justm ents increased 1984 nom inal GNP by $44.1 billion. (2) Im p ro v ed m ethodology and new data f o r re s i dential investm ent. Residential investm ent w as revised upw ard by $25.2 billion in t9 8 4 .' About half of this increase w as attributable to the new p ro ced u re of capitalized m ajor re p lacem en ts to stru ctu res; the rest reflected statistical ch an ges due to new data. (3) T he shift in the base p e rio d fr o m 1972 to 1982. This shift red u ced the rate of real grow th and, for a given path of nom inal GNP, increased the rate of change in the GNP defla tor. See box on opposite page for an exam ple. (4) Im pro v ed p ric e index f o r co m p u ters. This ch an ge had no effect on nom inal GNP; how ever, it im proved estim ates of real p ro d u ce rs’ durable equipm ent exp en d itu res. Previously, the C om m erce D epartm ent had assu m ed that co m p u te r p rices had rem ain ed u n ch an ged. The C om m erce D epartm ent now incorporates a 10 p ercen t decline p e r y e a r in co m p u te r prices from 1970 to 1984. This ch an ge substantially in creases estim ates of real co m p u te r exp en d itu res over the period. 2For example, GNP does not include illegal activities. For a de tailed discussion of the underground economy, see Carson (1984) and Parker (1984). 'A listing of these changes is provided in U.S. Department of Commerce (1985b, 1985c). http://fraser.stlouisfed.org/ 20 Federal Reserve Bank of St. Louis 3This leaves $14.9 billion of statistical changes affecting nominal GNP that are attributable to other changes. The Commerce Department did not allocate these remaining changes. JANUARY 1986 FEDERAL RESERVE BANK OF ST. LOUIS T h e Effect of Shifting th e B ase P e rio d o n R eal GNP G row th The effect on real GNP of shifting to a m ore recent base period can be show n by using a simplified exam ple in w hich there are only two com m odities, A and B. Heal GNP can be obtained by multiplying the quantities of A an d B sold in each y ear by their prices in the base period. F o r exam ple, real GNP grow th can be calculated as follows: (2) Using y e a r 2 as base period: Quantity (number) Price (dollars) A B Value (dollars) Year 1 Year 2 Year 1 Year 2 Year 1 Year 2 $5 6 $9 10 10 11 $ 90 70 $ 99 $160 $211 7 16 112 $211 Real GNP growth = (1) Using y ear 1 as base period: Quantity Price (dollars) A B (number) Value (dollars) Year 1 Year 2 Year 1 Year 2 Year 1 Year 2 $5 6 $9 7 10 10 11 16 $ 50 60 $ 55 96 $110 $151 $151 Real GNP growth = 1.373 or 37.3% crease of the GNP deflator w as revised upwar d slightly. Because these ch an ges are due chiefly to the shift of the base period from 1972 to 1982, they h ad no signifi can t effect on the general m ovem ent of prices and real GNP over the post-W orld W ar II period. Table 4 sum m arizes real GNP grow th over exp an sions and co n tractio n s on the old an d the revised basis. An exam ination of the quarterly m ovem ents of real GNP arou n d turning points reveals no ch an ges in the timing of the business cycle. T here w ere, however, som e ch anges in the severity of co n tractio n s an d the strength of expansions. The left side of table 4 reveals th at real grow th in all eco n om ic expansions w as revised dow nw ard, excep t for the Korean W ar expansion of 1949—53. Real grow th during the 1 9 7 0 -7 3 an d 1 9 8 0 -8 1 exp an sion s w as re d u ced m ost by the revisions; all revisions, however, w ere m inor. Moreover, the ordering of the expansion periods from strongest to w eakest w as left u n chan ged by the revision. The right side of table 4 sum m arizes the effect of the revision on the severity of recession s. The effect w as $160 1.319 or 31.9% In both cases, the grow th rate of real GNP is a w eighted average of grow th rates of A and B. In case 1, the w eights are based on the p rices in y e a r 1; in case 2, the w eights are based on prices in y e a r 2. The exam ple reflects the assu m p tion that the price of A rises m ore than the price of B, while the quantity of A increases less than B. As a result, A receives m ore weight w hen y e a r 2 is used as the base period than w hen y e a r 1 is used. not as uniform as for expansions: recession ary d e clines in real GNP w ere revised upw ard during som e co n tractio n s and dow nw ard during others. Five co n traction s w ere found to be m ore severe than previ ously estim ated, although in no case w as the revision dram atic. The largest dow nw ard revision in real grow th w as for the 1948-49 recession . TH E E F F E C T O F TH E REVISION ON KEY MACROECONOMIC RELATIONSHIPS One question of interest to econ om ists is w heth er the revision influenced certain key m acroecon o m ic relationships that are used in analyzing the econ om y an d form ulating eco n om ic policy. While m any rela tionships cou ld be exam ined, this section focuses specifically on four of th e m .1 Simple sum m ary rela tionships w ere estim ated for the 1 9 5 6 -8 4 period using 3For a summary and discussion of such relationships for the 1956-81 period, see Carlson and Hein (1983). 21 FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 Table 4 Real GNP Growth over the Business Cycle: Old and Revised Series (compounded annual rates of change) Expansion iV/1949 —11/1953 11/1954-111/1957 1/1958-1/1960 IV/1960-111/1969 IV/1970 - IV/1973 1/1975-1/1980 11/1980-111/1981 111/1982-111/1985 Previous Revised 7.3% 3.9 5.5 4.7 5.3 4.4 3.3 4.6 8.0% 3.4 5.4 4.5 4.6 4.1 2.7 4.3 Change Contraction + 0.7 -0 .5 -0 .1 - 0 .2 - 0 .7 -0 .3 - 0 .6 -0 .3 IV/19 4 8 - IV /1 949 11/1953-11/1954 111/1957-1/1958 1/1960- IV/1960 111/1969- IV /1 970 IV/1973-1/1975 1/1980-11/1980 111/1981 -111/1982 p ercen tage ch an ges (where applicable) on a fourthq uarter-to-fou rth -q uarter basis. No attem p t w as m ade to search for the “b est” equation; rather, the equations w ere ch o sen for th eir illustrative simplicity. They are intended solely to illustrate the effect of the revision on the various relationships in the sim plest form p os sible. Money and Nominal GNP The relationship betw een m oney an d GNP is a fun dam ental one in term s of the m onetarist view of how total spending is determ ined. In a simple version, it ca n be estim ated as the relationship betw een the fourq u arter p ercen t ch an ge of nom inal GNP (Y4) and the four-quarter p ercen t change of m oney (M J.‘ The eq u a tion u sed h ere also includes a dum m y variable (D) for the 1982—84 period b ecau se previous studies have indicated that the relationship shifted significantly after 1981.5 W hen this equation w as estim ated over the 1 9 5 6 -8 4 period, using both the previously published an d re vised data, the results w ere those show n in lines l a an d lb of table 5. An insp ection of the estim ated equations indicates a slight strengthening in the rela tionship betw een nom inal GNP and m oney, w ith the coefficient on m on ey staying close to its theoretically exp ected value of one. The t-statistics (m easures of the precision of the coefficient estim ates) increased ; R2, a m easu re of the explanatory pow er of the equation, also rose. The stan dard erro r (SE) of the equation, a Previous Revised -1 .4 % - 3 .2 - 6 .6 - 1 .5 - 0 .5 - 3 .9 - 9 .0 - 3 .0 -2 .0 % -3 .0 - 7 .0 - 1 .4 - 0 .7 - 3 .5 -9 .1 - 3 .4 Change - 0 .6 + 0.2 - 0 .4 + 0.1 - 0 .2 + 0.4 -0 .1 -0 .4 m easure of the a ccu ra cy of the fitted equation in term s of its d ep en den t variable, w as red u ced by 4 p ercen t. The Durbin-W atson (DW) statistic, a m easu re of resid ual correlation, show ed a slight im provem ent. Inflation and Money Growth The relationship betw een inflation and m oney grow th is an oth er fundam ental one in m a cro e co n o m ics. Since, during the 1970s an d 1980s, ch an ges in the price of energy played a key role affecting m ovem ents of the p rice level, this variable w as also includ ed in the estim ation of the relationship. The estim ated equation for inflation (PJ includes the 16-q u arter rate of ch an ge of m oney (M J m easu red from fourth q u arter to fourth quarter, the four-quarter p ercen t ch an ge of the rela tive p rice of energy (P1,), an d the du m m y variable d iscu ssed earlier." W hen estim ated over the 1 9 5 6 -8 4 period, th e results w ere those show n in lines 2a an d 2b of table 5. As the statistics show, the revision im proved the inflation equation m arginally; both R- an d the stan d ard erro r im proved slightly, and the coefficient on m on ey stayed close to its exp ected value of one. In addition, the t-statistics all increased. Signs of positive a u to co r relation also ap p eared to be rem oved. Unemployment Rate and Real GNP A nother relationship of interest to m a cro e co n o m ists is the relationship betw een th e u n em ploym ent rate and the grow th of real GNP, a variant of w hat is called Okun’s law. In the simple relationship esti- 4For estimation purposes, only fourth-quarter data were used from each calendar year. 5With the exception of the unemployment-real GNP equation, results presented here include this dummy variable. 22 6The choice of 16 quarters for money growth reflects previous re search. See Carlson and Hein. FEDERAL RESERVE BANK OF ST. LOUIS JANUARY 1986 m ated below, AU4 is the change in the unem ploym ent rate from fourth q u arter to fourth quarter, and X4 is the p ercen t change in real GNP from fourth q u arter to fourth quarter. Table 5 Macroeconomic Relationships Using Old and Revised Data Money and Nominal GNP (1a) Using previously published data: Y4 = 3.83 + .89 M„ - 3.51 D (4.10) (5.10) (2.35) R2 = SE = DW = .46 2.25 2.12 R2 = SE = DW = .50 2.15 1.97 (1b) Using revised data: Y4 = 3.80 + .91 M4 - 3.77 D (4.26) (5.43) (2.64) Inflation and Money Growth (2a) Using previously published data: P4 = .08 + .97 M16 + .08 P4 - 3.05 D (.14) (7.68) (3.06) (3.46) R2 = SE = DW = .82 1.15 1.65 R2 = SE = DW = .84 1.10 1.89 R2 = SE = DW = = Pi .70 .64 2.01 - .2 8 R2 = SE = DW = = Pi .70 .67 1.99 - .3 7 R2 = SE = DW = = Pi .54 1.25 1.89 .74 R2 = SE = DW = = Pi .39 1.40 1.88 .85 (2b) Using revised data: P4 = .21 - 1.00 M)6 + .07 PE 4 - 3.10 D (.39) (8.31) (3.14) (3.67) Unemployment Rate and Real GNP (3a) Using previously published data: AU4 = 1.20 - .34 X„ (7.36) (8.15) (3b) Using revised data: AU4 = 1.15 - .35 X4 (7.34) (8.11) W hen this relationship w as estim ated from 1956 to 1984, the results w ere those show n in lines 3a an d 3b of table 5. B ecau se the residuals w ere negatively c o rre lated, the equations w ere adjusted for first-order serial correlation. The estim ates indicate th at the exp lan a tory p ow er of the relationship w as u n ch an ged using the revised d ata and that a 1 p ercen t increase in output still red u ces the unem ploym ent rate by about one-third of a p ercen tage point. The stan d ard erro r increased only slightly, and the estim ated coefficients did not change significantly. Short-Term Interest Rate and Inflation Interest rates generally m ove w ith the exp ected rate of inflation. B ecau se e x p ected inflation can n o t be ob served directly, estim ates of its effect on interest rates require the use of “p roxies”; the actu al rate of ch an ge in the GNP deflator is u sed here as an approxim ation for the e x p ected rate in the interest rate equation. The four-m onth com m ercial p ap er rate (RS) w as estim ated as a function of the four-quarter rate of inflation (P4) m easured from fourth q u arter to fourth q u arter and the dum m y variable d escribed previously.7 It w as n e c essary to estim ate the equation using a first-order serial correlation adjustm ent. Lines 4a and 4b of table 5 show th e results. The sh o rt-term interest rate relationship d eteriorated w hen estim ated w ith the revised data. Such a result is probably n ot surprising, sin ce th e revised data are different than those that w ere used by m arket partici p an ts to form exp ectation s. Even though the coef ficient on inflation declined, it is not significantly different from one, its theoretically exp ected value. Short-Term Interest Rate and Inflation SUMMARY (4a) Using previously published data: RS = 2.40 + .91 P4 + 1.22 D (2.07) (5.67) (.92) (4b) Using revised data: RS = 3.47 + .72 P4 (1.98) (3.95) .69 D (.45) NOTE: Absolute value of t-statistics in parentheses. The D epartm ent of C om m erce h as recently revised the GNP acco u n ts. The revision results from a variety of changes, including a shift of the base period from 1972 to 1982. This ch an ge in base period affects constant-dollar, o r real, estim ates as well as serving as the base y e a r for the p rice indexes. 7A similar attempt was made to estimate a long-term interest rate equation but the results were meaningless. Conventional adjust ments were unsuccessful in removing the positive correlation of the residuals. 23 FEDERAL RESERVE BANK OF ST. LOUIS RO. BOX 442 ST. LOUIS, MISSOURI 63166 Subscriber: Please include address label with subscription inquiries or address changes. The revision resulted in substantial increases in the level of nom inal GNP from 1948 to 1984. It had little effect on the rates o f ch a n ge of GNP. The revised figures for real GNP yield a slow er p ace of eco n om ic grow th; it w as revised dow nw ard from a 3.4 p ercen t annual rate to a 3.2 p ercen t rate from 1948 to 1985. The rate of change of the GNP deflator w as revised upw ard, from a 4.1 p ercen t rate to a 4.3 p ercen t rate over the period. While the revision had no effect on b usiness-cycle turning points, it h ad som e im p act on th e strength of exp an sion s an d the severity of recession s. Revisions of the grow th of real GNP over the business cycle w ere w ithin the —0.7 to + 0 .7 p ercen tage-p oin t range. m arginally using the revised data. The o th e r relation ships deteriorated marginally. On net, th e revision had no m ajor effect on the p attern of recen t fluctuations in the econom y. R E FE R E N C E S Carlson, Keith M., and Scott E. Hein. “Four Econometric Models and Monetary Policy: The Longer-Run View,” this Review (Janu ary 1983), pp. 13-24. Carson, Carol S. “The Underground Economy: An Introduction," Survey of Current Business (May 1984), pp. 21-37. Parker, Robert P. “Improved Adjustments for Misreporting of Tax Return Information Used to Estimate the National Income and Product Accounts, 1977,” Survey of Current Business (June 1984), pp. 17-25. This article also exam ined the effects of the revision on sim ple versions of certain key m acro eco n o m ic rela tionships. These relationships cover the im p act of m on ey grow th on nom inal GNP and inflation, the relationship betw een real GNP grow th an d u nem p loy m ent, an d th e im p act of inflation on sh ort-term inter est rates. ________ _ “An Advance Overview of the Comprehensive Revision of the National Income and Product Accounts,” Survey of Current Business (October 1985b), pp. 19-28. The results w ere m ixed. The two relationships link ing m on ey grow th to GNP and inflation im proved ________ _ “Revised Estimates of the National Income and Prod uct Accounts of the United States, 1929-85: An Introduction,” Survey of Current Business (December 1985c), pp. 1-19. U.S. Department of Commerce, Bureau of Economic Analysis. “Simon Kuznets and the Early Development of National Income and Product Estimates,” Survey of Current Business (July 1985a), pp. 27-28.