The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 70, No. 5 September/October 1988 3 Have Li.S. Exports Been Larger Than Reported? 24 W hy Have State Per Capita Incom es Diverged Recently? 37 Ttesting the Expectations M odel o f the Term Structure: Som e Conjectures on the Effects o f Institutional Changes 4G The Puzzling Growth o f the M onetary Aggregates in the 1980s THE FEDERAL ARESERVE RANK of rtf ST. LOUIS 1 Federal Reserve Bank of Si. Louis Review Septem ber/O ctober 1988 In This Issue . . . In the first article o f this Review, M ack Ott investigates the relation b e tw een the in creasin gly large U.S. m erch andise trade deficits occu rrin g since the mid-1970s and a lesser-known, p u zzlin g attribute o f the U.S. bal ance o f paym ents du rin g this era — the in creasin gly large statistical d is crepancy. Starting from the accou n tin g relation betw een export reportin g errors and the disagreem en t b etw een the current and capital sides o f the balance o f paym ents, Ott builds the case for the discrepan cy b ein g evi d en ce o f a persistent u n d errep ortin g o f U.S. m erch andise exports. In direct evid en ce consistent w ith this v ie w from studies o f u n d errep ortin g in in ter national services trade and from U.S. d om estic studies o f understated tax able in com e are discussed to m otivate the statistical tests w h ic h follow . T h e test results are consistent w ith an affirm ative an sw er to the a rticle’s question: U.S. m erch andise exports have been u n d errep o rted since the mid-1970s. * * * From the early 1930s through the late 1970s, p er capita in com es rose faster in lo w -in co m e states than in h igh -in co m e states, resulting in a sub stantial red u ction in state p e r capita in com e inequality. This long-stan din g tren d has since reversed: since 1978, the interstate in eq u a lity o f p e r capita in com es has risen in all but o n e year. In this issu e’s secon d article, “W h y Have State Per Capita In com es D i verged R ecently?” Cletus C. C ou gh lin and T h om as B. M an delbau m fin d that the recen t increase in in com e in equ ality stem s from the rapid g row th in 10 h igh -in com e A tlantic Coast states alon g w ith relatively s lo w grow th in 10 lo w -in co m e states scattered throu ghou t the n a tio n ’s interior. A fter con siderin g several explanations o f regional grow th, in clu d in g the m o ve m ent o f industrial activity from the Frost Belt to the Sun Belt and the “ farm crisis,” the authors co n clu d e that d eclin in g en ergy prices du rin g the 1980s w as the p rim a iy con tribu tor to the rising in equ ality o f state p er capita in com e. * * * Explaining the beh avior o f interest rates has been a long-stan din g p r e o c cu pation fo r m an y econom ists. A lthou gh eco n o m ic th eory suggests that the relationship betw een short- and lon g-term interest rates is sim ple, em pirical research, alm ost u niform ly, has rejected it. In the third article o f this issue, M ich ael T. B elongia and K ees G. K oedijk re-exam ine a basic m o d el o f interest rate determ in ation b y co n sid erin g the effects o f several w ell-k n ow n p o lic y changes. In th eir “ T estin g the Expectations M o d e l o f the T erm Structure: Som e C on jectu res o n the Effects o f Institu tional Changes," B elongia and K oedijk co n sid er h o w changes in the Federal R eserve’s im p lem en ta tio n o f m o n e tary p o lic y and the op era tion o f the Eu ropean M on eta ry System m a y have affected interest rate behavior. U sing data fo r five countries, the authors SEPTEMBER/OCTOBER 1988 2 fin d that forw ard rates are not related o n e-to-on e w ith changes in actual three-m on th interest rates in the U nited States, G erm any and Switzerland. Even though con sid erin g the operatin g p roced u res o f the Fed and the EMS led to som e im p rovem en t in the results, the persistent m o d el re jec tions leave beh in d m an y u nsolved pu zzles. * * * In the 1980s, the relationships b etw een the g ro w th o f the m on etary base and the grow th o f the m ajor m on etary aggregates, M l, M2 and M3, has ch anged dram atically. Historically, M l had gro w n m o re s lo w ly than the m o n eta iy base, w h ile M2 and M3 had grow th about 2 to 3 percen tage points faster. Beginning in early 1984, how ever, M l began to g ro w m uch faster than the m o n eta iy base, w h ile M2 and M3 g re w m ore slo w ly than the base. In the final article in this Review, “T h e P u zzlin g G row th o f the M o n e ta iy A ggregates in the 1980s,” Albert E. Burger presents a fram ew ork o f analysis that h elps unravel this mystery. Burger derives m u ltipliers that link the m o n e ta iy base to M l, M2 and M3. T h e co m p o n en t ratios o f these m u ltipliers su m m arize the key asset po rtfo lio decision s m ade by the pu blic that affect the grow th o f the aggre gates. Th e author presents the recen t beh avior o f these ratios in a h istori cal context to em p h a size the dram atic nature o f the changes that occu rred in the 1980s. He traces the acceleration in M l relative to base grow th to the sharp rise in the p u b lic’s h oldin gs o f checkable dep osits relative to its h oldin gs o f currency, an unusual historical d evelop m en t. Burger also show s that M2 and M3, in addition to bein g affected by this developm en t, also have been affected b y a rise in the p u b lic’s h oldin gs o f checkable d e posits relative to the oth er financial assets that co m p o se these aggregates. T h e key develop m en ts associated w ith this ch an ged beh avior o f the co m p o n en t ratios — the financial innovations in the 1970s and 1980s, and the sh aip red u ction o f inflation and interest rates in the 1980s — are also discussed. T h e au thor show s that, although the relationships b etw een the grow th o f the m o n e ta iy base and the M l, M2 and M3 aggregates ch anged significantly in the 1980s, the grow th o f these m on etary aggregates rem ains tied to the grow th o f the m o n eta iy base. FEDERAL RESERVE BANK OF ST. LOUIS 3 Mack Ott Mack Ott is a senior economist at the Federal Reserve Bank of St. Louis. The author acknowledges the helpful suggestions and criticisms of Andrew Kamarck, Tom Mayer, Guy Meredith, Gordon Midgely, Marius van Nieuwkerk, Lois Stekler, A.C.J. Stokman, Bill Witte, Geoffrey Wood, Stephen Wright and seminar participants at Notre Dame and Appalachian State Universities. Nancy D. Juen and Rosemarie V. Mueller provided research assistance. Have U.S. Exports Been Larger Than Reported? J . N LATE 1987, Ihe U.S. C o m m erce D ep a rlm en l a n n ou n ced that in its m o n th ly trade reports, ex ports to Canada w o u ld h en ceforth use Canadian custom s data on im ports from the U nited States rather than U.S. export data. T h e rationale fo r this proced u re is the d o cu m en ted in accu racy since 1970 o f U.S. cu stom s data fo r exports to Canada. T h e discrepan cies betw een the U.S. and Canadian data have b ecom e substantial both in absolute term s — n early $11 billion in 1986 — and in term s o f th eir effect on the U.S. trade balance — a 42 p ercen t red u ction in the 1986 U.S. trade deficit w ith Canada. W hite these errors are co rrected in the annuai recon cilia tion o f U.S-Canadian trade data, th eir persisten ce raises a b roa d er question: A re U.S. exports to o th er cou ntries sim ilarly u nderstated? This possibility raises som e im portan t political and eco n o m ic issues. In recen t years, the trade balance has been the focus o f m u ch eco n o m ic p o lic y debate, rivaling o r co m p le m en tin g such traditional d om estic issues as em ploym en t, in flation and grow th . In this context, isolatin g large understatem ents in U.S. m erch an dise export data is clearly a top ic w ith im portan t p o licy im plications. In this article, the relationship b etw een export u n d errep ortin g and the statistical d iscrep a n cy in the balance o f paym ents, w h ic h also rose from in sign ifican ce to p ro m in en ce du rin g the 1970s, is d evelo p ed and is used to assess the validity o f estim ated U.S. export u n d errep ortin g in the 1970s and 1980s. BALANCE OF PAYMENTS ACCOUNTING, REPORTING ERRORS AND THE STATISTICAL DISCREPANCY T h e first p ostw ar U.S. trade deficit d id not occu r until 1971, a qu arter o f a cen tu ry a fte rW o rld W ar II. D uring the early 1970s, the U.S. m erch andise trade account alternated b etw een deficits and surpluses; despite the com paratively w eak grow th o f U.S. m erch andise exports relative to im ports, h ow ever, the d eclin in g U.S. current accou nt b al ance rem ain ed in surplus du rin g m ost years until 1982, prim arily becau se o f strong in com e from U.S. foreign investm ents. A lo n g w ith the d eclin in g current accou nt bal ance, a p ersisten tly large d iscrep a n cy arose b e tw een the current a n d capital accou nt balances. Since the first OPEC em bargo in 1973-74, this dis- SEPTEMBER/OCTOBER 1988 4 crepan cy has averaged nearly $22 billion.' Before 1975, it h ad been g en erally small and negative, averaging —$1.1 billion from 1960 to 1974. T h e relation b etw een the current accou nt balance, errors in exports and the statistical d iscrepan cy can be illustrated by review in g balance o f p a y m ents accou nting.2 The Rudiments o f Balance o f Payments Accounting Balance o f paym en ts accou n tin g is structured by tw o basic prin cip les: dou ble-en try accou ntin g and equ ality b etw een net sales m inus gifts and the change in financial claim s. Balance o f paym ents accounts record a cou ntry's sales (exports) and purchases (im ports) o f go o d s and services plus transfers to foreign ers as w e ll as its len d in g to (capital exports) and b o rrow in g from (capital im ports) o th er countries. T h e sum o f go o d s and ser vices pu rchased and sold to foreigners, minus transfers, in a given p erio d is called the current account balance; the con com itan t ch ange during the sam e p erio d in the co u n try’s financial position du e to capital ou tflow s and in flow s is ca lled its capital account balance. Oftentim es, discussion focu ses on bilateral balances — fo r exam ple, b e tw een the U nited States and Japan; h ow ever, cou ntries gen era lly have surpluses w ith som e countries and deficits w ith others, and the overall balance w ith all cou ntries is the m ost inform ative m easure o f a cou ntry's international eco n om ic con d ition . A n illustration o f these p rin cip les in a three-cou n try exa m p le w ill high ligh t the offsetting equ ality o f the current and capital accou nt bal ances assuming they are completely and accurately measured. An Illustration o f Balance o f Payments Accounting Suppose that total w o r ld m erch an dise trade du rin g a qu arter con sisted o f a $1 m illio n c o m p u ter s old b y the U nited States to Japan and $300,000 w o rth o f crystal im p o rte d by the U nited States from Ireland, each paid fo r w ith short-term 'Throughout this article, the statistical discrepancy reported will be the “total discrepancy” — that is, the statistical discrepancy as it would be without the reconciliation adjustment for unre ported trade with Canada. 2For a more detailed discussion of balance of payments account ing, see chapter 15, “The Balance of Payments and Foreign Exchange Rate," in Caves and Jones (1981). For an application of these principles to the U.S. trade deficit, see Chrystal and Wood (1988). FEDERAL RESERVE BANK OF ST. LOUIS notes. T h ese IOUs are capital im ports (inflow s) o f the borrow ers and capital exports (outflow s) o f the lenders. Suppose also that a corp ora tion in Ire land, o w n e d b y U.S. residents, h ad profits du rin g the p erio d o f $80,000, $50,000 o f w h ic h rem ained w ith the subsidiary as retain ed earnings and $30,000 o f w h ic h w ere paid to the U.S. ow n ers out o f the firm ’s dep osits in a U.S. bank. T h e profits o f the Irish firm, in effect, are the paym ent fo r the use o f m achines, bu ildin gs an d financial resources that the U.S. o w n ers have sent to Irelan d — capital services ex p o rted b y the U n ited States to Ireland. T h e balance o f paym en ts fo r each o f the three countries du rin g the qu arter is sh ow n in figure 1. Som e Accounting Principles. T h e figure d is plays the transactions b etw een the three countries in the T-accou nts in the u p p e r panel. Every trans action is en tered tw ice, u sually as a debit and a credit but also in a variety o f o th er ways, d e p e n d ing on the transaction. For exam ple, fo r the U.S. o w n e d Irish firm ’s transactions, an $80,000 debit fo r capital services im p orted, a m inus $30,000 debit fo r U.S. bank deposits draw n dow n, and a plus $50,000 credit fo r the reinvested retained earnings are the entries in the Irish accounts, w h ile the opposite, b alancing entries a p p ea r in the U.S. accounts. N ote that debits (left-hand side o f T-accou nt) are en tered w ith negative signs in the balance o f paym ents (lo w e r panel), w h ile credits (right-hand side o f T-accou nts) are en tered w ith positive signs. For exam ple, the co m p u ter ex p o rted b y the U nited States to Japan appears as a credit (export) in the U.S. current accou nt and a debit (im port) in the Japanese current account. In contrast, in the capital account, capital ou tflow s (exports) a p p ea r w ith a negative sign w h ile capital in flow s (im ports) a ppear w ith a p ositive sign. Thus, the Japanese n ote payin g fo r the co m p u ter appears as a deb it (capital export) in the U.S. ca p i tal accou nt and a credit (capital im port) in the Japanese capital account. The Balance o f Payments Identity. W h en the transactions fo r each cou n try are su m m ed up, the resulting statem ent is the balance o f paym ents SEPTEMBER/OCTOBER 1988 Figure 1 The Relation Between International Transactions and the Balance of Payments Transactions T-Accounts (-) United States $1,000,000 note 300,000 crystal 50,000 foreign investment (+ ) (-) $1,000,000 computer exported 300,000 note 80,000 services -3 0 ,0 0 0 U.S. deposits Ireland $300,000 note 80,000 capital services -3 0 ,0 0 0 (-) (+ ) $300,000 crystal exported 50,000 Corporate retained earnings of U.S. subsidiary $1,000,000 computer imported Japan (+ ) $1,000,000 note U.S. deposits Balance of Payments Accounts Irish balance of payments U.S. balance of payments Balance on current account - $1,080,000 300,000 $780,000 Balance on capital account Statistical discrepancy Exports Imports - Balance on current account $300,000 80,000 Exports Imports $220,000 Balance on current account -$1,050,000 270,000 -$7 80 ,0 00 $0 Increase assets Increase assets ( - ) in Irish abroad ( + ) in foreign in Ireland Balance on capital account Statistical discrepancy $0 - 1,000,000 -$1,000,000 Capital Account: Capital Account: Capital Account: Increase ( - ) in U.S. assets abroad Increase ( + ) in foreign assets in U.S. Current Account: Current Account: Current Account: Exports Imports Japanese balance of payments -$2 70 ,0 00 50,000 -$2 20 ,0 00 $0 Increase ( - ) in Japanese assets abroad Increase ( + ) in foreign assets in Japan Balance on capital account Statistical discrepancy $0 1,000,000 $1,000,000 $0 6 sh ow n in the lo w e r pan el o f figure 1. Since g o o d s and services exports (im ports) have positive (nega tive) signs in the current accou nt balance w h ile capital exports (im ports) have negative (positive) signs, the current accou nt balance (CAB) is equal and o p p o site in sign to the capital accou nt bal ance (KAB) fo r each country. Th is essential id e n tity o f balance o f paym ents accounting, (1) CAB + KAB = 0, m ust h o ld as lo n g as the international transac tions are p ro p e rly and c o m p le te ly recorded, as th ey are in figure 1. In o th er w ords, if there is a trade surplus, CAB > 0, there m ust b e a capital deficit (net capital outflow ) o f an equal absolute am ount, KAB = - CAB < 0, and vice versa. T h e co m m on sense o f this fu ndam ental iden tity is that if a cou n try sells m ore go o d s and services to foreigners than it buys from them , foreigners m ust balance this shortfall w ith real assets and financial claim s on them selves — equities, real property, bon d s and m on ey.3 C onsequently, the balance o f paym en ts statistical discrep a n cy for each cou n try in figure 1, a correction equal to the sum o f CAB and KAB w ith the o p p o site sign, is zero. In the exam ple in figure 1, the U nited States has an overall current accou nt surplus ($780,000), but it has a trade deficit w ith Ireland ($220,000) and a trade surplus w ith Japan ($1,000,000). If reportin g errors or o m ission s are m ade w ith any country, they w ill sh o w u p in eith er the statistical d isc rep ancy, the w o r ld current accou nt balance or both. T o see w hy, co n sid er w h at happen s w h e n re p o rt ing errors are m ade. crepan cy equal to the export u nderreporting, $100,000. Such errors can be la b eled relative er rors: th ey affect the current accou nt balance (e) or capital accou nt balance (k) relative to each o th er causing a statistical d iscrep a n cy o f equal m agn i tu de and o p p o site sign. Alternatively, som e errors affect b oth current and capital accounts. F or exam ple, su ppose the $1 m illion co m p u ter export w as co rrectly reported, but the $80,000 earnings o f the U.S. o w n e d firm in Ireland w e re not reported. As a result, the rise in U.S. claim s on Ireland ($50,000) also w o u ld be unrep orted in the U nited States as sh ow n in pan el b o f figure 2. In this case, the U.S. statistical d isc rep ancy w o u ld be $30,000 becau se o f the d o cu m en ted (bank reports) d eclin e in Irish -ow n ed U.S. assets; h ow ever, the o th er $50,000 o f the U.S. export u n derstatem ent w o u ld be offset so that the levels o f both current and capital balances are u nderstated b y the absolute am ou nt o f this error, $50,000. That is, the u n rep o rted $50,000 in retain ed earnings — u n rep o rted service in com e o n current accou nt — is m atch ed b y the u n rep o rted $50,000 reinvested in the firm — u n rep o rted capital o u tflo w o n ca p i tal account. T h ese offsetting errors, d en o te d b y a, can be called absolute errors since th ey change the absolute level o f both current and capital ac counts. T h e y d o not affect the relative levels o f the tw o accounts; thus, th ey have n o effect on the statistical discrepancy. T h e general relation o f the re p o rted balance o f paym ents data w ith the actual trade and financial transactions can then b e su m m a rized as follow s: (2) cAb = CAB + e + a (3) KAB = KAB + The Effects o f Errors in R ep orted Exports. In practice, the statistical discrepan cy typically is not zero; errors or o m ission s in the data result in a n o n zero discrepan cy. For exam ple, su ppose the U.S. ex p o rter h ad filed export d ocu m en ts listing the co m p u ter sale in correctly as $900,000 w h ile the earnings o f the Irish firm are correctly given as $80,000. If n o offsetting errors w ere m ade, the U.S. balance o f paym ents w o u ld be as sh ow n in figure 2, pan el a. In this case, there is a statistical d is 3This is, of course, the same rule which describes any voluntary exchange between two people. Any imbalance in the value of goods and services received over time is equal and opposite in sign to the net value of financial flows between them. Each person gives to the other a collection of goods, money and assets equal in value to what he receives. FEDERAL RESERVE BANK OF ST. LOUIS k - a w h ere the “ ” in dicates the re p orted data, e and k are relative errors in the re p o rted CAB and KAB, respectively, and a is an absolute error. T h e logic o f the accou n tin g con ven tion s requires that cAb + kAb + SD = 0, so the statistical d iscrep a n cy (SD) is d efin ed as the negative o f the sum o f the re p o rted balances, (4) SD = -[CAB + KAB], 7 Figure 2 Source of Statistical Discrepancy (a) Statistical Discrepancy: Underreported Exports without Offsetting Errors ( e = -$100,000) U.S. Balance of Payments Current account Exports Imports - $980,000 300,000 Balance on current account $680,000 Capital account Change in U.S. assets abroad Change in foreign assets in U.S. -$1,040,000 260,000 Balance on capital account -7 8 0,00 0 Statistical discrepancy $100,000 (b) Statistical Discrepancy: Underreported Exports with partly Offsetting Errors (a = - $50,000, e. = - $30,000) U.S. Balance of Payments Current account Exports Imports - $1,000,000 300,000 Balance on current account $700,000 Capital account Change in U.S. assets abroad Change in foreign assets in U.S. Balance on capital account -7 3 0,00 0 Statistical discrepancy $30,000 From (2), (3) and (4), SD = - [CAB + e + a + KAB + k - a], so that, b y (1), SD is sim p ly the negative o f the sum o f the relative errors, e and k ; that is, (5) SD = - [ e + -$1,000,000 270,000 k ]. W h ile absolute errors (a) d o not affect any c o u n try's balance o f paym en ts discrepancy, such errors do sh o w up in the w o r ld balance o f paym ents totals. Panel a o f figure 3 sh ow s that, w ith n o re p o rtin g errors, the current accou nt balance o f the w o r ld is zero. T h e co m m on sense o f this is that fo r the total tradin g system, the surpluses o f the na tions w ith m o re exports than im p orts m ust b al ance the deficits o f the nations w ith less exports than im ports.4 Panel b o f figure 3 show s that w ith relative current accou nt errors (e), the U.S. export 4ln macroeconomic theory, this is referred to as Walras’ Law of Markets — the sum of trades (planned or actual) must be zero — with excess demands ( + ) and supplies ( - ) cancelling. See Patinkin, (1965) pp. 34-36. SEPTEMBER/OCTOBER 1988 8 Figure 3 The World Current Account and the World Current Account Discrepancy (a) No Reporting Errors U.S. current account Exports Imports - $1,080,000 300,000 U.S. CAB $780,000 Irish current account Exports Imports - $300,000 80,000 Irish CAB 220,000 Japanese current account $0 -1,000,000 Exports Imports Japanese CAB -1,000,000 World CAB $0 (b) Underreported Exports With Relative Errors (e = - $100,000) U.S. current account Exports Imports $980,000 -300,000 U.S. CAB $680,000 Irish current account Exports Imports - $300,000 80,000 Irish CAB 220,000 Japanese current account Exports Imports Japanese CAB World CAB FEDERAL RESERVE BANK OF ST. LOUIS $0 -1,000,000 -1,000,000 -$100,000 9 Figure 3 cont’d. (c) Underreported Exports with Absolute Errors (a = - $50,000) and Relative Errors (e = - $30,000) U.S. current account Exports Imports - $1,000,000 300,000 $700,000 U.S. CAB Irish current account Exports Imports - $300,000 80,000 Irish CAB 220,000 Japanese current account Exports Imports Japanese CAB World CAB - $0 1,000,000 -1,000,000 -$ 8 0,00 0 u n d errep o rtin g results in figure 2, pan el a in an equivalent deviation from the logica l w o r ld zero current accou nt balance. Finally, pan el c show s that both the absolute (a) and relative (e) errors — the u n rep o rted U.S.-owned Irish firm ’s $50,000 retain ed earnings in figure 2, pan el b and the $30,000 o f u n rep orted dividends — are reflected in the w o rld CAB even thou gh the U.S. SD show s on ly the relative ($30,000) error. Som e in d irect evid en ce on the w o r ld current accou nt d iscrep a n cy (see sh aded insert) im p lies that the U.S. current accou n t reflects b oth absolute (a) and relative (e) errors, a m ix illustrated in the distribution o f the profits o f the U.S.-owned Irish corp ora tion in figures 2 and 3.5 By its defin ition in id en tity 5, the U.S. balance o f paym ents statistical discrep a n cy reflects o n ly relative errors. Still, the in direct im p lica tion o f u n rep o rted U.S. investm ent 5ln testimony before the Joint Economic Committee, Heller (1984), p. 67, argued that such unreported investment earnings might be large enough to offset the reported CAB deficit: foreign source interest income. Her estimates suggest that unreported interest income was substantial during the early 1980s: There is some reason to believe that the bulk of the unrecorded transactions is due to an underrecording of receipts of service items such as reinvested earnings abroad, investment income and fees. Consequently, the U.S. current account deficit, if measured properly, is likely to have been substantially smaller than indicated by the officially reported data. Thus it is entirely possible that the U.S. was in substantial current account surplus in 1983. Stekler provides evidence that U.S. service exports are under stated because of unreported interest; she uses differences between the data on U.S. claims on foreigners from three nonTreasury sources and the U.S. Treasury International Capital Reporting System (TIC) to generate estimates of unreported In summary, in the three cases where data on U.S. claims on foreigners from the TIC reports can be compared with data from other sources it appears that the TIC data seriously understate U.S. claims. The size of the discrepancy between the data sources can only be roughly measured, but for example, a total on the order of $ 100 billion would not seem impossible. This would imply that U .S. interest receipts are underestimated by about $12 billion a year currently (assuming an average return of 12 percent). Stekler (1984), p. 7. SEPTEMBER/OCTOBER 1988 10 The W orld’s Current Account Discrepancy A n v e x p o rted go o d from the e o u n tiy o f origin is an im p orted go o d fo r the cou n try o f destin a tion. As a consequ ence, if the data are co m p lete and accurate, the w o rld can have n eith er a trade deficit n or surplus; it must have a balance (see figure 3). Yet, as sh ow n in the a cco m p a n y ing table, the w o rld trade data d o not y ie ld a balance on current account. discrepan cies is that substantial export in com e is not bein g reported; that is, exports o f services are understated. T h e data in the table d ocu m en t a w o rld cu r rent account deficit averaging $70.9 billion du r ing the early 1980s. T h is w o rld CAB discrepan cy can be accou n ted fo r by a negative service a c count balance, w ith u n rep o rted sh ip p in g in com e, u n reported direct investm ent in com e and u n reported p o rtfolio investm ent in com e the largest contributors. S hipp ing in com e is irrelevant for the U nited States; the IM F w ork in g party fou nd it attributable to "several e c o n o m ies w ith large m aritim e interests (notably those o f Greece, H ong K on g and Eastern Eu ro p e)."' T h e o th er tw o d iscrep a n cy items, direct T h rou gh ou t the first h alf o f the 1980s, w orld m erch andise trade w as in “ surplus,” substan tially so in 1980 and 1981, and n egligibly so since then. M ore b roadly throu ghou t the 1980s, the current account — the sum o f m erch andise and seivice trade m inus transfers — has been in substantial deficit w ith no clear trend tow ard balance. T h e im p lication o f these statistical Selected Balances of World Current Account Transactions (billions of U.S. dollars) 1980 1981 1982 1983 1984 1985 $31.1 -4 6 .8 -3 1 .9 - 2 .9 - 0 .5 $25.5 -7 5 .7 -3 5 .3 - 5 .3 0.8 -$ 0 .9 -8 8 .5 -3 4 .5 -3 .8 1.6 $3.0 -7 3 .4 -3 3 .2 - 2 .8 4.1 $9.7 -8 5 .1 -3 3 .3 - 2 .3 4.7 $7.8 -6 0 .6 -2 7 .4 0.2 5.3 13.0 11.0 2.0 8.4 10.0 21.0 - 9 .9 -1 3 .1 - 9 .0 -1 1 .6 -1 4 .9 -1 2 .5 - 7 .9 -1 8 .6 -3 1 .1 -2 8 .8 -3 8 .9 -4 6 .6 -1 1 .6 -1 5 .2 -1 2 .2 -1 4 .0 -1 2 .1 - 4 .5 4.9 6.5 0.1 5.2 - 1 .5 3.6 4.6 6.3 1.6 6.6 4.0 2.7 Current account (excluding official transfers) Official transfers - 9 .2 -1 9 .6 -4 5 .0 -1 7 .6 -8 5 .8 -1 7 .5 -6 4 .1 -1 5 .3 -6 8 .8 -1 6 .9 -5 0 .1 -1 5 .5 Current account (including official transfers) -2 8 .8 -6 2 .6 -1 0 3 .4 -7 9 .4 -8 5 .7 -6 5 .6 Merchandise trade balance Service balance Shipment Other transportation Travel Reinvested earnings on direct investment Other direct investment income Other (portfolio) investment income Other official transactions Other private transactions Private transfers SOURCE: International Monetary Fund, Report on the World Current Account Discrepancy, table 6. 'International Monetary Fund (1987), p. 3. FEDERAL RESERVE BANK OF ST. LOUIS 11 and p o rtfo lio investm ent in com e, w e re fo u n d to be attributable in large part to U.S. in vestors’ u n rep orted o r m isrep orted foreign in com e.T h ere are several co m m on elem en ts in these m a jor u n rep o rted service exports com p risin g the w o r ld CAB discrepancy. First, in each case, the im p o rte r has rep orted receivin g a service and payin g fo r it, but the cred ito r has not ac k n ow led ged the in com e receipt o r financial arrangem ent. Second, U.S. investm ents are e i ther d irectly im p lica ted by the evid en ce (direct investm ent) o r in d irectly im p lica ted by the size 2For a discussion of direct investment adjustments attributed to U.S. nonreported or misreported income, see pp. 35-39 of International Monetary Fund (1987). For portfolio invest ment adjustments, the U.S. role is more conjectural in that the working party was able only to pin down adjustments to industrial countries and others. Yet, it is plausible that the United States, as the largest holder of foreign securities in the year (1983) analyzed in detail — 27.8 percent of world earnings is that U.S. exports have b een u n d er stated du rin g the 1980s and that this u nderstate m ent is reflected partly (e) in the U.S. statistical discrepan cy. It is esp ecially n otew o rth y h o w large and persistent both the statistical d iscrep a n cy and the w o rld current account balance have b een since the mid-1970s. The U.S. Balance o f Payments Statistical Discrepancy: 1960—86 As chart 1 shows, the statistical d iscrep a n cy has b ec o m e quite large since the mid-1970s. T w o ve r sions o f the d iscrep a n cy are sh ow n in chart 1: the rep orted SD (SDHAT) and the total SD (SDTOT). SD TO T in clu des the discrep a n cy du e to U.S. u n d errep ortin g o f U.S. exports to Canada. SDHAT has been p u rged o f this error b y the annual re c o n ciliation agreed u p o n betw een the U.S. Census Bureau o f the C o m m erce D epartm en t and its Ca nadian counterpart, Statistics Canada. o f po rtfo lio earnings (seivice exports). Th ird, fo r both direct and po rtfo lio u n rep orted earnings, there w ill be both absolute (a) and relative er rors ( e ) in the U.S. balance o f paym ents: an un rep orted credit fo r service export and an u n re ported debit fo r the capital o u tflo w represen ted b y the u nrepatriated earnings (see figure 3, panel c). Thus, these u n rep o rted exports d o not affect the U.S. SD, but they illustrate that the w o rld current accou nt statistical discrep a n cy is prim arily the result o f u n d errep o rted exports and that U.S. firm s and individu als are in volved in u n d errep ortin g exports. cross-border bond holdings and 44 percent of cross-border equities (p. 68) — is a substantial nonreporter. See pp. 45-80, in particular tables 29 and 30 where unreported U.S. nonbank deposit interest is estimated at $7.7 billion; see also Stekler (1984). statistical d iscrep a n cy in id en tity 5, the exp ected value o f this su m m ation o f errors and om ission s in each y e a r w o u ld be zero, if such errors and om is sions were not systematic. Thus, o ver several years' observations, the m ean o f the statistical d isc rep ancy w o u ld ten d to be close to zero. A bsent sys tem atic errors, a d eclin e in the data’s reliability m ight cause w id e r fluctuations in the SD; persist ent positive SDs since the mid-1970s, how ever, suggest system atic errors. The Source o f the Statistical Discrepancy: Capital or Current A ccou n t E rrors? By its defin ition in id en tity 5, the statistical dis crepan cy must be du e to eith er relative overstate m ent (e) o f the current accou nt deficit o r relative T h e p ersisten ce o f large positive values o f the statistical d iscrep a n cy from 1975 on w a rd suggests that there are n on -ra n d om errors in the U.S. bal u n derstatem ent ( k ) o f the capital accou nt surplus. If capital accou nt errors are responsib le fo r the SD, capital in flow s m ust have b een persisten tly u n d er stated: as equ ation 4 shows, the capital surplus w o u ld have to be in creased in o rd e r to drive SD to ance o f paym ents data. From the defin ition o f the zero .6 6From a strictly logical point of view, there is also the possibility of overstatement of U.S. gross capital outflows — that is, an exag geration of U.S. investment abroad; however, there is neither empirical evidence nor a priori behavioral foundation for its occurrence. SEPTEMBER/OCTOBER 1988 12 Chart 1 U.S. Balance of Payments Statistical Discrepancies, Unreconciled and Adjusted Billions of dollars 40 Billions of dollars 40 Annual Data -2 0 -2 0 1960 62 64 82 84 1986 NOTE: The reported statistical discrepancy, SDHAT, reflects the U.S.-Canadian m erchandise trade reconciliation; the unreconciled statistical discrepancy, SDTOT, is the statistical discrep ancy as it would be w ithout the U.S.-Canadian reconciliation. A lthou gh m ost observers argue that capital ac count understatem ents are to blam e fo r the SD’s large deviations, this h ypoth esis is im plausible from a behavioral standpoint.7 Capital in flow s prim arily represent increases in debt fo r U.S. firms and individuals, and th ey have strong incentives to report them since the interest paym en ts to ser vice these debts are tax-dedu ctible. This su p p osi tion has b een su p p o rted b y the IM F W orking G rou p’s study, The World Current Account Dis- 7The Department of Commerce intimates that the statistical discrepancy is likely to be relative capital account errors («): “ If one assumes that a large part of cumulative net unrecorded inflows of about $140 billion from 1979 through 1984 was accounted for by capital inflows, foreign assets would have been understated by that am ount. . . ” Jack Bame, quoted in Scholl (1984), p. 26. Stekler (1983), p. 3, observes that “When the Interagency Work Group on the Statistical Discrepancy was set up in mid-1980, it was assumed that the bulk of the huge positive statistical discrepancy in 1979 and 1980 was ac counted for by unrecorded capital inflows.” Amuzegar (1988), p. 18, a former IMF Executive Director, reinforces this: “ . . . capital inflows into the United States are probably under recorded.” Pluckhahn (1988) reports that Commerce officials still downplay the notion of current account errors explaining the discrepancy: “ More likely, they say, capital flow statistics — measuring international financial transactions — have not kept up with the ongoing deregulation of financial markets.” That SD has been KAB error is also assumed in textbook discussions, such as Krugman and Obstfeld (1988), p. 299, and empirical applications of the balance of payments data; for example, see Hooper and Morton (1982), p. 45: “The sum of the current account plus official intervention purchases of domestic cur rency (I) define net private capital flow s. . . ” [italics added] FEDERAL RESERVE BANK OF ST. LOUIS 13 crepancy, and b y the Internal Revenue Service (1979) stu dy o f U.S. d om estic u n rep o rted in com e. T h e W orking Group fou n d that borrow ers w o r ld w id e d o consistently report international capital inflow s, w h ile len ders have been fou nd consist en tly to u n d erreport th eir capital exports: The main result o f analyzing the gaps in portfolio investment income reporting is that the discrep ancy results mainly from the understatement of receipts by the private nonbank sector and that this deficiency is widespread across countries." U n rep orted capital in flow s are the requisite explan ation if the U.S. SD is du e to capital account relative errors (K);yet, debt in crem en ts have b een fou n d to be d ep en d a b ly reported. U nreported capital in flow s w o u ld be inconsistent w ith both w o r ld w id e findings and the debtors' taxm in im izin g incentives to report such debt in cre ments. If anything, the IM F fin d in g suggests that the capital accou nt m a y be overstated because som e capital ou tflow s associated w ith reinvested earnings m a y be u n reported.9 Conversely, if U.S. m erch andise exports can be sh ow n to be u n derstated g en erally — as th ey have been in the specific case o f Canada — then u n d er statem ent o f the CAB is a plausible culprit. T h ere are three behavioral fou ndations fo r U.S. export understatem ent. First, is sim ple n egligen ce o r the 8lnternational Monetary Fund (1987), p. 78. Consistent with these IMF findings indirectly implicating U.S. investors, Stekler (1984), p. 3, observes that: Some have argued that since the United States accounts for about 20 percent of world services exports, that the United States probably accounts for the sam e share of the global services discrepancy ($15 billion in 1982). 9Note that in the 1980s, while the world current account discrep ancy has been a substantial deficit, the world merchandise discrepancy has been slightly in surplus; see table a in the shaded insert. The world current account discrepancy and the large U.S. holding of foreign assets creates a presumption that U.S. service exports are understated. By itself, this provides a counter argument to the claim that unreported capital inflows are the explanation for the statistical discrepancy. In contrast, the absence of a worldwide merchandise export understate ment does not in and of itself imply anything about errors in U.S. merchandise exports data. 10The first explanation is documented by the Commerce Depart ment and is one of the reasons implied for the late 1960s episode of export underreporting in the United Kingdom. See “ Under-recording of exports” (1969). The second has been substantiated by the IMF Working Party Report on the World Current Account Discrepancy, by the IRS (1979) study of unre ported U.S. income, in the OECD study by Veil (1982) and in Stekler (1983). The third conjecture receives a variety of sup porting argument in terms of costs and competitive disadvan tage imposed on U.S. producers in the National Academy of Sciences (1987) study of U.S. export controls. "F or example, the cover page of the U.S. Department of Com merce release, “ Summary of U.S. Export and Import Merchan dise Trade” for March 1987 described the discrepancy in costs o f reporting, esp ecia lly if the pen alties for n on rep ortin g are small. Second, sellers have an in cen tive to u n d errep o rt sales because, if u n d e tected, it redu ces th eir taxable in com e. Th ird, the U nited States im p oses restrictions on about 40 percen t o f U.S.-manufactured m erch andise ex ports; to avoid outright export proh ibition s or red u ce the h igh er costs im p o se d on foreign bu y ers o f U.S. m ach in ery b y such restrictions, som e u n rep orted sales are likely.10 U.S. MERCHANDISE EXPORTS: THE COMPARATIVE RELIABILITY OF U.S. EXPORT DATA VS. COUNTRY-OF-DE STIXATION IM PO R T DATA In principle, as illustrated in the balance o f p a y m ents figures 1-3, U.S. exports co u ld be m easured b y U.S. data o r cou n tiy-of-d estin a tion im p ort data. Yet, begin n in g in 1970, the U.S. C o m m erce D epart m en t has d o cu m en ted a p ersisten t u nd erstate m en t o f U.S. exports to Canada. R eferred to as “ u n d ocu m en ted exports,” the extent o f this p ro b lem is revealed in the annual recon ciliation o f U.S. and Canadian trade data throu gh com parison s o f U.S. export and Canadian im p ort data." export reporting as follows: The annual trade data reconciliation study with Canada (sched uled for release in June) indicates a substantial and growing undercount of exports from the United States to C anada in 1986, amounting to approximately 20 percent. This is due primarily to the non-filing of export documents with the U.S. Customs Ser vice. A number of joint U .S./Canadian efforts are underway to address this issue (informational mailings, bilateral collection of export documents, data exchange, etc.). The annual reconcilia tion studies also confirm that import data are more accurate than export data. See also Daily Report for Executives for August 5,1987. Such discrepancies are not unprecedented — see below, table 2 and footnotes 21, 22, 24 and 25. More generally, smuggling is a topic of longstanding interest to economists, both theoretically and empirically — see Bhagwati (1974). In industrial countries, the United Kingdom documented a pervasive period of export understatement in the late 1960s, amounting to about 3 per cent of exports and, more significantly, as high as 58.2 percent of the reported trade balance in 1966. See “ Underrecording of Exports" (1969), p. 667. While greatly reduced from the trou blesome levels of the late 1960s, export underreporting in the United Kingdom continues and is accommodated in the na tional income accounts by a 1 percent allowance in exports in the CIF/FAS conversion procedure (private correspondence, Stephen Wright, Bank of England). There is also evidence that the Canadian export data are subject to similar lapses: During 1978-79, a refinery in New Brunswick did not file customs reports on exports to the United States; this resulted in a $700 million understatement in petroleum products exported by ship to the United States. See Rose (1979). SEPTEMBER/OCTOBER 1988 14 Table 1 U.S.-Canadian Merchandise Trade, 1980-86 (billions of dollars) Northbound Trade’ U.S. exports 1980 1981 1982 1983 1984 1985 1986 $35.4 39.6 33.7 38.2 46.5 47.3 45.3 Southbound Trade’ U.S.-Canadian Trade Balances2 Undocumented3 Canadian imports (FAS) U.S. imports4 (FAS) Canadian exports U.S. compiled Canadian compiled Reconciled $4.9 5.0 4.2 5.1 5.3 6.0 10.2 $41.2 45.2 38.5 44.2 53.0 54.6 56.1 $41.2 45.9 45.9 51.5 65.6 68.1 67.3 $41.1 46.5 46.5 53.8 66.3 68.3 67.2 -$ 6 .1 - 6 .9 -1 2 .8 -1 3 .9 -2 0 .0 -2 1 .7 -2 2 .9 $0.2 - 1 .2 - 7 .9 - 9 .9 -1 2 .4 -1 3 .6 -1 1 .4 -$ 1 .4 -2 .8 - 9 .7 -1 1 .7 -1 5 .4 -1 5 .7 -1 3 .3 'Reported exports and imports from IMF Directions of Trade Statistics Yearbook, 1987. 2U.S.-Canadian trade balances from U.S. Bureau of Census, Department of Commerce, "Reconciliation of Canada-United States Merchandise Trade, 1986." Undocumented exports from U.S. Department of Commerce (1987b), table 14. “U.S. FAS imports estimated from CIF data, adjusted using 2.0 percent CIF/FAS margin; this choice is based on a comparison of FAS and CIF Canadian import data in the 1980s; see footnote 14. T h e persisten t u n derstatem ent o f U.S. exports to Canada and the resulting overstatem ent o f the U.S. bilateral trade deficit w ith Canada in the 1980s is sh ow n in table 1. T h e first five colu m n s in the b o d y o f the table sh ow the n orth bou n d trade (U.S. exports/Canadian im ports) and sou th bou nd trade (U.S. im ports/Canadian exports) as re co rd ed by each o f the co u n tries’ cu stom s authorities, and th eir re co n ciled estim ate o f u n d ocu m en ted U.S. exports. W h ile the sou th bou nd trade evinces no substantive disparities b etw een the U.S. and Cana dian data, the n orth b ou n d trade data exhibit d if feren ces ranging from 14 p ercen t to 24 p ercen t o f the U.S. export figures. As the u n d ocu m en ted ex ports colu m n shows, m ost o f this d iscrep a n cy has been ack n ow led ged b y the U.S. authorities as an u nderstatem ent o f exports. T h e sum o f the c o m p iled and u n d ocu m en ted U.S. exports a p p ro x i m ate the Canadian im p ort data, in dicating that ports results in an u nderestim ate o f the U.S. trade balance — that is, an overstatem ent o f the trade deficit. T h e ack n ow led ged U.S. errors — U.S. ex ports — ran ged from 2 7 p ercen t to 80 percen t o f the U .S.-com piled bilateral deficit w ith Canada and from 4 p ercen t to 19 p ercen t o f the U.S.c o m p ile d total trade deficit w ith the w o r ld in the 1980s.12 In summary, the Canadian data are substan tially m ore accurate than the U.S. data as the re c o n ciled bilateral balance is far clo ser to the initial Canadian balance. M o re generally, these d o cu m en ted errors suggest that o th er country-ofdestin ation im p ort data m a y also offer a su perior alternative to U.S. export data. o f U.S. exports. Two Problems with Using Country-of-Destination Import Data to Estimate U.S. Exports T h e last three colu m n s o f the table sh o w the bilateral trade balances du rin g the 1980s as c o m T h ere are tw o basic p rob lem s w ith using cou ntry-of-destin ation im p ort data. First, m ost p iled b y each cou ntry a nd as re co n ciled during conferen ces b etw een th eir respective custom s authorities. O f course, the u nderstatem ent o f ex im p ort data are re p o rted CIF (Cost + Insurance the Canadian im p ort data are a far su perior gauge 12Computed from data in U.S. Department of Commerce (1987b), Table 14. FEDERAL RESERVE BANK OF ST. LOUIS + Freight), w h ile export data are rep orted FAS (Free A lon gsid e Ship) — that is, not in clu d in g in- 15 surance and freight charges.11T h ese CIF im p ort data m ust b e adju sted to approxim ate the FAS export data.14This adjustm ent has been the sub ject o f som e research w ith in con clu sive results.15 Second, there is the issue o f sm uggling, esp ecially in less-d evelo p ed o r nonindustrial countries, in w h ic h the o m itted im ports in the cou n tiy-ofdestin ation data cou ld w e ll ex ceed the om itted exports in the export data.16 Ch oosin g the CIF/FAS Margin. One solu tion to the first p rob lem is sim ply to ch oose a reasonable CIF/FAS m argin to convert CIF data to FAS data. That is, the adjustm ent sh ou ld make sense in light o f w h at is know n, at least anecdotally, about freigh t and insurance charges, but sh ou ld not bias statistical tests o f the export understatem ent hypothesis. T h e evid en ce suggests a true m argin fo r the industrial countries w e ll b e lo w the 10 p ercen t traditionally u sed b y the IM F in its Directions o f Trade Statistics (DOTS) data on bilateral m erch a n dise trade. For exam ple, the U.S. C om m erce D e partm ent reports that, fo r U.S. im ports, the average CIF-FAS m argin is 5.2 percen t; the Bank o f England estim ates 5.0 p ercen t fo r U.K. im ports; the Bank o f N eth erlan ds estim ates a 5.6 p ercen t CIF/FAS m ar gin fo r D utch im ports du rin g 1980-87; and Geraci "Another reporting valuation, FOB (Free On Board) is frequently used as a synonym for FAS as it will be here. Strictly, FAS and FOB differ by the amount of loading and cargo handling charges included in the latter. 14Of the 151 IMF member countries whose bilateral trading volumes are covered in the Directions of Trade Statistics, 15 countries report imports FAS: Australia, Bermuda, Canada, Dominican Republic, Mexico, Papua New Guinea, Paraguay, Peru, Poland, Romania, Solomon Islands, South Africa, Vene zuela, Zambia, Zimbabwe. Moreover, the IMF’s annual IFS Yearbook reports CIF/FAS margins for each of the member countries; however, these margins are multilateral and cannot be used to isolate the appropriate margin on imports from the United States. 15Since insurance and freight are services, they should not appear in the merchandise trade account; moreover, these services may be rendered by a domestic or a foreign seller. Thus, they must be removed from the import data in order to make valid comparisons. See Geraci and Prewo (1977) and Yeats (1978). 16For an important collection of theoretical and empirical papers on this issue, see Bhagwati (1974). 17The U.S. CIF/FAS margin was published in Daily Report for Executives, No. 159, August 19, 1987, p.2. The U.K. margin was obtained by telephone from Gordon Midgely of the Bank of England and the Dutch estimate was supplied by M. van Nieuwkerk and A.C.J. Stokman of De Nederlandsche Bank in private correspondence. and P rew o (1977) fou n d a 5.2 p ercen t transport m argin fo r intra-European trade in 1970.17For the 15 cou ntries in DOTS (see fo otn ote 14) w h ic h re port both FAS and CIF im p ort data, the co m p u ted m argins fo r the 1980s range from 2.4 percen t for Canada to 20 p ercen t fo r Peru, S olom on Islands and Zambia. In general, these co m p u ted CIF/FAS margins w ere lo w e r fo r industrial than fo r n onindustrial cou ntries and fo r cou ntries w h o se trade is p re d om in a n tly w ith nearby tradin g p artners.18For exam ple, M exico, a non indu strial country, has a relatively lo w 4.6 p ercen t margin, w h ile Australia, an industrial country, has a m oderate, but h igh er 10.0 p ercen t m argin. M e x ic o ’s m argin is kept lo w by short transport lines w ith the U nited States from w h ich it obtains nearly tw o-th irds o f its re po rted im ports; Au stralia’s m argin is raised b y its relatively lo n g transport lines w ith N orth A m erica and E u rope from w h ich it obtains m ore than h alf its im ports. In light o f the re p o rted estim ates and the c o m pu ted CIF-FAS ratios, the em pirical tests in this article assume that the CIF/FAS m argin fo r in du s trial countries is 5.2 percent, the sam e as the aver age co m p u ted b y the C o m m erce D epartm ent for all U.S. im p orts.19 imports, 22.9 percent for Canadian imports and 18.3 percent for U.S. imports; however, their estimates were obtained from the ratio of CIF imports in country of destination to FAS exports in country of origin. If, as we argue here, exports are under stated, their approximation to the CIF/FAS margin will be biased upward. See Yeats (1978). ,9This margin also conforms with anecdotal evidence on current U.S. shipping charges and insurance rates for both transAtlantic and trans-Pacific routes. In fact, it is actually somewhat high relative to examples of transport and insurance rates for ocean-shipped containers quoted in the St. Louis area in April 1988: $1400-$1600 pier-to-pier, for a 40-foot container (2680 cubic feet) Los Angeles to Yokohama, Japan. Examples of products a 40-foot container could transport include $1 million worth of small sporting firearms or $80,000 worth of liqueurs. With insurance at $4 per $1000 of declared value, these exam ples would have CIF/FAS margins of 0.6 percent and 2.4 percent, respectively. (I am indebted to Jerry Kausch, Interna tional Import-Export Services, St. Louis, for these examples). Bulk grain shipping rates, conversely, bracket the traditional 10 percent margin. From U.S. Gulf of Mexico ports to Rotterdam, the Netherlands, large deep draft bulk carriers of up to 110,000 tons displacement charge $15/metric ton (April 1988) and insurance of 0.15 percent of value. This implies a 4.95 percent CIF/FAS margin for soybeans, 16.3 percent for corn and 12.2 percent for hard red winter wheat given their April 1988 prices per metric ton, $248, $92 and $123, respectively. (I am in debted to John Muller of Bunge Grain Co., St. Louis, for these examples). 18Both of these tendencies concur with the findings of Geraci and Prewo (1977); however, their point estimates (based on 1970 OECD data) are much higher: for example, 13.8 percent for UK SEPTEMBER/OCTOBER 1988 16 Table 2 Trade Discrepancies — Selected Areas and Country Imports from the World Compared with World Exports to Those Areas and Countries, 1980-86 (annual averages, billions of dollars)_______________________________________ World exports to Imports from the world by1 Discrepancy $522.4 $492.1 $30.3 95.5 11.8 12.0 7.4 18.6 6.4 7.4 27.0 11.5 80.9 8.6 9.1 9.1 13.3 1.7 6.5 24.5 16.7 14.6 3.2 2.9 - 1 .7 5.3 4.7 0.9 2.5 - 5 .2 15.3 27.1 24.2 -2 3 .0 28.5 73.4 12.2 9.3 -4 5 .2 Industrial - 202 1,240.9 1,260.6 -2 0 .7 -1 .7 Netherlands Switzerland Industrial-183 Industrial-174 United States 80.1 36.0 1,124.8 843.1 281.7 64.1 31.8 1,166.6 873.8 292.8 16.0 4.2 -4 1 .8 -3 0 .7 -11 .1 20.0 11.7 -3 .7 -3 .6 -3 .9 Nonindustrial-131 Western Hemisphere Egypt Greece Israel Mexico Panama Phillipines Singapore South Africa Discrepancy as percentage of world exports to 5.8% SOURCE: Data from Directions of Trade Statistics, Yearbook 1987, World exports and imports table. 'FAS imports estimated from CIF data using 10 percent CIF/FAS margin for nonindustrial countries and the IFS Yearbook CIF/FAS margin for industrial countries (see footnote 14). 2The 20 countries classified as industrial are Australia, Austria, Belgium-Luxembourg, Canada, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Spain, Sweden, Switzerland, the United Kingdom and the United States. (Note that Belgium and Luxembourg are counted as one country.) industrial countries less the Netherlands and Switzerland. 4lndustrial-18 less United States. Screening f o r Valid Im p o rt Data. T h e oth er em pirical p rob lem w ith using country-ofdestin ation im p ort data to estim ate U.S. exports is that the im p ort data m ay not b e valid. If all co u n tries' im p ort data w e re equ ally valid, then an esti Table 2 provid es a com parative assessm ent o f the validity or com p leten ess o f the im p ort data o f the nonindu strial and industrial countries. An im partial basis fo r evaluating the validity o f a m ate o f the w o r ld w id e U.S. export u nderstatem ent co u n try’s im p ort data is to com pare its o w n data cou ld be obtain ed easily from data on im ports co m p ilin g total im p orts from all o f the cou ntries in the w o rld w ith the sum o f the data c o m p ile d by from the U nited States fo r all 151 cou ntries in DOTS. T h e IM F classifies 20 o f these countries as “ in du strial” and the others as “ non indu strial.” 20 “ The 20 countries classified as industrial by the IMF in its DOTS are Australia, Austria, Belgium-Luxembourg, Canada, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Spain, Sweden, Switzer land, the United Kingdom and the United States. (Note that Belgium and Luxembourg are counted as one country.) FEDERAL RESERVE BANK OF ST. LOUIS the IM F o f all the in dividu al cou n tries’ exports to that country. Since cou ntries obtain revenues from 17 tariffs and p o lice quotas on politically sensitive im ports, a strong presu m ption exists that im port data shou ld be m ore co m p le te — as in the U.S.Canadian case — than export data. By this postu late, a country's trade data can be ju dged invalid if its rep orted FAS im ports are less than the sum o f w o rld exports to it. For exam ple, du rin g the 1980s, as sh ow n in table 2, the rep orted level o f w o rld exports to M exico ex ceed ed by 28.5 percen t the level o f FAS im ports from the w o rld rep orted by M exico.-' For G reece and the Phillipines, the c o r resp o n d in g shortfalls w ere 24.2 percen t and 12.2 percent, respectively, w h ile for Panam a it w as a w h o p p in g 73.4 percent. For n onindustrial co u n tries in the W estern H em isphere, the u nderstate m ent w as 15.3 percent, w h ile for all 131 n on in d u s trial countries, it averaged 5.8 percent. Such u n d errep ortin g o f im ports in d evelop in g nations has been w id e ly d o cu m en ted in the trade litera ture and often used as a m easure o f sm uggling in d u ced b y tariff a vo id a n ce.-Th ese illustrations are not isolated; they reflect gen erally the characteristics o f the nonindustrial co u n tries’ data. A m ore system atic analysis re jected all but 6 o f the 131 nonindustrial cou n tries’ im p ort data.-1Given these problem s, such data are not useful in testing the relationship b etw een U.S. export u nderstatem ent and the U.S. SD. A p p ly in g the same criterion to the industrial cou n try data results in a general accep tan ce o f the validity o f the im p ort data fo r 18 o f the 20 co u n 2,The full discrepancy between the U.S. and Mexican data is further complicated by the U.S. Commerce Department’s rough estimate that exports to Mexico are underreported by about 10 percent. (I am indebted to Gerald Kotwas, Assistant Chief Foreign Trade Division of Census Bureau, U.S. Department of Commerce, for this estimate.) “ See Bhagwati (1974), especially Part III — “ Partner-Country Data Comparisons and Faked Invoicing.” Sometimes, the errors are positive: Probably resulting from ineffective embar goes, the level of imports from the world by South Africa has exceeded acknowledged world exports by an average of 33.7 percent during the 1980s. Similarly, the level of Israeli imports has exceeded acknowledged world exports to Israel by 22.6 percent during the 1980s. 23The general testing of the nonindustrial countries was accom plished using a three-part screen: tries. O nly the data o f the N eth erlan ds and S w itz erland are rejected (discrepan cies statistically significant at 1 percen t level). Excluding these tw o cou ntries m ore than dou bles the average p ercen t age discrep a n cy b etw een im p orts from the w o rld and w o rld exports to the industrial countries from — 1.7 percen t to — 3.7 percent. T h ese tw o co u n tries have a lon g tradition o f re-exp ortin g im p orted goods, referred to as “ m erch an tin g” in the Dutch data; re-exp orted go o d s are o m itted from their im por t data. Consequently, w o rld exports to them exceed their re co rd ed net im p orts by substantial amounts, as the table shows.-4 Th e exclu sion o f re-exp orted g o o d s suggests that som e U.S. exports m ay sim ply be u n record ed anyw here. That is, if a U.S. sh ipm ent to the N eth er lands that is re-exp orted by a D utch m erchant to France is not rep orted as a N e th erlan d s’ im port from the U nited States, but is m easured solely as a D utch export to France, foreign im p ort data u n derstate U.S. exports. T h e om ission o f the re exp orted go o d s w o u ld cause the im port-based estim ate o f U.S. exports to be understated; h o w ever, it w o u ld not cause errors in the tw o co u n tries’ o w n international data.2’ Given the evid en ce o f inaccurate im p ort data illustrated in table 2, the estim ates o f the U.S. ex port u nderstatem ent and tests o f its h yp o th esized relationship to the U.S. balance o f paym ents d is crepan cy em p lo y a data set that in clu des 17 o f the industrial countries: o n ly the Netherlands, Switz- 24Net imports are imports less re-exported goods. The Nether lands, for example, does not count a landed shipment of mer chandise as a Dutch import if it neither a) changes title to a Dutch resident, nor b) crosses the border (i.e. — passes through customs). Hence, goods landed in the Netherlands and reexported apparently have been counted by exporting countries as an export to the Netherlands; however, according to the Bank of the Netherlands, which compiles the Dutch trade data, the Netherlands has not counted them as an import. 25ln principle, since the Netherlands and Switzerland report net exports as well as net imports, the omission of U.S. exports to any of them should be captured in their exports to other coun tries being similarly understated relative to the importing coun try’s data; that is, the sum of the two discrepancies should be approximately zero. This offsetting does occur in the data for Switzerland but not for the Netherlands trade data (billions of dollars) 1980-86 averages: (1) Availability of data on imports from the United States in each year, 1 9 6 0 -8 6 ; (2) Substantial trade volume with the United States (annual imports from the U.S. of at least $400 million 1 9 8 0 -8 6 ); and (3) Imports (FAS) reported from the world at least as large as reported world exports to the country. Only 6 of the IMF 131 nonindustrial countries passed this screen: Indonesia, Israel, Korea, South Africa, TrinidadTobago and Venezuela. These countries accounted for only about 20 percent of U.S. exports to nonjndustrial countries and about 7 percent of total U.S. exports in 1986. Discrepancy Discrepancy between world between world exports and imports and country imports country exports Netherlands Switzerland 16.00 4.20 1.55 -5.05 Sum 17.55 -0.85 SEPTEMBER/OCTOBER 1988 18 Table 3 U.S. Balance of Payments Statistical Discrepancies, Observed and Adjusted, 1960-86 (billions of dollars) 1960-86 Data Mean Standard error SDHAT SDTOT SDAI2 SDAINC3 $7.11 9.03 3.24 5.67 $2.32 2.69 1.80 2.28 1960-74 t-test' Mean 3.07** 3.36** 1,80 2.28* -$ 1 .4 3 -1 .0 7 -2 .8 4 -2 .5 0 1975-86 Standard error t-test1 Mean $0.64 0.58 0.75 0.69 2.25* 1.83 3.76** 3.64** $17.78 21.64 10.85 15.88 Standard error t-test1 $3.05 3.44 2.63 3.13 5.82** 6.28** 4.12** 5.08** 'Test of statistical significance of mean SD; ** indicates significance at 1 percent level and * indicates significance at 5 percent level. JSDT0T adjusted by U.S. export discrepancy with industrial countries other than the Netherlands and Switzerland. 3SDTOT adjusted by U.S. export discrepancy with industrial countries other than Canada, the Netherlands, and Switzerland. erland and, o f course, the U nited States are o m it ted. A d eta iled descrip tion and listing o f the data are con tain ed in the appen dix. TESTS OF THE UNDERSTATED U.S. EXPORT HYPOTHESIS Testin g the p rop o sitio n that U.S. m erch andise exports have been understated em ploys the d is crepan cy b etw een cou n tiy-of-destin ation im p ort data and U.S. export data to determ in e h o w much, if any, o f SDTOT can be accou n ted for by u n d er reportin g o f U.S. m erch an dise exports.211First, the cou ntrv-of-destination im p ort data are used (anal o gou sly to the C om m erce D epartm en t’s use o f Canadian im p ort data) to revise the U.S. balance o f paym ents statistical discrep a n cy data; the m ean o f the revised SD series is then tested fo r statistical significance. Second, regression analysis is u sed to test w h e th er the export adjustm ent variable signi ficantly explains the U.S. statistical discrepancy. “ Since underreported service exports, conjectured in Heller (1984) and documented in Stekler (1984), also form part of e in identity 5, a portion of SDs should depend on non-merchandise export errors. 27See the data appendix for a more detailed explanation of SDTOT. It may appear to be possible to test the relationship between the data on the U.S. statistical discrepancy either with or without the Canadian errors — SDTOT and SDHAT, respec tively — against corresponding data on the U.S. export under reporting (compiled from the IMF DOTS) with or without the Canadian component — XDI17 and XDINC, respectively. Yet, this cannot be accomplished consistently because the corres ponding data are not available. SDTOT contains the U.S. errors as compiled and, likewise, XDI17 contains the U.S.country-of-destination discrepancies as compiled; however, the adjustment RAUSCA to obtain SDHAT from SDTOT in identity 6 removes less than the total U.S.-Canadian export discrep ancy but also deletes some import discrepancies. This distinc tion can be seen in table 1 by comparing the column of undoc umented U.S. exports against the difference between the U.S. FEDERAL RESERVE BANK OF ST. LOUIS The Adjusted U.S. Balance o f Payments Statistical Discrepancy T h e U.S. balance o f paym en ts statistical d isc rep ancy, as re p orted in the U.S. balance o f paym ents data, SD, is net o f the U.S.-Canadian trade d iscrep ancy. T h e inclusive m easure o f the d iscrepan cy is the a ppropriate form to test its relationship to export u nderreporting, since n eith er U.S. data are adju sted n o r is any cou n try e x clu d ed a p r i o r i on the basis o f an assum ed relationship. Therefore, w e use SDTOT, the inclusive m easure as in chart 1, (6) SDTOT, = SDHAT, - RAUSCA,, w h e re RAUSCA, is the re co n ciled adju stm ent to the U.S.-Canadian m erch an dise trade balance.27 In o th er w ords, SDTOT, is the statistical d iscrepan cy that w o u ld exist if U.S. m erch an dise trade w ith Canada had b een co m p iled , unadjusted, in the and the reconciled bilateral trade balance. In each year, RAUSCA, the difference between the U.S. compiled and the reconciled trade balance, is a smaller adjustment than the undocumented exports. Moreover, as can also be seen in the table, the undocumented exports agreed upon between the two countries’ customs authorities do not incorporate the year’s full difference between the U.S. and the Canadian measures of northbound trade as obtained from the IMF DOTS. Conse quently, RAUSCA adjusts the statistical discrepancy in a fash ion that does not correspond with deleting the DOTS Canadian export discrepancy from the total 17-country DOTS U.S. export discrepancy. While the agreed-upon changes predominantly reflect northbound trade statistics, southbound trade (U.S. imports) data are also affected. Data separating RAUSCA into northbound and southbound changes are not available. None theless, there is a high correlation between RAUSCA and the bilateral U.S.-Canadian export discrepancy from DOTS during 1970-86; .943; moreover, a regression of SDTOT on XDINC, reported in table 4, has results similar to the regressions based on equation 7. 19 Chart 2 U.S. Balance of Payments Statistical Discrepancies, Total and Adjusted 1960 62 64 66 68 70 72 74 76 78 80 82 84 1986 NOTE: The adjusted statistical descrepancies are SDTOT less the estimated U.S. export discrepancy: SDAI is adjusted by the 17-country discrepancy; SDAINC is equal to SDAI with Canada omitted. sam e fashion as m erch an dise trade w ith o th er countries. Using the d iscrep a n cy in the U.S. exports to the industrial co u n tries’ (less the N eth erlan ds and Sw itzerland) XDI17,, an adju sted statistical dis crepancy, SDAI,, w as com p u ted : SDAI, = SDTOT, - XDI17,. SDAI and SDAINC are d isp layed in table 3 fo r the full p eriod 1960-86 and fo r the tw o subperiods, before and after 1975. T h e re p o rted d iscrep a n cy in the balance o f p a y ments, SDHAT, averaged about $7 b illion w h ile SD TO T averaged about $9 billion du rin g the 196086 period, both statistically significant; h ow ever, each w as com paratively sm all and negative du rin g 1960-74 and large and positive du rin g 1975-86. T h e industrial cou n try a dju sted SDs, SDAI and See the a p p en d ix fo r details. T o assess the p ossi bility that o n ly the U.S.-Canadian export d isc rep ancy is m ean in gfu l in the analysis o f SDTOT, a d ju sted SDs both w ith and w ith ou t the Canadian SDAINC, are sm aller but still substantial and sta tistically significant in both subperiods. As chart 2 shows, the industrial cou n try d iscrep a n cy (XDI17) d iscrep a n cy — SDAI and SDAINC, respectively, — accounts fo r about h a lf o f the total discrepan cy since 1975. Chart 2 also show s that the non- are co m p u ted and rep orted in table 3. T h e m ean and standard errors o f m eans fo r SDHAT, SDTOT, Canadian co m p o n en t o f the exp ort d iscrep a n cy is large and persistent. SEPTEMBER/OCTOBER 1988 20 Table 4 Regression Analyses of Total Statistical Discrepancy’s Relation to Industrial Countries Export Discrepancy Estimated Coefficients1 Specification i ii iii iv V5 vi6 Intercept (a) -4 .7 8 (2.85)** -4 .6 8 (2.78)** -1 .4 1 (0.74) -0 .6 2 (0.30) Dummy (b) 4.32 (0.98) -4 .6 1 (0.91) -1 .3 2 (1.13) 0.20 (0.07) Export discrepancy slope (c) Summary Statistics Export discrepancy dummy (d) 2.39 (11.10)** 2.04 (4.88)** 0.01 (0.02) -0 .2 5 (0.27) 2.16 (2.83)** 2.74 (2.74)* -0 .0 7 (0.13) -0 .7 7 (0.46) 2.24 (4.62)** 4.45 (3.18)** Hypotheses Tests2 R2 DW P Specification F-test Slope coefficients «1.0* t-test .82 2.42 - .2 2 N/A 6.46** .82 2.41 - .2 2 ii vs. i: 0.96 2.48* .86 3.05 -.5 3 iii vs. i: 7.99** 5.72** .86 3.18 - .5 9 .91 — iv vs. i: 4.38* iv vs. ii: 7.52* iv vs. iii: 0.82 N/A .77 2.13 — - .1 9 N/A 3.67** 9.36** 5.01** 'The letters under the coefficient-column headings refer to the coefficients in equation 6; absolute value of t-statistics appear in parentheses beneath estimated coefficients; * indicates significance at 5 percent level and ** indicates significance at 1 percent level. ^Indicates rejection at 5 percent level; ** indicates rejection at 1 percent level. 3Test of null hypothesis that added variables in unrestricted specification are zero. “One-tail test of null hypothesis that slope coefficient is less than or equal to 1.0. In i and ii, the test reported is for full period; in iii-vi, the test reported is for slope coefficient (c + d) for period 1975-86. Specification v is specification iii with corrected for serial correlated residuals, AUTOREG procedure in SAS. Specification vi is specification iii with the U.S.-Canadian export discrepancy removed from the independent variable; see footnote 25. Regression Analysis o f the Relation Between SD and XD T h e m ean SDs re p o rted in table 3 fo r each subp e rio d are each statistically significant, and the industrial country-based adjustm ent fails to redu ce SDTOT to a level in sign ifican tly different from zero. Consequently, the n on -zero m eans o f the adju sted SDs im p ly that o th er errors remain, in clu d in g u n d errep o rted service exports not in clu d ed in the DOTS m erch andise trade data as w e ll as u n rep o rted m erch andise exports to co u n tries not in clu d ed in XDU7. Thus, it is still u nclear that the U.S. m erch an dise export discrep a n cy is substantively related to the SDTOT. A direct w a y to test this hypothesis can be in ferred from iden tity 5. coefficien t if each o f three con d ition s are met: 1. the discrepancy is due entirely to CAB errors, e; 2. these errors arise totally from merchandise trade export omissions; and 3. U.S. errors in reported exports to nonindustrial and the three omitted industrial countries are negligible. A llo w in g fo r shifts in this relationship b etw een the tw o subperiods, 1960-74 and 1975-86, w e have (7) SDTOT, = a + bX, + c XDI17, + d\, XDI17, + in,* 0, t < 1975 1 1, t 5* 1975. Equation 7 provid es three tests o f the relation o f Id en tity 5 im p lies that a regression o f SD TO T on XDI17 sh ou ld have an in tercept not significantly SD TO T to XD. First, it perm its tests o f the re le vance o f the U.S.-industrial cou n try export d is differen t from z e ro and a positive, unitary slope crepan cy in the significance o f the coefficien ts c FEDERAL RESERVE BANK OF ST. LOUIS 21 and d oil XDI17: If u n reported LJ.S. exports o f m er ch andise to industrial countries have been the sole source o f SDTOT, c shou ld be statistically signi ficant and not significantly d ifferent from unity. On the o th er hand, if eith er u n reported LJ.S. service exports o r m erch andise exports to cou ntries not in clu ded in XDI17 also matter, then c (or c + d) shou ld be significantly larger than unitv. If XIJ117 is irrelevant to SDTOT, n either c n or d w ill be signi ficantly differen t from zero. Second, equ ation 7 p er mits testing fo r the differen ces in the tw o su b p e riods by m eans o f the du m m y variable \. Third, it p erm its a test o f om itted variables’ relevance in the significance test o f the intercept: If the in tercep t is not significantly different from zero, then eith er o m itted variables are h igh ly correlated w ith XDU7 o r they have zero means. Th e results o f the regres sion estim ates and these specification tests are rep orted in table 4. T h e estim ates o f specification s (i) — (iv) test the relevance o f the su bperiod du m m y A.. T h e F-tests fo r the three specifications w ith in tercep t o r slope du m m ies (ii, iii, iv) against the null h ypothesis o f no du m m ies (i) indicate that (iii), the specification w ith the slope dum m y, rejects the null h ypothesis and is n ot rejected b y the specification w ith both slope and in tercep t du m m ies (iv). Uniform ly, how ever, the strong form o f the h ypothesis — that is, o n ly the 17 industrial cou ntry m erch andise exports are re le vant and, consequently, that the coefficien t on XDI17 is 1.0 — is rejected by the t-test in the last colu m n o f the table. T w o additional specifications, v and vi, are also rep orted in table 4. Th e specification tests require the use o f the sam e data in the alternative specifica tions i, ii, iii, iv. Yet, th eir D urbin-W atson statistics in dicate that specifications iii and iv have negatively serially correlated residuals. Since this biases the estim ated standard errors o f their coefficients, a c o rrected estim ate o f the preferred specification iii, d esign ated as specification v, is also rep orted in table 4. A com parison o f v w ith iii show s only n egligible differences. Finally, specification vi is a regression o f SDTO T on the non-C anadian export discrepancy, XDI17NC. Th e significance o f the esti m ated coefficien t d refutes “ Regression tests parallel to those reported in table 4 were also run on a sample including the selected nonindustrial countries described in footnote 23. Tests of the explanatory power of the nonindustrial countries against the null specifications omitting them established that the sample of nonindustrial countries did not add explanatory power to specifications restricted to indus trial countries. the con ten tion that on ly the Canadian export d is crepan cy is related to SDTOT. T h ese test results dem on strate that the U.S. ex port d iscrep a n cy w ith the industrial cou ntries has a statistically significant relation w ith the balance o f paym ents discrepan cy; that is, the claim that U.S. m erch andise export u n d errep o rtin g is a cause o f the statistical d iscrep a n cy is not rejected. T h e in dustrial cou ntry m erch andise export d iscrepan cy is not the w h o le s to iy since the c oefficien t is greater than unity; how ever, the DOTS nonindu strial data are o f no avail in explain in g it.28 C onsistent w ith the IM F study fin din gs (see pp. 10-11), the leadin g can didate fo r addition to the m o d el seem s to be U.S. service exports.2'1 Finally, the coefficien ts on n eith er the intercept n or its du m m y variable are significantly different from zero in the preferred specification s (iii, v, vi). This suggests that if any variables have been o m it ted — for exam ple, service exports — th ey are eith er h igh ly correlated w ith the U.S.-industrial countries' m erch an dise export discrep a n cy or have a m ean o f zero. CONCLUSION U.S. m erch an dise exports have been u n d er rep orted d u rin g 1960-86, p rim arily du rin g 1975-86. This u n d erreportin g, m easu red by cou ntry-ofdestination m erch an dise im ports from the U nited States, parallels the export d iscrep a n cy d o c u m en ted by the U.S. C om m erce D epartm en t fo r U.S. exports to Canada since 1970. A n estim ated export correction based on industrial co u n tries’ im ports from the U nited States re d u ced the statistically significant U.S. balance o f paym en ts d iscrepan cy from $9 billion to $3.2 billion fo r 1960-86 and from $21.6 billion to $10.9 billion fo r the 1975-86 su bpe riod. M oreover, regression tests o f the industrialcou n try im p ort-b ased adjustm ent explain m ost o f the variation in SD TO T du rin g the last 12 years. Th ese results in dicate that U.S. exports o f m erch an dise and services have been larger than rep orted and, consequently, that U.S. m erch an dise and cu r rent accou nt deficits have been sm aller than re p o rted since the mid-1970s. ^See also Heller (1984) and Stekler (1984). SEPTEMBER/OCTOBER 1988 22 REFERENCES Amuzegar, Jahangir. “The U.S. External Debt in Perspective,’’ Finance and Development (June 1988), pp. 18-19. Bhagwati, Jagdish N., ed. Illegal Transactions in International Trade, Theory and Measurement, (North-Holland Publishing Company, 1974). Caves, Richard E., and Ronald W. Jones. World Trade and Payments, An Introduction, 3d ed. (Little, Brown and Com pany, 1981). Chrystal, K. Alec, and Geoffrey E. Wood. “Are Trade Deficits a Problem?” this Review (January/February 1988), pp. 3-11. Daily Report for Executives. “ Trade Balance: FAS/Customs Monthly Trade Gap Widened to $14.1 billion in June, Com merce Says,” No. 159, August 19, 1987, pp. N5-N7. Daily Report for Executives. “ Trade Balance: Commerce De partment Ready to Upgrade Monthly Reports on U.S. Trade Deficit, Official Says,” No. 149, August 5, 1987, pp. L2-L3. Geraci, Vincent J., and Wilfried Prewo. “ Bilateral Trade Flows and Transport Costs," Review of Economics and Statistics (February 1977), pp. 67-74. Heller, H. Robert. Statement of H. Robert Heller, Vice Presi dent for International Economics, Bank of America N.T. & S.A., in The Foreign Trade Dilemma: Fact and Fiction, Hearing before the Joint Economic Committee, 98 Cong. Sess. (GPO, 1984), pp. 48-70. Laney, Leroy O. “The Case of the World’s Missing Money," Economic Review, Federal Reserve Bank of Dallas, (January 1986), pp. 1-9. National Academy of Sciences. Balancing the National Interest, U.S. National Security Export Controls and Global Economic Competition, (National Academy Press, 1987). Patinkin, Don. Money, Interest and Prices, 2d ed. Row, 1965). (Harper & Pluckhahn, Charles W. “ Measurement Errors Could Distort Trade Gap Figures, Economist Says,” Investor's Daily Janu ary 18, 1988. Rose, Frederick. "Error in Canada Report of Trade Surplus Draws Fire From Economists, Politicians,” Wall Street Journal, November 26, 1979. Stekler, Lois. “The Statistical Discrepancy in the U.S. Interna tional Transactions Accounts,” Board of Governors of the Federal Reserve System, Mimeo, October 17, 1983. ----------------“The Statistical Discrepancy in the U.S. Interna tional Accounts: Are We Missing Interest Income Receipts?’’ Board of Governors of the Federal Reserve System, Mimeo, 1984. “ Under-recording of Exports,” Board of Trade Journal [United Kingdom] September 10, 1969, pp. 665-67. Hooper, Peter and John Morton. “ Fluctuations in the Dollar: A Model of Nominal and Real Exchange Rate Determination,” Journal of International Money and Finance, (April 1982), pp. 39-56. U.S. Department of Commerce, “ United States Foreign Trade: Summary of U.S. Exports and Import Merchandise Trade,” Release No. FT900-87-03, March, 1987a. International Monetary Fund, Report on the World Current Account Discrepancy, September 1987. “ United States Foreign Trade: Summary of U.S. Exports and Import Merchandise Trade,” Release No. FT900-87-09, September 1987b. Internal Revenue Service. “ Estimates of Income Unreported on Individual Income Tax Returns,” Department of the Trea sury, Publication 1104, September, 1979. Veil, Erwin. “The World Current Account Discrepancy,” OECD Occasional Studies (June 1982), pp. 46-63. Krugman, Paul R., and Maurice Obstfeld. International Eco nomics, Theory and Policy, (Scott, Foresman/Little, Brown College Division, 1988). Yeats, Alexander J. “ On the Accuracy of Partner Country Trade Statistics,” Oxford Bulletin of Economics and Statistics (November 1978), pp. 341-61. Appendix Data Sources for the U.S. Export Discrepancy and the U.S. Balance of Payments Statistical Discrepancy p iled from the IM F D irection s o f Trad e Statistics States. T h e estim ated U.S. export discrep a n cy for the 17-country sam ple o f industrial countries, tape and the U.S. balance o f paym en ts statistical XDI17, w as obtain ed as follow s: T h e bilateral im p ort and export data w ere c o m discrep a n cy w as obtain ed from International Fi nancial Statistics tape. T h e U.S. export d iscrepan cy w as estim ated us ing 17 industrial countries — the 20 countries classified as industrial by the IM F less the N eth er lands, S w itzerlan d and, o f course, the U nited http://fraser.stlouisfed.org/ FEDERAL RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 17 XDI17, = 2 (MUSt)/1.052) -X U S tj, j= l 23 w h ere MUSti = CIF im p orts o f cou n try j from the U nited States in y e a r t. XUSt| = FAS exports o f the U nited States to cou n try j in y e a r t. T h e in clu d ed countries in XDI17 are: Australia, Austria, Belgium -Luxem bourg, Canada, Denmark, Finland, France, Germany, Iceland, Ireland, Italy, Japan, N e w Zealand, Norw ay, Spain, Sw eden and the U nited K in gdom . T h e U.S. balance o f paym ents statistical d isc rep ancy, SD,, w as obtain ed from the IFS tape o f the IM F. Since the re co n ciled adjustm ent to the bilat eral U.S.-Canadian m erch andise trade balance is rem oved from the data (1970-86), the annual U.S.Canadian recon ciliation , RAUSCA,, is subtracted from the rep orted SD, SDHAT,, to get SDTOT,. That is, from id en tity 4, SDHAT, = - [CAB, + k A b ,] + RAUSCA,, so that SDTOT, = SDHAT, - RAUSCA,. RAUSCA, w as obtain ed from U.S. D epartm ent o f C om m erce (1987b), table 14. P rior to 1970, RAUSCA, is zero, so SDHAT, and SDTOT, are equal. Source Data and Constructs (billions of dollars) Year USCAB SDHAT SDTOT XDI17 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 $2.82 3.82 3.38 4.40 6.82 5.41 3.03 2.59 0.59 0.42 2.33 -1 .4 5 -5 .7 8 7.07 1.92 18.13 4.17 -1 4 .4 9 -1 5 .4 5 -0 .9 7 1.84 6.87 -8 .6 4 -4 6 .2 8 -1 0 7 .0 9 -1 1 6 .4 3 -1 4 1 .4 6 -$ 1 .0 2 -1 .0 0 -1 .1 1 -0 .3 6 -0 .9 1 -0 .4 2 0.63 -0 .2 2 0.46 -1 .4 6 -0 .1 7 -9 .7 6 -1 .9 5 -2 .6 0 -1 .5 2 5.88 10.53 -2 .0 5 12.59 25.45 25.01 19.96 36.12 11.18 26.81 17.87 24.06 -$ 1 .0 2 -1 .0 0 -1 .1 1 -0 .3 6 -0 .9 1 -0 .4 2 0.63 -0 .2 2 0.46 -1 .4 6 0.43 -8 .8 6 -0 .9 5 -1 .2 0 - 0 .0 2 7.58 11.93 - 0 .0 5 15.09 29.85 29.71 24.06 39.22 13.38 31.41 23.87 33.66 $0.5422 1.0101 1.3890 0.1014 0.2590 0.5876 0.5828 0.7807 1.7465 2.0261 2.3949 2.7078 2.8999 4.1984 5.3438 4.9660 5.6111 5.8041 7.9439 11.5811 16.5480 12.2240 12.9268 12.0168 11.1532 13.3568 15.4334 XDI17NC' $0.5851 0.9723 1.4175 0.2399 0.3888 0.6523 0.6600 0.5358 1.4384 1.6202 2.0144 2.2254 2.2613 2.7936 3.6280 3.6362 3.5931 5.5742 3.4205 6.0699 10.7420 6.5639 8.1274 6.0545 4.6521 5.9909 4.6719 1XDI17-XDCANADA SEPTEMBER/OCTOBER 1988 24 Cletus C. Coughlin and Thomas B. Mandelbaum Cletus C. Coughlin is a senior economist and Thomas B. Mandel baum is an economist at the Federal Reserve Bank of St. Louis. Thomas A. Pollmann provided research assistance. Why Have State Per Capita Incomes Diverged Recently? J . ROM the early 1930s to the late 1970s, differences in p er capita in com e across states n arrow ed substantially. By 1978, for exam ple, on e m easure o f state p e r capita in com e in equality h ad fallen to less than o n e-th ird o f its 1932 value. Since 1978, h ow ever, this tren d tow a rd greater in com e equ al ity across states has been sharply reversed; by 1987, state p e r capita in com e in equ ality h ad risen back to its 1966 level. Historically, disparate regional in com e grow th has g en erated political pressures to alter federal policies. For exam ple, faster in com e grow th in the South and W est relative to the Northeast and M id w est in the 1970s led to charges that these differ ential grow th rates w ere due, in part, to the distri bu tion o f federal govern m en t exp en d itu res.1Yet, the Sun Belt-Frost Belt controversy arose du rin g a p erio d in w h ic h state p er capita in com e grow th w as converging. Pressures fo r in creased federal action in the realm s o f farm policy, trade p o licy and industrial targeting are even m o re likely to 1For example, see “The Second War Between the States” (1977) and “ Federal Spending: The Northeast's Loss is the Sunbelt’s Gain” (1976). d iffe re n t views of the appropriate federal role can be found in Reich (1988) and Weinstein and Gross (1988). 3The reversal of the income inequality trend was confirmed statistically by regressing state per capita income inequality on time. To allow for the possibility of a structural break in 1978, a piecewise linear regression model was estimated. The results, based on conventional hypotheses tests, indicated a negative relationship between inequality and time until 1978 and a positive relationship thereafter. FEDERAL RESERVE BANK OF ST. LOUIS appear because o f the in creasin g in com e diver gen ce across states in the 1980s.This study pursues tw o objectives. First, it id e n tifies the specific states responsib le fo r the in creasing in equality o f state p e r capita incom e. Second, it exam ines w h e th er w ell-k n ow n d es crip tions o f regional g row th and m a jor eco n om ic changes can explain this n e w p h en om en on . INCREASING INEQUALITY — WHICH STATES ARE DIVERGING? T h e recen t sharp reversal o f the 45-year trend tow ard lesser state p e r capita in com e in equ ality is sh ow n in chart l . 3 T h e m easure o f in com e in eq u a l ity across states used in the chart is the annual coefficient o f variation o f state p er capita in com e; its p recise calcu lation is d eta iled o n page 28. In co m e in equality across states gen erally d eclin ed from 1932 to 1978; since then, it has risen g ra d u 25 Chart 1 Inequality of State Per Capita Income Percent ally, but consistently. By 1987, it had clim b ed back to its mid-1960s levels.4 Differential in com e g row th across states has tw o o p p o sin g effects on state p er capita in com e in equ ality m easures. In co m e in equ ality is red u ced w h e n states w h o se p e r capita in com es e x ceed (are less than) the average fo r all states exp erien ce s lo w er (faster) than average grow th in in com e. S im ilarly in com e in equ ality rises w h e n states w h o se p er capita in com es ex ceed (are less than) the average fo r all states ex p erien ce faster (slow er) than average in com e grow th . T h e net effect on in com e in equality d ep en d s on w h ic h o f these tw o Percent possible gro w th patterns p redom in ate. As chart 1 indicates, the fo rm er pattern p red o m in a ted until the en d o f the 1970s, but the latter result has o c cu rred since then. Table 1 iden tifies the im pact o f each state on in com e in equ ality since 1978. T h e analysis in this table, and throu ghou t the article, focuses o n the state’s relative p e r capita in com e — the state’s p er capita in com e expressed as a p ercen t o f the p er capita in com e o f all (continental) states. For exam ple, if M ississippi's p e r capita in com e in 1978 w as three-fourths o f the average p e r capita in com e o f all states fo r that year, its relative p e r capita in "Personal income consists of labor and proprietor income, dividends, interest, rent and transfer payments. Transfer pay ments differ from the other components in that they are not derived from current economic activity. The interstate inequality of per capita income minus transfers followed similar trends as the inequality of total per capita income; the coefficient of variation of non-transfer per capita income for the 48 states trended downward from 23.3 percent in 1946 to a minimum of 13.8 percent in 1976, then rose to 19.1 percent by 1987. SEPTEMBER/OCTOBER 1988 26 Table 1 Classification of States Based on Per Capita Income Levels and Changes State Per Capita Income as a Percent of State Average Upwardly Divergent1 Connecticut Massachusetts New Jersey New Hampshire New York Virginia Maryland Rhode Island Delaware Florida Downwardly Divergent2 Idaho Montana Louisiana Utah North Dakota West Virginia Oklahoma Indiana New Mexico Texas Upwardly Convergent3 Georgia Maine Vermont North Carolina Downwardly Convergent4 Wyoming Nevada Oregon Iowa Michigan Washington 1978 1987 123% 109 119 100 113 101 113 98 108 101 146% 131 139 119 125 113 123 107 113 106 Percentage Point Change 1978-87 23 22 20 19 12 12 10 9 5 5 93 96 88 87 99 84 94 102 87 102 82 85 79 78 91 76 87 96 81 96 89 86 90 86 98 95 98 91 9 9 8 5 117 124 107 107 113 114 89 111 96 99 106 107 -2 8 -1 3 -1 1 -8 -7 -7 com e fo r 1978 w o u ld equal 75 percen t. A state is ju d ge d to have h ad an im p act on in com e in equ al ity if its relative p er capita in com e ch an ged by 5 p ercen tage points or m ore b etw een 1978 and 1987. T h e in com e changes o f 20 states ten d ed to in crease inequality. T en states w ith above-average p er capita in com e in 1978 — C onnecticut, Massa -1 1 -1 1 -9 -9 -8 -8 -7 -6 -6 -6 Idaho, M ontana, Louisiana, Utah, N orth Dakota, W est Virginia, Oklahom a, Indiana, N e w M exico and Texas — that ex p erien ced substantially slo w er than the average grow th . W e call these states “ d o w n w a rd ly divergent.” W e have also id en tified 10 states w h o se in com e chusetts, N e w Jersey, N e w H am pshire, N e w York, Virginia, M aryland, R h od e Island, D elaw are and F lorid a — e x p erien ced substantially faster grow th changes have ten d ed to red u ce inequality. Four o f th em — Georgia, M aine, V erm ont and N o rth Caro b etw een 1978 and 1987 than the average. W e call these states "u p w a rd ly d ivergent.” T h ere w ere 10 states w ith below -average p e r capita in com e — b e lo w the average across states in 1978, but w h o have grow n faster than this average since then. Th ese states are called “ u p w a rd ly convergent.” FEDERAL RESERVE BANK OF ST. LOUIS lina — w e re states w h o se p e r capita in com es w ere 27 Table 1 cont’d. No Substantial Change5 Illinois Ohio South Dakota Kentucky Mississippi Nebraska Arkansas Wisconsin Kansas Pennsylvania Alabama Colorado Missouri Arizona California South Carolina Tennessee Minnesota 118 105 91 86 74 103 81 104 105 105 82 109 100 95 121 80 86 106 114 101 87 83 71 100 79 102 104 104 82 110 101 97 123 82 88 110 -4 -4 -4 -3 -3 -3 -2 -2 -1 -1 0 1 1 2 2 2 2 4 ’States with above-average per capita income in 1978 and with a 5 or more percentage-point in crease in per capita income as a percent of the state average. For Rhode Island, a state with below-average per capita income in 1978 and above-average per capita income in 1987, the rise in relative income resulted in the state's income absolutely further from the average in 1987 than in 1978. 2States with below-average per capita income in 1978 and with a 5 or more percentage-point drop between 1978 and 1987 in state per capita income as a percent of state average. For Indiana and Texas, states with above-average income in 1978 and below-average income in 1987, the drops resulted in the states’ being absolutely further from average per capita income in 1987 than in 1978. 3States with below-average per capita income in 1978 and with a 5 or more percentage-point in crease between 1978 and 1987 in state per capita income as a percent of the state average. “States with above-average per capita income in 1978 and with a 5 or more percentage-point de cline between 1978 and 1987 in state per capita income as a percent of the state average. For Wyoming, Oregon and Iowa, states with above-average per capita income in 1978 and belowaverage per capita income in 1987, the drop resulted in per capita income closer to the state aver age in 1987 than in 1978. 5States whose absolute percentage-point change in per capita income as a percent of the states was less than 5 percent between 1978 and 1987. Six states — W yom ing, Nevada, Oregon, Iowa, M ich igan and W ashin gton — w ere “ d o w n w a rd ly con vergen t.” Th ese states, w h o se p e r capita in com es ex ce e d e d the average across states in 1978, but w h o have gro w n slo w er than this average, also contribu ted to red u ced inequality. O f all the states, W yo m in g is the hardest to categorize. Be tw een 1978 and 1987, it ex p erien ced the largest percen tage p oin t d eclin e in relative p er capita in com e o f the 48 states. This 28-point declin e d ro p p e d W yo m in g from an above-average in com e level in 1978 to b elow -average b y 1987. I f the analy sis had focu sed on changes from 1984 to 1987, W yo m in g w o u ld have been labeled as d o w n w a rd ly divergent rather than d o w n w a rd ly convergent. Finally, 18 states had relative p e r capita in com es that ch anged less than 5 p ercen tage points b e tw een 1978 and 1987. Th ese states had little im pact on the recent changes in inequality. T o p rovid e a geogra p h ic o verview o f the results p resen ted in table 1, a m ap is presen ted. As the m ap reveals, states exp erien cin g relatively rapid p er capita in com e grow th are, w ith ou t exception, Atlantic Coast states. Since these states ten d to have p er capita in com es above the average across states, their rapid grow th tends to contribu te to in creasin g inequality. On the o th er hand, states ex p erien cin g relatively s lo w p e r capita in com e grow th are scattered across the rem a in d er o f the contin en tal U nited States. Th e fo llo w in g analysis exam ines som e o f the p op u la r description s o f regional grow th and som e m a jor eco n om ic changes to see if they can explain this rising in equality. SEPTEMBER/OCTOBER 1988 28 Measuring Income Inequality T h e m easure o f in com e in equality used in this article is the coefficien t o f variation o f an nual state p e r capita in com es across the 48 contin en tal states (INEQ).' T h e coefficien t o f variation is the standard de\iation o f a series d ivid ed by its mean. For each year, IN EQ m ea sures the degree o f dispersion o f state p er ca p ita in com es about the m ean state p er capita in com e (M EAN). W ith each state w eigh ted equally, M EAN is calcu lated as follow s: 48 M EAN = 1 SPCIt / 48, i= 1 w h ere i = subscript den o tin g the individual states and SPCI = state p er capita in com e. Thus, the IN EQ is calcu lated as follow s: 48 IN E Q = [{ 2 (SPCI, - M E A N )2/ 47 H i= 1 /MEAN, A larger value o f IN E Q indicates greater varia tion b etw een state p e r capita in com es and, thus, greater inequality.2 If p e r capita in com e rose (fell) in a state w ith below -average p er c a p 'Data for the continental, rather than the entire, United States are used because no consistent income series is available for Hawaii or Alaska for the postwar period. 2Because state income data do not correct for cost-of-living differences among states, the inequality measure may not accurately reflect the real variations in per capita income levels among states. No reliable state cost-of-living data exist to make such adjustments. A related issue is interstate differences in price changes over time. If states with aboveaverage per capita income in 1978 experienced substan tially higher inflation between 1978 and 1987 than lowincome states, the rise in inequality could be due to these differences with no change in the inflation-adjusted distribu tion of per capita income. Price deflators for individual states are unavailable: however, regional deflators show little difference in inflation between 1978 and 1987. Using a December 1977 base, the consumer price index (for all urban consumers) for November 1987 was 186.2 for the ita in com e o r d ec lin ed (rose) in a high p ci-ca p ita in com e state, o th er things equal, INEQ w ou ld d ec lin e (increase)/1 Unlike the standard deviation, the coefficient o f variation used in co m p u tin g INEQ reflects dispersion relative to the m ean and can be used to com p a re the d eg ree o f in equ ality in different years w ith differin g m eans. For exam ple, if per capita in com e in each state d o u b led b etw een 1970 and 1980, the standard deviation fo r 1980 w ou ld be tw ice that o f 1970. T h e coefficien t o f variation, how ever, w o u ld s h o w n o change since it is stan d ardized by the m ean p e r capita incom e. For the coefficien t o f variation to be an u nbi ased m easure o f inequality, the u n d erlyin g data must be n orm ally distributed.4 Using the Shapiro-W ilk (1965) statistic, the state p er capita in com e series w as tested fo r n orm ality for each year. T h e null hypothesis, that the state p er capita in com e data are a ran dom sam ple from a norm al distribution, co u ld not be rejected at the 5 percen t level fo r any years in the p ostw ar period. Northeast, 184.7 for the North Central Region, 185.1 for the South and 187.4 for the West. 3A related measure of income inequality, the standard devia tion of the ratio of regional to national per capita income was used in Browne (1980) and Ray and Rittenoure (1987). The simple correlation between INEQ and the standard deviation of the ratio of state to national per capita income was 0.999 in the 1948-87 period. Williamson (1965) p. 11, also used a related inequality measure: a populationweighted coefficient of variation of per capita income; the measure is computed identically to INEQ except each region's squared deviation from the mean is multiplied by its share of the national population. For the 1946-87 period, a correlation of 0.985 was found between INEQ and a population-weighted coefficient of variation using state per capita income. 4See Yotopoulos and Nugent (1976), pp. 242-43. THE SHIFT TO THE SUN BELT en ed in equality du rin g the 1970s. Businesses, par ticularly m anufacturing, m igrated to the Sun Belt T h e shift o f industrial activity from the n a tio n ’s Frost Belt to the Sun Belt contribu ted to the less from the Frost Belt fo r various reasons, in clu d in g lo w e r w a ge rates.5 Since m anufacturing w ages are 5See Crandall (1986), pp. 124-27, for a brief survey of empirical research documenting and explaining manufacturing’s shift to the Sun Belt. FEDERAL RESERVE BANK OF ST. LOUIS 29 States Classified by 1978-87 Per Capita Income Change Upwardly Convergent I I Downwardly Convergent Upwardly Divergent I I Downwardly Divergent w ell above the average w age o f all industries in all regions o f the nation, this shift o f labor d em an d from h igh er-w age to lo w er-w a ge states p ro d u ce d h igh er relative grow th in per capita in com e in the lo w er-in com e states and relatively lo w e r in com e grow th in the h igh er-in com e states.6 For exam ple, using on e listing o f Frost Belt and Sun Belt states (see table 2), the Sun Belt’s share o f (continental) U.S. m anufacturing em p loym en t in creased from 34.4 p ercen t in 1969 to 39.0 percen t in 1978, w h ile the Frost Belt's share d ecreased from 51.3 percen t to 46.2 percent. D uring the sam e period, average relative p er capita in com e for the Sun Belt states in creased from 91.2 percen t in 1969 to 92.6 percen t in 1978; in the Frost Belt states, it fell from 112.4 p ercen t in 1969 to 106.3 p ercen t in 1978. This shift has con tin u ed in the last 10 years. Th e Sun Belt's share o f m anufacturing em p loym en t in creased from 39.0 percen t in 1978 to 43.7 percen t Non Substantial Change in 1987, w h ile the Frost B elt’s share decreased from 46.2 p ercen t to 41.1 percent. A lthou gh the shift, by itself, tends to redu ce in com e inequality, the actual p er capita in com es fo r the tw o regions have not con tin u ed to converge over this period. W h ile the average p e r capita in com e fo r the Sun Belt states as a p ercen tage o f the average in com e fo r all states rose slightly from 92.6 p ercen t to 93.1 p ercen t b etw een 1978 and 1987, it ju m p ed from 106.3 percen t to 111.1 p ercen t in the Frost Belt states. One reason w h y p e r capita in com es in the Frost Belt and the Sun Belt have sto p p ed con vergin g since 1978 is that the shift o f m anufacturing activ ity to the Sun Belt is less w id esp rea d than in p re vious decades; since 1978, m anufacturing trends in m any states d iffered sharply from that o f their region. For exam ple, the Frost B elt’s share o f m a n ufacturing w orkers con tin u ed to d eclin e after 6ln 1987, for example, average weekly earnings for production workers in the nation's manufacturing sector was $406, 30 percent higher than the private-sector average. SEPTEMBER/OCTOBER 1988 30 Table 2 Impact of Sun Belt and Frost Belt States on Inequality Sun Belt States Alabama — Arizona — Arkansas — Delaware — California — Florida — Georgia — Kentucky — Louisiana — Maryland — Mississippi — New Mexico — North Carolina — Oklahoma — South Carolina — Tennessee — Texas — Virginia — West Virginia — No Substantial Change No Substantial Change No Substantial Change Upwardly Divergent No Substantial Change Upwardly Divergent Upwardly Convergent No Substantial Change Downwardly Divergent Upwardly Divergent No Substantial Change Downwardly Divergent Upwardly Convergent Downwardly Divergent No Substantial Change No Substantial Change Downwardly Divergent Upwardly Divergent Downwardly Divergent Frost Belt States Maine New Hampshire Vermont Massachusetts Rhode Island Connecticut New York New Jersey Pennsylvania Ohio Indiana Illinois Michigan Wisconsin — — — — — — — — — — — — — — Upwardly Convergent Upwardly Divergent Upwardly Convergent Upwardly Divergent Upwardly Divergent Upwardly Divergent Upwardly Divergent Upwardly Divergent No Substantial Change No Substantial Change Downwardly Divergent No Substantial Change Downwardly Convergent No Substantial Change SOURCE: Weinstein, Gross and Rees (1985) and table 1. 1978, but m anufacturing in m ost N e w England states g rew as fast as, o r faster than, the nation. these states, rapid in com e grow th w as fu eled by the expansion o f constru ction and services, e sp e cially health, business and financial services.8 M anufacturing job shares rem ained constant b e tw een 1978 and 1987 in Maine, Massachusetts and Connecticut, w h ile rising in N e w H am pshire and Verm ont. Th e rapid grow th o f h igh -tech n ology m anufacturing b etw een 1978 and 1984, particu larly com pu ter- and defense-related produ ction, was largely responsib le fo r the rapid grow th o f per capita in com e in N e w England.7This grow th c o n tribu ted to the Frost Belt’s relatively rapid in com e grow th and the nation's increasing in com e in equ ality since 1978. As table 2 shows, the higherin com e states o f Connecticut, N e w H am pshire and M assachusetts are classified as u p w a rd ly divergent. At the sam e time, som e Sun Belt states have not shared in that region's industrial expansion. M an ufacturing em p loym en t from 1978 to 1987 g rew substantially slo w er in W est Virginia and Lou isi ana and no faster in Kentucky, M aryland, Okla h om a and Ten n essee than it did in the nation. T h e slo w er grow th in these states m ay have stem m ed, in part, from th eir specialization in en ergy-related industries, an issue discussed later in this article. As table 2 indicates, Louisiana, Okla h om a and W est Virginia w ere am ong the d o w n w a rd ly divergent Sun Belt states. D espite a sharp loss o f m anufacturing jobs since 1978, N e w York, N e w Jersey and Bhode Island T o sum m arize, m anu facturing activity has c o n tinu ed to shift from the Frost Belt to the Sun Belt have had relatively rapid p er capita in com e states in the 1980s, but n ot as w id e ly as in p re grow th, contribu tin g to the rising inequality. In vious decades: in fact, a n u m ber o f states in both 7See Bradbury and Browne (1988). Manufacturing, however, was not entirely responsible for New England’s per capita income growth, especially since 1985. Rapid growth of earn ings in construction and in service-producing industries (espe cially finance, insurance, real estate, medical and business services) combined with relatively slow population growth to spur New England's expansion. 8U.S. Department of Commerce (1987), p. 2, and Ray and Rittenoure (1987) p. 244, briefly discuss sources of growth in FEDERAL RESERVE BANK OF ST. LOUIS Mid-Atlantic States. Gross and Weinstein (1988) argue that the rapid growth of the New England and Mid-Atlantic economies in the 1980s is at least partially due to a rise in federal spend ing in those regions, particularly grants-in-aid and procurement. The slower economic growth of some Sun Belt states, mean while, allegedly stems from a decline in the federal expendi tures they receive. 31 “b elts” have ex p erien ced m anufacturing grow th cou n ter to that o f th eir region as a w h o le. Thus, rather than con tin u in g to converge as they had in the early and m id d le 1970s, the gap b etw een per capita in com es in the Frost Belt and Sun Belt states has w id e n e d since 1978. THE BI-COASTAL ECONOMY A cco rd in g to a stu dy released in 1986 by the D em ocratic staff o f the Joint E con om ic C om m ittee o f the U.S. Congress, national e co n o m ic grow th betw een 1981 and 1985 w as con cen trated in states on the East Coast and in California." T h e rapid expansion o f these states relative to the nation's in terior states led to the ch aracterization o f the U nited States as a bi-coastal econ om y, despite the absence o f O regon and W ashington from the list o f fast-grow in g states. For exam ple, the study noted that real earnings g rew at a 4 p ercen t rate in the coastal states du rin g the 1981-85 period, c o m pared w ith a 1.4 percen t rate in the non-coastal states. D oes the bi-coastal econ om y, w h ic h is prim arily a d escrip tion rather than an explanation o f the pattern o f grow th, p rovid e insights into the in creasing in equality o f state p e r capita in com e? T w o questions must be an sw ered affirm atively. First, are the bi-coastal states ex p erien cin g m ore rapid g row th o f p er capita in com e? T h e an sw er to this qu estion is “yes." Table 3 lists the bi-coastal states and their p er capita in com e perform an ce fo r 1978-87. O f the 16 bi-coastal states, 14 g rew substantially faster in p e r capita in com e than aver age. California, the sole W est Coast state, and South C arolina ex p erien ced no substantial change in th eir relative p e r capita in com e grow th. Second, did these rapid ly g ro w in g states also have above-average p er capita in com es? If so, the rapid grow th causes th eir p e r capita in com e to rise fu rther above the average, thus, increasing state in com e inequality. O f the 14 states w ith ra p idly g row in g p er capita in com e, 10 are classified as 9The study, The Bi-Coastal Economy, was released in July 1986 by the Joint Economic Committee of the U.S. Congress. See U.S. Congress (1986). ,0The Joint Economic Committee study suggested a number of reasons for the uneven pattern of regional growth during the first half of the 1980s. The study suggests that “ a central cause is trade and the current massive imbalance in trade that exists between the United States and its trading partners” that dispro portionately affects interior states. U.S. exports of both agricul tural and nonagricultural commodities had declined to some extent, according to the authors, because of increased compe tition from Third World nations attempting to earn foreign cur Table 3 Impact of Bi-Coastal States on Inequality California — No Substantial Change Connecticut — Upwardly Divergent Delaware — Upwardly Divergent Florida — Upwardly Divergent Georgia — Upwardly Convergent Maine — Upwardly Convergent Maryland — Upwardly Divergent Massachusetts — Upwardly Divergent New Hampshire— Upwardly Divergent New Jersey — Upwardly Divergent New York — Upwardly Divergent North Carolina — Upwardly Convergent Rhode Island — Upwardly Divergent South Carolina — No Substantial Change Vermont — Upwardly Convergent Virginia — Upwardly Divergent SOURCE: U.S. Congress (1986) and table 1. divergent; on ly fo u r o f these states are convergent. In fact, tin; 10 divergent states accou nt for all the u p w a rd ly d ivergent states in the contin en tal U nited States and the fou r convergent states ac count fo r all the u p w a rd ly convergent states. Thus, relatively rapid East Coast in com e grow th w as a prim ary in flu ence in increasing the in equ al ity o f state p er capita in com e. W h ile explanations fo r the relatively rapid grow th o f in com e in the coastal states are specu la tive, explanations o f w h y in com e grow th in in te rior states lagged b eh in d are m ore p recise.10Fall ing en ergy prices and the agricultural crisis are tw o frequ en tly cited reasons fo r the below -average perform an ce. The Influence o f Falling Energy Prices Th e e co n o m ic grow th o f states e n d o w e d w ith substantial en ergy resources tends to be directly related to en ergy prices, w h ile the eco n o m ic rency to pay interest on their loans. Also, increased competi tion from imported manufactured goods in domestic markets was claimed to be partially responsible for the observed pattern of regional growth. The study’s final explanation relates to the strong job growth in the service industry, particularly in firms engaged in importing, advertising, financing and selling foreignmade goods. Such industries are strongly concentrated on the coasts, according to the study, and their growth helped boost the coastal states. SEPTEMBER/OCTOBER 1988 32 Chart 2 Relative Energy Prices and Relative Per Capita Income in Energy and Non-Energy States Index 1982=100 175 Percent of average 105 150 Average per capita income in non-energy states scale > 125 100 100 95 —-—/ 75 / Relative energy price < scale 50 25 1968 70 72 74 76 grow th o f en erg y-p o o r states tends to be inversely related.11As chart 2 shows, en ergy prices relative to the general p rice level rose rapid ly from 1973, peaked in 1981, then fell through 1987.12 If energyrich states are also gen erally lo w er-in com e states, the d eclin e in en ergy prices in the 1980s has c o n tributed to the increasing interstate in equ ality bv "S ee Manuel (1982) and Brown and Hill (1987) for empirical studies documenting the relationship between energy prices and state economic growth. Miernyk (1977) and Manuel (1982) discuss why energy prices and state economic growth are linked. As they rise, energy costs become an increasingly important factor in determining where to locate an energyintensive industry. Such relocation tends to shift employment opportunities from energy-poor regions to energy-producing states. Higher energy prices may also stimulate greater invest ment in energy production and exploration, increasing jobs in energy-producing states. Although profits from relocating manufacturing firms are likely to be distributed to owners throughout the nation, the increased employment tends to increase income in energy-producing states. In contrast, energy-poor states are burdened with higher costs for fuel and inputs in which energy costs are an important component. When energy prices fall, the advantages shift to states that heavily import oil rather than produce it. FEDERAL RESERVE BANK OF ST. LOUIS 78 Average per capita \ income in energy \ states \ scale > 80 82 84 86 90 85 1988 slo w in g in com e grow th in these states relative to those that pu rchase m ost o f th eir en ergy re sources from out-of-state sources. Th e evid en ce supports this explanation. As chart 2 shows, relative p er capita in com e in en ergy states generally fo llo w e d the rise and fall o f en ergy prices, w h ile the relationship w as an inverse one 12Relative energy prices in this article are indicated by the pro ducer price index for fuels, related products and electric power divided by the GNP implicit price deflator for the private busi ness sector. The oil embargo in 1973-74 contributed directly to the price increases for petroleum and indirectly to price in creases for other energy sources as energy users searched for oil substitutes. Relaxation of price controls during the period contributed to the price increases of natural gas. The easing of energy prices in the current decade reflects a worldwide in crease in global oil supplies as international oil cartels are unable to agree on production quotas. Also, heavy investment to increase energy efficiency by car makers, businesses and households has caused the quantity of energy demanded to grow substantially slower than the rest of the nation’s econ omy, according to Schmidt (1988). 33 Table 4 Impact of Energy-Producing States on Inequality___________________ Wyoming — West Virginia — Oklahoma — Louisiana — Kentucky — Texas — North Dakota — New Mexico — Colorado — Montana — Utah — Downwardly Convergent Downwardly Divergent Downwardly Divergent Downwardly Divergent No Substantial Change Downwardly Divergent Downwardly Divergent Downwardly Divergent No Substantial Change Downwardly Divergent Downwardly Divergent NOTE: Energy-producing states are those in which earn ings from oil and gas extraction and coal mining produced at least 3 percent of the state’s total earnings in 1981. States are ordered from those with the highest to the lowest percentage. SOURCE: table 1. fo r the o th er states.13Table 4 lists the 11 en ergy states in the continental U.S. in w h ic h earnings from oil and gas extraction and coal m in in g p r o d u ced at least 3 percen t o f the state’s total earn ings in 1981, the y e a r in w h ich en ergy prices peaked and oil and gas extraction and coal m in in g p rovid ed its largest share o f total U.S. earnings in the postw ar p erio d .14Th e en ergy states are listed in d escen d in g o rd er a ccord in g to the p rop o rtio n o f th eir earnings d erived from oil and gas extrac tions and coal m ining, ranging from W yo m in g w ith 18.6 p ercen t to Utah w ith 3.1 percent. In 1969, before the sharp rise in en ergy prices, p e r capita in com e in the en ergy states averaged 88.7 p ercen t o f that fo r all 48 continental states. This p ro p o rtio n rose to 95.4 p ercen t b y 1978 and peaked at 96.7 p ercen t by 1981. By 1987, after e n ergy p rices had d ec lin ed substantially, the average p e r capita in com e in en ergy states d ec lin ed to 86.8 percen t o f the average o f all states. 13ln the 1947-87 period, the correlation between relative energy prices and the average relative per capita income of energy states is 0.54, significantly different from zero at the 1 percent level. The correlation of relative energy prices and the relative per capita income of non-energy states, - 0.54, is identical in absolute value, but negatively signed. This correlation is also significant at the 1 percent level. 14The validity of this classification is suggested by the substantial overlap between this list of energy states and those suggested in two previous studies. Nine of the 11 states shown in table 4 were among the 10 continental U.S. states with a ratio of energy production to energy consumption greater than unity in O f the 11 en ergy states, all but Kentucky, C o lo rado and W yo m in g w ere classified as d o w n w a rd ly d ivergent (see table 4).lr' In h a lf o f these eight d o w n w a rd ly divergent states (Oklahom a, N e w M exico, Louisiana and Texas), relative p e r capita in com e rose from 1978 through the early 1980s, then fell sharply in subsequent years, fo llo w in g en ergy p rice trends. W yo m in g also exhibited this pattern o f grow th : its relative p er capita in com e g rew to 121 percen t o f the state average by 1980, rem ained high in 1981, then p lu m m eted to 89 percen t by 1987. A lthou gh classified as d o w n w a rd ly convergent, W y o m in g ’s p er capita in com e fell b elo w the national average in 1984 and, thus, has contribu ted to the greater in equality o f state in com e since that year. In the rem ainin g d o w n w a rd ly divergent en ergy states (West Virginia, N orth Dakota, Utah and M o n tana), relative p er capita in com e tren ded d o w n w ard throughout the 1978-87 period. A lthou gh the fall in en ergy prices u n d ou b ted ly con tribu ted to th eir slo w in g after 1981, their sluggish in com e grow th in previous years suggests that o th er fac tors w ere at w ork as w ell. Th e im p orta n ce o f the en ergy price d eclin e as a con tribu tor to in creasin g interstate in equality can be seen m ore clearly b y con sid erin g the list o f d o w n w a rd ly d ivergent states in table 1. Energy states accou nt fo r eight o f the 10 d o w n w a rd ly divergent states. In addition, W yom in g, has c o n tributed to in creasin g in equality since 1984. N on e o f the states w ith substantial u p w a rd m ovem en t o f relative p e r capita in com e w ere en ergy-rich states. Instead, these states w ere heavy im porters o f en ergy resources w h o g e n er ally b en efited from the ch ea p er en ergy resources in the 1980s. Since m ost states w ith substantial post-1978 in com e g ro w th had above-average p er capita incom es, the fall in en erg y prices also ten d ed to increase in equ ality by b oostin g their grow th fu rth er above the average. Thus, the de- 1976 (Corrigan and Stanfield, 1980). Eight of the 11 states identified as energy states in our study were among the nine continental U.S. states in which oil-price declines were associ ated with declines in total state employment in Brown and Hill (1988). 15Research by Hunt (1987) suggests that Colorado’s economy was not adversely affected by declining energy prices because of its diversified economic base which captured enough benefi cial effects of oil price declines to offset the negative effects. SEPTEMBER/OCTOBER 1988 34 Chart 3 Economic Indicators of U.S. Agriculture Dollars p er acre d in e in en ergy prices w as an im portant factor in increasing in equality in the 1980s."’ The Influence o f the “Farm Crisis” T h e first h a lf o f the 1980s has been a cco m p a n ied bv a w id e ly p u b licized e co n o m ic d eteriora tion o f the n atio n ’s agricultural sector.17Chart 3 show s tw o sym ptom s o f the so-called farm crisis. T h e value o f both the n atio n ’s farm exports and farm land g rew ra p id ly d u rin g the 1970s but d e clin ed du rin g the current decade. A d eclin e in the farm sector affects non-farm sectors d irectly linked to agriculture. T h ese in clu de su ppliers o f fertilizer and farm equ ipm en t 16Ray and Rittenoure (1987) found that declining energy prices contributed to the increasing inequality of regional per capita income in the 1980s. 17See Petrulis et al. (1987) for a discussion of the reasons for the farm crisis. 18Since the purpose of this analysis is to assess the possible effects of the farm sector downturn on state per capita personal FEDERAL RESERVE BANK OF ST. LOUIS B illions of dollars as w e ll as firms that transport, process and market agricultural produ cts. Less directly, a d eclin e in farm ing and agribusiness co u ld adversely affect o th er sectors as w ell, such as those p rovid in g ser vices to agricultural w orkers. A d eclin e in the n atio n ’s agricultural sector w o u ld m ost adversely affect state in com e in agriculture-intensive states. One m easure o f this intensiveness is the p rop o rtio n o f total state earn ings accou n ted fo r b y farm labor and p ro p rieto r earnings.18Table 5 displays the 12 states that d e rived at least 4 percen t o f th eir earnings from farms in 1981, the m ost recent peak in both agri cultural exports and farm land values. N orth Da- income, farm labor and proprietor earnings (a component of personal income) is a more appropriate measure of farm in come than net farm income. While real net farm income is a better measure of farm profitability, it includes corporate in come, which is excluded from the personal income series. 35 Table 5 Impact of Farm States on Inequality South DakotaNorth Dakota Iowa Nebraska Idaho Arkansas Montana Kentucky Minnesota Wisconsin Vermont Kansas - No Substantial Change - Downwardly Divergent - Downwardly Convergent - No Substantial Change - Downwardly Divergent - No Substantial Change - Downwardly Divergent - No Substantial Change ■No Substantial Change - No Substantial Change ■Upwardly Convergent - No Substantial Change NOTE: Farm states are those in which 4 percent or more of total 1981 state earnings were derived from farming. States are ordered from those with the highest to the lowest percentage. SOURCE: table 1. kota and South Dakota w ere the states m ost reli ant on farm ing, w ith 11.9 percen t and 15.1 percen t o f th eir total earnings directly d erived from agri culture. A verage p er capita in com e has d ec lin ed in farm states relative to nonfarm states since 1978. Be tw een 1978 and 1987, relative p er capita in com e in farm states d ro p p ed from 97 percen t o f the aver age to 93 percent. D uring the same period, the average o f relative p e r capita in com e in all oth er states rose from 101 percen t to 102 percent. D esp ite this divergence, fe w farm states con trib u ted substantially to interstate in com e inequality. As table 5 shows, on ly three o f the 12 farm states — Idaho, M ontan a and N orth Dakota — are clas sified as d o w n w a rd ly divergent. On the oth er hand, farm states accou nt for tw o o f the 10 convergent states. Belative p er capita in com e also fell substantially in Iowa, a state w ith aboveaverage p e r capita in com e in 1978, and p er capita in com e rose in Verm ont, a state w ith below average p e r capita in com e in 1978. Little change in relative p e r capita in com e o ccu rred in the rem a in in g seven farm states. Overall, the im pact o f the farm crisis on the recen t increase in in equality appears m inim al. CONCLUSION T h e 45-year d o w n w a rd trend in inequality en d ed in the late 1970s. T w en ty states, even ly d i vid e d b etw een below -average and above-average p er capita in com e states, are prim arily responsible fo r the in creasin g inequality. A ll states w ith aboveaverage p er capita in com e and relatively rapid in com e grow th are located on the A tlantic Coast. The states w ith below -average p e r capita in com e and relatively slow' grow th are scattered throu gh out the nation's interior. T h e Sun Belt-Frost Belt d escrip tion o f regional grow th has lim ited success in explain in g this p h e n om enon . T h e shift o f m anufacturing activity from the Frost Belt to the Sun Belt, w h ic h contribu ted significantly to the n arrow in g o f regional in com e differentials in the 1970s, has con tin u ed in the 1980s, but has affected few'er states. In deed, in recent years, m anu facturing has grow n relatively rapid ly in som e N e w England states, w h ile g r o w ing no faster than the national average in several Sun Belt states. T h e descrip tion o f the U.S. eco n o m y as a b i coastal e co n o m y w ith ra p id ly g ro w in g coastal and slo w ly gro w in g in terior states provid es a better insight in to the location o f states responsib le for the rising in com e inequality, but not necessarily the reasons fo r this result. T h e relatively p o o r p e r form an ce o f the in terior states has been attributed to various p roblem s related to agriculture as w ell as to falling en ergy prices. T h e agriculture crisis has little explanatory p o w er. A lthou gh the agricu l tural sector has w eaken ed in the 1980s, farm states accou nt fo r o n ly three o f the 10 d o w n w a rd ly diver gent states. On the o th er hand, d eclin in g en ergy prices have been a m a jor factor in increasing interstate in co m e inequality. Energy states account fo r eight o f the 10 d o w n w a rd ly d ivergent states. A n o th er en ergy state, W yom in g, has contribu ted to in creas ing in com e in equality since 1984. REFERENCES Bradbury, Katharine L., and Lynn E. Browne. “ New England Approaches the 1990s,” New England Economic Review (January/February 1988), pp. 31-45. Brown, S. P. A., and John K. Hill. “Lower Oil Prices and State Employment,” Contemporary Policy Issues (July 1988), pp. 60-68. Browne, Lynn E. “ Narrowing Regional Income Differentials,” New England Economic Review (September/October 1980), pp. 35-56. Corrigan, Richard, and Rochelle L. Stanfield. “ Rising Energy Prices — What’s Good for Some States is Bad for Others," National Journal (March 1980), pp. 460-69. Crandall, Robert W. “The Transformation of U.S. Manufactur ing,” Industrial Relations (Spring 1986), pp. 118-30. SEPTEMBER/OCTOBER 1988 36 “Federal Spending: The Northeast’s Loss is the Sunbelt’s Gain,” National Journal (Government Research Corporation, June 1976). Shapiro, S. S., and M. B. Wilk. “An Analysis of Variance Test for Normality (complete samples),” Biometrika (Volume 52, 1965), pp. 591-611. Gross, Harold T., and Bernard L. Weinstein. “ Frost Belt vs. Sun Belt in Aid Grants: Not a Fair Fight,” Wall Street Journal, August 23, 1988. “The Second War Between the States.’’ 17, 1977). Hunt, Gary L. The Impact of Oil Price Fluctuations on the Economies of Energy Producing States,” Review of Regional Studies (Fall 1987) pp. 60-76. U.S. Congress, Joint Economic Committee. “The Bi-Coastal Economy: Regional Patterns of Economic Growth During the Reagan Administration,” Staff Study by the Democratic Staff, mimeo (July 1986). Manuel, David P. "The Effects of Higher Energy Prices on State Income Growth,” Growth and Change (July 1982), pp. 26-37. Miernyk, William H. “ Rising Energy Prices and Regional Eco nomic Development,” Growth and Change (July 1977), pp. 2-7. Petrulis, Mindy, Bernal L. Green, Fred Hines, Richard Nolan and Judith Sommer. How is Farm Financial Stress Affecting Rural America? Agricultural Economic Report Number 568, U.S. Department of Agriculture (June 1987). Ray, Cadwell L., and R. Lynn Rittenoure. "Recent Regional Growth Patterns: More Inequality,” Economic Development Quarterly (August 1987), pp. 240-48. Business Week (May U.S. Department of Commerce. "Regional Differences in Per Capita Personal Income Widen in the 1980s.” Commerce News (August 20, 1987). Weinstein, Bernard L., and Harold T. Gross. “The Rise and Fall of Sun, Rust, and Frost Belts.” Economic Development Quarterly (February 1988), pp. 9-18. Weinstein, Bernard L., Harold T. Gross, and John Rees. Regional Growth and Decline in the United States, 2nd ed. (Praeger Publishers, 1985). Reich, Robert B. “The Rural Crisis, and What to Do About It,” Economic Development Quarterly (February 1988), pp. 3-8. Williamson, J.G. “ Regional Inequality and the Process of National Development: A Description of the Patterns.” Eco nomic Development and Cultural Change (July 1965), pp. 3-83. Schmidt, Ronald H. “Oil and the Economy,” Federal Reserve Bank of San Francisco Weekly Letter (May 27,1988). Yotopoulos, Pan A., and Jeffrey B. Nugent. Economics of Development: Empirical Investigations (Harper & Row, 1976). FEDERAL RESERVE BANK OF ST. LOUIS 37 Michael T. Belongia anil Kees G. Koedijk Michael T. Belongia is a research officer at the Federal Reserve Bank of St. Louis and Kees G. Koedijk is an assistant professor of monetary economics at Erasmus University, Rotterdam, The Netherlands. Anne M. Grubish and Rosemarie V. Mueller pro vided research assistance. Testing the Expectations Model of the Term Structure: Some Conjectures on the Effects of Institutional Changes T ■M. HE TR A D ITIO N A L expectation s m o d el of the term structure o f interest rates attem pts to explain h o w interest rates on a sim ilar debt instrum ent are related across different m aturities. It posits that, in a w o rld w ithou t risk or o n e in w h ich assets are perfect substitutes, the o n e-p erio d interest rate sh ou ld equal the ex p ected return to h old in g an instrum ent o f lon ger m aturity fo r on e period. Because the m o d el is based on the m ost fu n da m ental e co n o m ic assum ptions — rational behav io r b y individu als w h o act on all available in form a tion — it has h eld considerable appeal in a pplied research. E m pirical tests fo r data across a range o f cou ntries and sam ple periods, h ow ever, have ten d ed to reject this sim ple statem ent o f the ex pectations m od el.' M oreover, exp a n d in g the basic m o d el by a d d in g o th er explanatory variables, such as a tim e-varying risk p rem iu m or latent in form a tion variables, still has fou n d lim ited em pirical success in explain in g interest rate behavior.2Thus, a p u z z le rem ains: w h y is such a basic theoretical m o d el so frequ en tly rejected b y the data? In this article, using short m aturities in the E u rocu rren cy market, w e isolate several institu tional factors that m ight explain som e rejection s o f the expectation s m o d el. Alternatively, the analy sis m ay be view ed as an attem pt to suggest sp e cific characteristics o f p o lic y p roced u res that are inconsistent w ith the theoretical m o d el's assu m p tions. O ur results suggest that sin gle-cou n try esti mates o f the expectation s m o d el m ay om it im p o r tant in form ation because financial markets are h igh ly in tegrated across countries. M oreover, it appears as if the m an n er in w h ich m on etary p o l icy is co n d u cted has effects on interest rates that contribute to rejection s o f the theory. In particu lar, the expectation s m o d el does not h old in co u n tries w h e re the central bank — at least p e rio d i cally — fo llow s an exch an ge rate rule. A cco u n tin g fo r relationships across markets and fo r the m a n n er in w h ich m on etary p o lic y is co n d u cted re verses, in som e cases, the negative con clu sion o f sim ple, single equ ation estim ates o f term structure relationships. 'For a survey of these results, see Bisignano (1987). 2Examples of work along these lines are Shiller, et al. (1983) and Campbell and Clarida (1987). SEPTEMBER/OCTOBER 1988 38 THE EXPECTATIONS M ODEL APPLIED TO SHORT MATURITIES T h e em pirical version o f the expectation s m o d el can be w ritten as: <1* <r,., +, - r j = a + b(,F,,,, - r j + e, w h ere rlt is the y ie ld on a o n e-p erio d bill in p eriod t and ,FU+, is the current, observed fo iw a rd rate on a o n e-p erio d bill, o n e p erio d into the future.3 C o ef ficients to be estim ated are d en o ted a and b; e, is an erro r term w ith ze ro m ean and variance equal to a2. Thus, in equ ation 1, the d ep en d en t variable is the differen ce b etw een actual yield s on onep e rio d bills in con secu tive p eriod s and the explan atory variable is the differen ce b etw een the cur rent fo iw a rd and spot rates on o n e-p erio d bills. Equation 1 predicts that the change in onep erio d yie ld s sh ou ld be related to the forecasted change, as rep resen ted by the fo rw ard rate — spot rate spread. T h e expectation s h ypothesis im plies that, if the forw a rd rate is an unbiased p red icto r o f the future spot rate, the regression ’s slope c o e f ficient, b, sh ou ld not be significantly differen t from one and its in tercept, a, shou ld not be significantly differen t from zero. This p oten tia lly rich area fo r em pirical research has y ie ld e d fe w d efinitive results because tests o f the expectation s m o d el inevitably have been joint tests o f several m ain tain ed h ypotheses. T o cite just a fe w o f the p roblem s that arise, the m odel assumes a ze ro o r constant risk prem iu m . Th e p rob lem fo r estim ation, h ow ever, is that the risk (or, term ) prem iu m — som e system atic differen ce b etw een the long-term interest rate and the ex p ec te d future values o f short-term interest rates that reflects relative degrees o f u ncertainty — is unobservable. Thus, if an em pirical test rejects the h ypothesis a = 0 and b = 1, it is not possible to discrim inate b etw een true m o d el rejection and the possible effects o f a term prem iu m that has been assum ed, incorrectly, to be zero. In part for this reason, as w ill b e the case below , m an y stud 3For one derivation of this result, see Mankiw and Miron (1986), p. 214. Strictly speaking, this specification holds up to a con stant (the term premium), which we have ignored. The as sumption was that, for the short maturities used in this paper, term premium effects, if any, should be negligible. Also see Bisignano (1987). Cosset (1982) found that forward rates in this market are unbiased, but not optimal, predictors of future interest rates. He also found this market to be efficient in the sense that past information on interest rates is not useful in predicting future values of interest rates. Values for the forward rate, ,F1t+1, were calculated as twice the two-period interest rate minus the one-period rate. Because FEDERAL RESERVE BANK OF ST. LOUIS ies have ch osen to test a w ea k er form o f the e x p e c tations m o d el (b = 1) and in terpret the statistical significance o f the regression ’s in tercep t as in d ica ting the existen ce o f a term prem iu m .4 Th ere are o th er testing p roblem s as w ell. W h en data fo r lo n g er m aturities are studied, interest rate data often are estim ated from a fitted y ie ld curve rather than taken from observed market transac tions. In this instance, negative results m ight be a rejection o f the form ula used to approxim ate un observable interest rates rather than the exp ecta tions m odel. Finally, the rationality o f expectation s by market agents is assum ed but, again, this is difficult or im possible to test directly. A lthou gh m ore attention has been paid in recent research to m odels that isolate these assum ptions, it rem ains im possible to say w h e th e r negative results in d i cate a rejection o f the expectation s m o d el itself or sim ply one (or m ore) o f its u n d erlyin g assu m p tions. ESTIMATION OF THE EXPECTATIONS M ODEL As n oted in the in troduction , equations sim ilar to (1) have been estim ated w ith data fo r m an y countries and sam ple periods. W e illustrate these results b y estim ating equ ation 1 w ith Harris Bank data on spot three-m on th dep osit rates from the E u rocu rren cy market fo r the U.S., U.K., W est G er many, Japan and Sw itzerland; six-m onth d ep osit rates also w ere used, as exp la in ed in fo otn ote 3, to calculate values fo r the fo rw ard rate. T h e interest rates are calculated as sim ple rates. T h e data are Friday closin g quotes fo r the Friday closest to the begin n in g o f each m onth.5 T h e sam ple p erio d spans February 1981 through O ctober 1986. A l thou gh data p rior to 1981 are available, the Euroyen market w as th in ly traded and, in 1980, the C arter A dm in istration a d o p ted its Special Credit C on trol program . Because these factors the data in the study use three-month rates to represent the theoretical “one period,” the forward rate is calculated as twice the six-month (two period) rate minus the corresponding threemonth rate. 4See, for example, Shiller, et al. (1983). 5First-Friday-of-month data, rather than monthly averages of daily or weekly data, were used to avoid questions about how to treat partial weeks in adjoining months, months with different numbers of weeks and the gap between three, four-week months and a thirteen week quarter. See Hakkio and Leiderman (1984) for a discussion of these measurement issues. 39 that interest rate tiirre series closely approxim ate a ran dom walk. Overall, these m ixed results rep re sent the typical findings o f previous em pirical w ork on the expectation s m odel. Table 1 Estimates of the Basic Expectations Theory Relationship (monthly data, T h e m ixed results in table 1 can be in terpreted in tw o wavs. One in terpretation is that the e x p e c 1 9 8 1 . 0 2 - 1 9 8 6 . .1 0 ) a b R2 -0 .3 2 (0.92) 0.00 (0.00) -0 .4 6 (3.25) —0.17 (1.72) -0 .2 3 (0.82) -0 .2 6 * (2.68) 0.90 (0.31) 0.42* (4.13) 0.92 (0.27) 0.04* (2.97) 0.01 Country United States United Kingdom Germany Japan Switzerland 0.09 0.08 0.25 0.00 NOTE: Absolute values of t-statistics are in parentheses, t-statistics for b apply to the null hypothesis b = 1. An asterisk indicates a slope coefficient significantly different from one at the 0.05 level of significance. c o u ld adversely affect the test results, data p rio r to February 1981 are not used in estim ation.6 Finally, a co m m en t on the initial approach to estim ation is necessary. Because the data consist o f observations on three-m onth yie ld s sam p led m onthly, the changes in interest rates overlap and in trod u ce a secon d o rd er m ovin g average process into the data. Because this p ro p e rty o f the data w ill affect the estim ated co efficien ts’ standard errors, it must be co n sid ered by the estim ation tech niqu e. T h e H ansen-H odrick p roced u re w e use accounts fo r this p ro p e rty b y correctin g the m o d e l’s error term fo r serial correlation .7 tations m o d el is rejected because it appears not to h o ld fo r m ost o f the cou ntries exam in ed. A n o th er in terpretation is that institutional o r o th er co n sid erations, w h ich the pu re theory regards eith er as given o r unim portant, m ay have had adverse ef fects on the em pirical tests. A m o n g others, im p o r tant structural changes that w ill affect the results in clu de the co n d u ct o f U.S. m on etary policy, changes in interest rate ceilin gs and general fin an cial market deregu lation. Given the results sh ow n in table 1, previous research gen erally has left these results u n explain ed o r has a d d ed som e ad hoc m easure o f risk to accou nt for the possible effects o f an unobservable term prem iu m . In the sections that follow , w e first revise the estim ation p roced u re to see h o w this change affects the test results. W e then discuss som e w ell-d efin ed events and changes in institutions that co u ld affect the term structure relations and p ro d u ce the results that a ppear to reject the m od el. ONE POSSIBLE REASON FOR REJECTION OF THE EXPECTATIONS MODEL: CORRELATED ERROR TERMS T h e in creasin g integration o f w o rld capital m ar kets suggests that an alternative statistical a p proach sh ou ld be u sed to estim ate equ ation 1. As capital flow s freely am on g nations, m on etary p o l power- fo r the equ ations is gen erally lo w (w ith the notable excep tio n o f Japan).8This result is typical icy actions (for exam ple) undertaken in on e co u n try can be ex p ected to affect financial variables in o th er countries as w ell. Consider, fo r exam ple, a change in Bundesbank p o lic y that affects Germ an interest rates and then is transm itted to interest rates in the o th er fou r nations via capital flow s cau sed b y the change in Germ an interest rates. Th is effect, w h ic h w ill a p p ea r o n ly in the error term o f the Germ an interest rate equ ation w h en separ ate regressions are estim ated, co u ld be ex p lo ite d as a n e w source o f in form ation fo r each in estim ates o f the expectation s m odel, in dicating regression if the cou n try equations w ere estim ated 6ln fact, the U.S. results are extraordinarily sensitive to these few data points. The dramatic increase in interest rate volatility during the first and second quarters of 1980, relative to the remaining sample, would suggest this sensitivity in OLS re gression estimates. account for the effects of the third-order serial correlation, see Hansen and Hodrick (1980) and Campbell and Clarida (1987). BASIC RESULTS T h e results from estim ating equ ation 1 are re p o rted in table 1. T h e expectation s m o d el is clearly rejected for the U nited States, G erm any and Sw itzerland; their estim ated slope coefficients are significantly differen t from one. In contrast, the results fo r the U nited K in gd o m and Japan support the expectation s m odel. Explanatory 8Durbin-Watson statistics are not reported because, as indi cated in the text, the reported standard errors reflect correc tions for serial correlation in the data. 7For an extensive description of the econometrics used to SEPTEMBER/OCTOBER 1988 40 in form ation exists in the erro r term s is substanti ated by the co m p u ted value o f 56.34 fo r a likeli h ood ratio statistic testing w h e th er covariances Table 2 Revised Expectations Model Estimates Using Seemingly Unrelated Regressions (SUR): 1981.02 -1986.10 Country United States United Kingdom Germany Japan Switzerland a b -0 .5 2 (2.26) 0.00 (0.02) -0 .4 7 (5.62) -0 .1 8 (3.07) -0 .3 9 (2.37) 0.20* (2.60) 0.93 (0.20) 0.45* (4.15) 1.07 (0.42) 0.49* (2.21) am on g the erro r term s are zero: this value is to be com p a red w ith the 5 percen t critical value o f 18.30. T h e erro r covariance and correlation m atrices rep orted in table 3 in dicate w h ere the significant correlation s b etw een countries w ere found. Note, in particular, the high correlations betw een the U.S. and G erm any and b etw een G erm any and Sw itzerland. C on jectu res to explain these correlations and, possibly, m o d el rejection s are discussed later in referen ce to the table 4 results. A lthou gh OLS and SUR sh ou ld p rod u ce sim ilar coefficient estimates, both the U.S. and Swiss NOTE: Absolute values of t-statistics are in parentheses. For b, the t-statistic applies to the null hypothesis b = 1. An asterisk indicates a slope coefficient significantly different from one at the 0.05 level of significance. F-test for null hypothesis: bus = bUK = bGER = bj = bsw = 1 is 5.63 compared with a critical value of 2.21. jointly. In o th er w ords, the error term o f a single equation (w hich reflects ' n ew s,” o r u npredictable events w ith in that country) also m ay contain in for m ation — du e to linkages am on g markets — that is relevant to explainin g interest rate beh avior in an oth er country. T h e im portan t p oin t is that the expectation s m o d el bein g tested assumes that this in form ation is b ein g u sed b y the rational agents w h o se co llective actions determ in e changes in interest rates. Single equ ation estim ates, how ever, exclu de the in form ation im p licit in these linkages because they look at data for each cou ntry in iso lation. One w a y to accou nt fo r this m issing in form ation is to estim ate equ ation 1, as a p p lied to the live countries u n d er study, as a system o f seem in gly u nrelated regressions (SUR).9Th is p roced u re co n siders con tem poran eou s correlation s that m ight exist am on g the error term s o f the five equations and, by d o in g so, im proves the efficien cy w ith w h ich the coefficients are estim ated. The SUR Results T h e results from estim ating the five equations bv SUR are rep orted in table 2. That im portant 9Edwards (1982) has made the same point and reported muchimproved results for a similar model applied to the exchange rate. Krol (1987) also reported substantial integration of these markets across countries. Mankiw (1986), however, finds little FEDERAL RESERVE BANK OF ST. LOUIS slope coefficien ts re p o rted in table 2 are m arkedly differen t from th eir values in table 1. In v ie w o f the lo w values fo r R- in both the U.S. and Swiss equ a tions, how ever, these changes m erely indicate that, for these data, the basic specification o f the expectation s m o d el sim p ly d oes not p rod u ce p re cise estim ates o f the slope coefficient. T h e m ore im portant p o in t is that, after using the SUR estim a tor, the h ypothesis that all five slope coefficients are jointly equal to o n e still is rejected. Finally, the Japanese intercept, w h ic h d id not change n u m eri cally, n o w is significantly d ifferen t from zero. Be cause the Germ an and U.K. results are largely u n affected by the SUR estim ation, how ever, this sim ple change in estim ation p roced u re to in c o r porate linkages a m o n g financial markets, w h ile in dicatin g that significant in form ation exists in the correlation s am on g erro r term s across equations, still rejects the expectation s m odel fo r m ost o f the countries exam ined. OTHER SOURCES OF EXPLOITABLE INFORMATION A n o th er assu m ption b eh in d em p irica l tests o f the expectation s m o d e l is that the data u sed for estim ation w ere gen erated du rin g a p erio d ch ar a cterized b y a stable eco n o m ic structure. M o re over, the data sh ou ld be draw n from markets in w h ic h interest rates can adjust freely. Thus, the basic m o d el shou ld not be estim ated w ith data from period s associated w ith m a jor p o licy correlation across countries and speculates that capital con trols may “ prevent effective international arbitrage (p. 66)” . See Zellner (1962) for details on the estimation procedure. 41 Table 3 Error Correlation and Covariance Matrices From The SUR Estimation Country United States United Kingdom Germany Japan Switzerland Covariance Across Models United States United Kingdom Germany Japan Switzerland 2.49 0.43 0.44 0.06 0.14 United States United Kingdom Germany Japan Switzerland 1.00 0.19 0.43* 0.08 0.08 2.03 0.25 0.43 0.07 0.11 0.49 0.30 Correlation Matrix 1.00 0.27* 0.10 0.29* 1.00 0.33* 0.38* 0.23 0.00 1.00 -0 .0 1 5 percent significance level for correlation is 0.25. For the null hypothesis that the off-diagonal elements of the covariance matrix are zero, the likelihood ratio statistic is 56.34 vs. a 5 percent critical value of 18.30. Table 4 Revised Expectations Model SUR Estimates Country United States United Kingdom Germany Japan Switzerland a b MTARGET -0 .5 6 (2.23) 0.00 (0.02) -0 .4 7 (5.50) -0 .1 8 (3.07) -0 .4 0 (2.42) 0.45 (1.08) 0.94 (0.19) 0.47 (3.83)* 1.07 (0.42) 0.58 (1.60) -0 .4 9 (0.88) — — — — — — — — EMS — — — — -0 .5 6 (0.72) — — -0 .6 2 (1.27) NOTE: Absolute values of t-statistics are in parentheses. For b, t-statistic applies to the null hypothesis b = 1. An asterisk indicates a slope coefficient significantly different from one at the 0.05 level of significance. F-test for null hypothesis: bus = bUK = bGER = bj = bsw = 1 is 3.55 versus a critical value of 2.21. changes o r im p ed im en ts to m arket adjustm ents. In the case o f the form er, m ajor p o lic y changes cu rred du rin g the p erio d used for estim ation and assess h o w th ey affect the results re p o rted above. m ay cause large discrete changes in expectation s or changes in the variability o f expectation s that can not b e m easured o r m o d elle d p rop erly. Simi larly, taking data from , say, a p erio d ch a racterized by interest rate controls w o u ld be in ap propriate fo r testing the m o d e l becau se th eory assumes that interest rates can adjust freely in p erfectly c o m petitive, efficien t markets. In w h at fo llo w s below , w e describe som e m ajor changes that have o c Changes in U.S. Monetary Policy Since O ctober 1979, the Federal Reserve has used tw o distinct operational p roced u res in its con d u ct o f m on etary p olicy. B etw een O ctober 1979 and O cto b er 1982, the Fed established a tar ge te d path fo r n o n b o rro w ed reserves; this a p proach p erm itted short-term interest rates to fluc- SEPTEMBER/OCTOBER 1988 42 Chart 1 Changes in Federal Funds Rate Percent 4 Percent 4 1977 tuate w ith in w id e r bands than had the previous procedu re, w h ic h had fo cu sed on k eep in g the federal funds rate w ith in a n arrow range. In O cto ber 1982, the Federal Reserve an n ou n ced that, due to increasing u ncertainties about the defin ition o f the M l aggregate, it w o u ld co n d u ct m on etary p o lic y b y setting an objective fo r b o rro w e d re serves; this latter strategy resu lted in less variation in short-term interest rates.10Thus, the first part o f the sam ple p erio d used in the estim ation is char a cterized b y a Fed op era tin g p roced u re that p e r m itted greater variation in short-term interest rates; this p erio d is fo llo w ed bv fou r years o f data associated w ith a p roced u re that, o n ce again, re d u ced the variation in short-term interest rates. T h e beh avior o f the fed eral funds rate, w h ic h su p ports this d ep ictio n o f events, is sh ow n in chart 1. H o w w o u ld this sw itch in p o licy im p lem en ta tion affect tests o f the expectation s m odel? A c ,0See Wallich (1984) and Gilbert (1985) for more discussion about changes in the implementation of U.S. monetary policy over time. FEDERAL RESERVE BANK OF ST. LOUIS 1988 co rd in g to M an kiw and M iron (1986), Fed p o lic y based o n sm ooth in g short-term interest rates can be ch a racterized as: (2) E, (Ar,+1) = 0 or, the exp ected change in the short-rate at each m om en t in tim e is ze ro even if the Fed has been observed to change short rates in response to, say, real GNP g row th o r inflation rates that deviated from p rio r expectations. If equ ation 2 describes Fed p o lic y since O cto b er 1982 (and p rio r to O cto b er 1979), the value o f (,F11+1 — r ,,) in equ ation 1 w ill always be zero and short-term interest rates w ill behave, approxim ately, as a ran dom walk. In this case, the expectation s m o d el o f the term structure w o u ld be in capable o f explain in g the b eh avior o f short-term interest rates. M an kiw and M iron (1986) investigated this p ro b lem using annual U.S. data from 1890-1914 and 43 1915-79. T h e y fou n d that su pport fo r the e x p e c tations m o d el varied w ith m o n eta iy regim e. W h ile the expectation s m o d el “ h o ld s” fo r the pre-Fed period, w h e n there was n o m onetary authority to sm ooth interest rates, the m o d el is rejected fo r the later p erio d w h en the F e d ’s approach to p o licy ten d ed to sm ooth fluctuations in short-term in ter est rates. T h e ir results, therefore, suggest that the U.S. results rep orted in table 2 — and perhaps o th er rejection s o f the m o d el using post-1979 U.S. data — co u ld be d o m in a ted b y the sub-sam ple associated w ith the post-O ctob er 1982 change in Federal Reserve operatin g procedu res. Effects o f Exchange Rate Intervention Rules T h e fo u n d in g o f the European M o n e ta iy System (EMS) is an oth er im portan t change that occu rred in 1979 and is a possible source o f the negative results for G erm any and Sw itzerland. T h e EMS agreem ent established ranges fo r bilateral ex change rates o f the m em b er countries and called fo r coop era tive interventions b y the central banks o f the countries in volved w h e n rates deviated from their s p ecified ranges. Thus, German m on etary p o lic y since 1979 has been constrained b v its p a r ticipation in the exch an ge rate agreem ent and its p led ge to in terven e." In practice, G erm any has b ec o m e the lea d er o f the EMS due to the size o f its e c o n o m y and its lo w inflation rate; oth er EMS countries have fo llo w e d its n on in flation a iy m o n e ta iy policy. M u ch research lias sh ow n that the EMS agreem ent really has beh aved as if a dollar/ DM o bjective w ere pu rsued b y the German central bank.12 In addition, Swiss m o n eta iy p o licy is in flu en ced by the DM/Swiss franc exchange rate even though S w itzerlan d is not an EMS m em ber.13 Because standard m od els typically explain the beh avior o f the exch an ge rate as d ep en d in g o n the spread b etw een foreign and d om estic interest rates, at tem pts b y the Bundesbank to in flu en ce the dollar/ DM exch an ge rate also w o u ld create a strong link ,1The history of the EMS and a discussion of how it functions can be found in lingerer, et al. (1986). ,2See, for example, Fels (1987) for a discussion of the EMS as a dollar/DM commitment by the Bundesbank. "Because trade represents 39 percent of Swiss GDP and trade with Germany accounts for one-fifth of total trade, the Swiss franc/DM exchange rate has been particularly important to the conduct of Swiss monetary policy. The Swiss National Bank, at times, has abandoned its objectives for the growth rate of the monetary base and, instead, pursued an exchange rate objec tive. See Rich and Beguelin (1985). betw een Germ an and Swiss interest rates.14Sup pose, for exam ple, that the d olla r w ere d ep recia t in g against the DM because U.S. interest rates w ere falling. T h e Bundesbank co u ld attem pt to stop or reverse this dolla r d ep recia tion bv e x p a n d in g the Germ an m o n ey stock and lo w e rin g G erm an sh ort term interest rates. Such an action, h ow ever, w o u ld cause the value o f the Swiss franc to rise against tin; DM. In the past, the Swiss N ational Bank has resp o n d e d to this (or sim ilar) sequ en ce o f events by fo llo w in g the Bundesbank w ith a m ore expansion ary m on etary p o lic y a nd lo w e r short-term interest rates as it attem p ted to re establish som e d esired value for the DM/Swiss franc exch an ge rate. This close linkage o f German and Swiss interest rates, from a Swiss objective fo r stability o f the bilateral exch an ge rate, is likely to be the source o f the h igh ly correlated Swiss and Germ an erro r term s rep orted in table 3.15In sum, both Germ an and Swiss m o n e ta iy p olicies are in flu en ced b y exch an ge rate considerations that cou ld affect em pirical estim ates o f the ex p ecta tions m odel. Empirical Implementation T o investigate these possibilities, the system o f SIIR equations rep orted in table 2 w as reestim ated w ith changes in the U.S., German and Swiss regressions. For the U.S., the w h o le-sa m p le slope coefficien t w as split to represent the tw o distinct period s o f Federal Reserve operatin g p r o cedures. A slop e du m m y (M TARGET) w as in tro duced, w h ic h took a value o f on e b etw een Febru ary 1981 and S eptem ber 1982 and a value o f zero fo r the rem ainin g m onths. If the M an kiw -M iron h ypothesis is correct, the slope coefficien t fo r the first part o f the sam ple (b plus M TARGET) should not be significantly d ifferen t from on e w h ile the coefficien t fo r the latter p erio d (b alone) sh ou ld be significantly d ifferen t (less than) from o n e.16 A lthou gh the p recise w a y to quantify the im pact o f the EMS agreem en t on Germ an and Swiss finan cial markets is not clear, the p eriod s w h en the 15A related point that suggests this sort of influence across countries is based on results from Belongia and Ott (1988). They show that the dollar exchange rate risk premium and the amount that the exchange rate adjusts to a given domesticforeign interest differential both vary with the choice of Federal Reserve operating procedure (interest rate vs. money stock objectives). If nothing else, their result would be suggestive of a time varying risk premium in the expectations model. ,6An intercept dummy also was tried but it was not significant individually and had no material effects on the magnitudes or significance of other coefficients. '“See, for example, the model presented by Dornbusch (1980). SEPTEMBER/OCTOBER 1988 44 m em b er countries agreed to m a jo r realignm ents o f the official exchange rate levels and ranges are know n. O ther things the same, on e can h yp o th e size that interest rates m ade discrete adjustm ents to these realignm ents w ith in one m onth after they w ere ann ou nced. T o test the p rop o sitio n about exchange rate linkages and interest rates, a du m m y variable w as created to represent EMS realignm ents and w as in trod u ced into both the German and Swiss regressions. This variable took a value o f o n e du rin g the m onths associated w ith the eight EMS realignm ents and a value o f zero du rin g all o th er m on th s.17As w ith the U.S. case, m u ltiplyin g the forw ard rate — spot rate spread in the German and Swiss regressions b y this du m m y variable perm its the estim ation o f tw o different values fo r the regressions’ slop e coefficients: o n e coefficient fo r “ n orm a l” period s and the sum o f tw o coefficients fo r m onths w h e n a realignm ent occu rred. In table 4, the revised SUR results are reported. Th e null h ypothesis that all five slope coefficients are jo in tly equal to on e is rejected, o n ce again, at the 0.05 level o f significance. T h e expectation s m o d el is re jected even after a u gm entin g the in for m ation set to in corporate changes in the im p le m entation o f U.S. m on etary p o lic y and the EMS realignm ents. Lookin g at in dividu al cou n try results, the ta b le’s top row , associated w ith the slope d u m m y fo r the p erio d o f m on etary targeting in the U nited States, in dicates that estim ates o f the expectation s m o d el are sensitive to changes in the F e d ’s operatin g p roced u re. Even thou gh the M TARG E T d u m m y is not significant, the m o d e l’s w h o le-p e rio d slope coefficien t increases from 0.20 to 0.45 and n o w is n ot significantly d ifferen t from one. This apparent im p rovem en t in the U.S. results, how ever, is in direct contrast to M an kiw and M iro n ’s results in tw o respects. First, w h e n they attem pted to investigate the effects o f post-1979 data on the expectation s m odel, th ey rep orted that “w e obtain standard errors so large that one can reject no interesting h yp o th esis” (p. 227). M ore im portant, th ey h yp o th es ize d that the ex pectations m o d el sh ou ld not be rejected fo r the p erio d o f m o n ey stock targeting, but sh ou ld be rejected fo r the post-Septem ber 1982 period ; e m pirically, this im p lies that b plus M TARG E T shou ld ’T h e dates of EMS realignments were March 23 and October 5, 1981; February 22 and June 14,1982; March 21,1983; July 22, 1985; April 7 and August 4, 1986 and are provided in Fels, p. 217. FEDERAL RESERVE BANK OF ST. LOUIS not be statistically different from o n e w h ile b alone sh ou ld be significantly differen t from (less than) one. In fact, the results are reversed; the ex p ecta tions m o d el is rejected fo r the p eriod o f m o n ey stock targeting. Thus, w h ile the du m m y variable im proves the overall results and provides perhaps a stron ger test o f th eir m odel, the exact process at w ork is in con sisten t w ith the o n e h ypoth esized, leaving an u n explain ed p u zzle. T h e revised estim ates fo r the Germ an and Swiss equations p rovid e w eak su pport fo r the con jectu re that the intervention p o licies o f th eir central banks have significant effects o n tests o f the expectation s m odel. T h e signs on the slope d u m m ies are nega tive and sim ilar in m agnitude, to the w h o le p erio d slope coefficient, w h ich in dicates that the fo rw ard rate-spot rate spread has zero effect during m onths o f EMS realignm ents. M oreover, the w h o le p e rio d Swiss slope coefficien t n o w both is larger n u m erically and n ot significantly d ifferen t from one. For Germany, h ow ever, the results are not altered w h e n the dates o f EMS realignm ents are co n sid ered and the data contin u e to reject the expectation s m odel. CONCLUSIONS Th e expectation s m o d el o f the term structure o f interest rates has been a p p lied to data fo r a n u m ber o f countries and sam ple p eriod s w ith g e n er ally n egative results. In this article w e have investi gated som e con d ition s u n d er w h ic h the expectation s m o d el m ight be rejected in the c o n text o f its traditional single equ ation test. W e fou n d substantial correlation s across the errors o f the in dividu al equ ation s w hich, w h e n e x p lo ite d by using SUR estim ation, im p ro ved the efficien cy o f estim ation. W e also fo u n d that, although du m m y variables used to represent changes in the a p proach to m on etary p o lic y o r EMS exch an ge rate targets w ere n ot significant individually, th ey c o n tribu ted som ew h at to im p ro ved overall character istics o f the equations. A lthough, as in previous studies, m any p u zzles still remain, these results suggest that tests o f the expectation s m o d el sh ou ld use m o re gen eral m o d els and m o re ef ficient estim ation p roced u res than the sim ple OLS equation typically e m p loyed . 45 REFERENCES Belongia, Michael T., and Mack Ott. 1The U.S. Monetary Policy Regime, Interest Differentials and Dollar Exchange Rate Risk Premia,” Journal of International Money and Finance, (Decem ber 1988), forthcoming. Bisignano, Joseph R. “A Study of Efficiency and Volatility in Government Securities Markets." Bank for International Settlements, processed June 1987. Campbell, John Y., and Richard H. Clarida. "The Term Struc ture of Euromarket Interest Rates: An Empirical Investiga tion,” Journal of Monetary Economics (January 1987), pp. 25-44. Cosset, Jean-Claude. “ Forward Rates as Predictors of Future Interest Rates in the Eurocurrency Market,” Journal of Interna tional Business Studies (Winter 1982), pp. 71-83. Dornbusch, Rudiger. “ Exchange Rate Economics: Where Do We Stand?,” Brookings Papers on Economic Activity (1980:1), pp. 145-85. Edwards, Sebastian. “ Exchange Rates and News': A MultiCurrency Approach," Journal of International Money and Finance (December 1982), pp. 211-24. Fels, Joachim. “The European Monetary System 1979-87: Why Has It Worked?," Intereconomics (September/October 1987), pp. 216-22. Gilbert, R. Alton. “ Operating Procedures for Conducting Mone tary Policy," this Review (February 1985), pp. 13-21. Hakkio, Craig S., and Leonardo Leiderman. “ Intertemporal Asset Pricing and the Term Structure of Exchange Rates and Interest Rates: The Eurocurrency Market,” European Eco nomic Review (April 1986), pp. 325-44. Hansen, Lars P., and Robert J. Hodrick. “ Forward Exchange Rates as Optimal Predictors of Future Spot Rates: An Econo metric Analysis,” Journal of Political Economy (October 1980), pp. 829-53. Krol, Robert. “The Interdependence of the Term Structure of Eurocurrency Interest Rates,” Journal of International Money and Finance (June 1986), pp. 245-53. Mankiw, N. Gregory, and Jeffrey A. Miron. “ The Changing Behavior of the Term Structure of Interest Rates,” Quarterly Journal of Economics (May 1986), pp. 211-28. Mankiw, N. Gregory. “The Term Structure of Interest Rates Revisited,” Brookings Papers on Economic Activity (1:1986), pp. 61-96. Rich, Georg, and Jean-Pierre Beguelin. “ Swiss Monetary Policy in the 1970s and 1980s: An Experiment in Pragmatic Monetarism,” in Monetary Policy and Monetary Regimes, Center Symposium Series, CS-17, Karl Brunner, ed., Center for Research in Government Policy and Business, University of Rochester (1985), pp. 76-111. Shiller, Robert J., John Y. Campbell, and Kermit L. Schoenholtz. “ Forward Rates and Future Policy: Interpret ing the Term Structure of Interest Rates,” Brookings Papers on Economic Activity (1:1983), pp. 173-217. Wallich, Henry C. “ Recent Techniques of Monetary Policy," Federal Reserve Bank of Kansas City Economic Review, (May 1984), pp. 21-30. Ungerer, Horst, Owen Evans, Thomas Mayer, and Philip Young. The European Monetary System: Recent Develop ments, Occasional Paper No. 48 (International Monetary Fund, December 1986). Zellner, Arnold. “An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias,” Journal of The American Statistical Association (June 1962), pp. 348-68. SEPTEMBER/OCTOBER 1988 46 Albert E. Burger Albert E. Burger is a vice president at the Federal Reserve Bank of St. Louis. Laura A. Prives provided research assistance. The Puzzling Growth of the Monetary Aggregates in the 1980s M l w i O D E R N m a croecon o m ic analysis assigns the key role in aggregate d em an d m an agem ent to m on etary policy. This role is carried out through changes in the m o n e ta iy aggregates. Since there are several m on etary aggregates — M l, M2 and M3 — con siderable confu sion m ay d evelop about the m eaning o f th eir behavior, particularly w h e n they d o not m ove in lock step w ith each o th er o r w ith the grow th o f the m o n eta iy base. Such confusion is esp ecially likely to h ap pen w hen, as has h ap p en ed in the 1980s, th eir m ovem en ts are quite unusual bv historical standards. In the 1980s, these relationships changed quite dram atically. From 1984 through 1987, the m o n e ta iy base grow th averaged about 6 p ercen t to 8 percent. In sharp contrast to its previous historical relationship, M l grow th averaged 7 p ercen t to 12 percent; in 1986 alone, M l g re w 4 percen tage points faster than the base. M ean w hile, the grow th rates o f M2 and M3 d ec lin ed relative to the grow th o f the base: in 1986, th ey fell b e lo w base grow th, and in 1987, base grow th ex ceed ed the grow th o f M2 and M3 b y m ore than 2 p ercen tage points. T h e m onetary base can be thought o f as the fou ndation on w h ic h all the m o n eta iy aggregates are built; it is also the set o f m on etary assets m ost M a jor shifts in the p u b lic’s h oldin gs o f m o n eta iy assets have a ccou n ted for these ch anged relation ships. This article describes a fram ew ork that both incorporates the relative am ounts o f different closely related to Federal Reserve actions. Prior to the early 1980s, there w as a fairly stable relation m on etary assets the pu blic desires to h old and relates the grow th o f M l, M2 and M3 to the m o n e ship on an annual basis b etw een the grow th rate ta iy base. This fram ew ork is then used to analyze o f the m on etary base and the grow th rates o f M l, the unusual m ovem en ts o f these aggregates d u r M2 and M3. Th e m on etary base g re w about 1 p er ing the past fe w years. centage point faster than M l; and the o th er tw o aggregates, M2 and M3, g r e w about 2 o r 3 p ercen t age points faster than the m on etary base. Thus, w h en Federal Reseive actions resulted in a 6 p e r SOURCES AND USES OF THE MONETARY BASE cent annual grow th rate o f the m on etary base, M l w o u ld g ro w at about 5 percent, M2 at 8 percen t and M3 at about 9 percent. FEDERAL RESERVE BANK OF ST. LOUIS T h e m o n eta iy base is essentially d erived from the Federal Reserve’s balance sheet and can be 47 Table 1 Components of the Monetary Base: December 1987 (billions of dollars, not seasonally adjusted) Sources Federal Reserve holdings of government securities Federal Reserve loans Float plus other Federal Reserve assets Other items' Source base Reserve adjustment2 Monetary base Uses $227.8 0.8 17.3 19.4 265.0 7.7 272.8 Depository institution deposits at Federal Reserve banks Currency held by depository institutions Currency held by nonbank public $ 37.7 30.9 196.5 Source base Reserve adjustment2 Monetary base 265.0 7.7 272.8 'Other items include: Treasury deposits at Federal Reserve Banks, special drawing rights, Treasury currency outstanding, Treasury cash holdings, foreign and other deposits with Federal Reserve Banks, service-related balances and adjustments, and other Federal Reserve liabilities and capital. Adjustment for reserve requirement ratio changes. co m p u ted eith er from the sources side — the item s that su pply base — o r from the uses side — the item s that absorb base.1As table 1 shows, the m a jor source o f the m on etary base is Federal Re serve h oldin gs o f govern m en t securities. Changes in this item reflect the F e d ’s o p en market op era tions; during the last 10 years, it has accou n ted for about 80 percen t o f the total change and m ost o f the vear-to-vear fluctuations in the base. W h en the Federal Reserve makes an o p en m ar ket purchase o f govern m en t securities, o th er fac tors the same, m ore m o n eta iy base is su p p lied to the financial sector and the public. This increase in the base is then "u s e d ” by the pu blic and d e pository institutions as additions to th eir h oldin gs o f cu rren cy and reserves. Th e increase in reserves form s the base from w h ic h to expan d derivative m o n e ta iy assets created b y financial institutions. Because the pu blic chooses the relative p ro p o r tions o f these types o f assets th ey w an t to hold, it determ in es the relationship betw een the grow th o f the base and the resulting grow th o f the various m o n e ta iy aggregates. 'For a discussion of the concept and derivation of the monetary base, see Burger and Balbach (1976). There are two available measures of the monetary base, one published by the Federal Reserve Board and the other by the Federal Reserve Bank of St. Louis. The Board’s measure is a “ uses” concept and the Federal Reserve Bank of St. Louis’ is a “ sources” concept. The major difference is that the St. Louis Fed treats all vault cash contemporaneously while the Board lags the vault cash com ponent of total reserves, reflecting its treatment as total re THE LINK BETWEEN THE MONETARY BASE AND THE MONETARY AGGREGATES T h e relationship b etw een the m o n e ta iy base and any m o n e ta iy aggregate can be expressed in the fo llo w in g manner: M = mB. T h e m o n e ta iy base (B) is related to the specified m o n e ta iy aggregate (M) by a m o n ey m u ltip lier (m). Given the m o n e ta iy base, the m u ltip lier sum m a rizes the effect o f po rtfo lio decision s b y the public and financial institutions on a m o n e ta iy aggregate. In term s o f gro w th rates, this expression can be w ritten : iVI = ill + B, w h ere the dot above each item den otes its grow th rate. If the m o n ey m u ltipliers w ere constant over time, then the grow th rates o f the m o n e ta iy aggre gates w o u ld fo llo w the sam e pattern as the grow th serves. In analyzing periods of two or more quarters, the differ ences in results between the two base concepts is very small. For a further discussion of these measures, see Burger (1979) The source base is usually “ adjusted” to incorporate the influence of reserve requirement changes into movements in the adjusted monetary base. For a discussion of this adjust ment, see Burger and Rasche (1976), Burger (1979) and, for the most recent method of calculating this adjustment, Gilbert (1987). SEPTEMBER/OCTOBER 1988 48 Chart 1 M1 Multiplier 1960 1965 1970 1975 1980 1985 1990 o f the m on etary base, and all aggregates w o u ld g r o w together. b etw een the grow th o f the m on etary base and M l. As the next section shows, how ever, these m u lti pliers have not been constant. Consequently, al thou gh the grow th rates o f M l and the m on etary base have b een h igh ly correlated, there have still been period s such as 1974-76 and 1985-87 w h en they diverged substantially. Th e grow th rates o f M2 and M3 have been less closely tied to the grow th o f the m on etary base and, although both For the rem a in d er o f the 1970s, the M l m u lti p lier d eclin e slo w ed to about its 1962-73 pace. Then, about mid-1980, the M l m u ltip lier flattened out and sh o w ed little grow th on average until early 1985, w h e n its beh avior ch anged m arkedly. It rose at a 1.7 percen t annual rate in 1985; in 1986 its grow th in creased to 4 percent. T h e M l m u ltip lier d eclin ed som ew h at in mid-1987; h ow ever, w h e n have been h igh ly correlated, th ey have frequ ently diverged from the grow th o f M l. EXAMINING THE BEHAVIOR OF THE MULTIPLIERS As chart 1 shows, from the early 1960s through the 1970s, there w as a long-run d o w n w a rd trend in the M l m u ltiplier. Th e m u ltip lier d rifted lo w e r from the early 1960s through 1973, d eclin in g at about a 1 percen t annual rate. D uring the next three years, it fell faster at about a 3 p ercen t an FEDERAL RESERVE BANK OF ST. LOUIS nual rate. This w as reflected in a w id e n in g spread m easured on an annual basis, it still rose another 2 percen t in 1987. As chart 1 indicates, this p ro lo n g ed and substantial rise w as w ith ou t preced en t since the early 1960s. Chart 2 show s that the M2 and M3 m o n ey m u lti pliers have fo llo w e d v ery different paths. T h ey gen erally rose fo r m ost o f the p e rio d since the early 1960s, w h ile the M l m u ltip lier w as falling. In the last fe w years, w h ile the M l m u ltip lier has been rising, how ever, the M2 and M3 m u ltipliers have fallen. D uring the p e rio d sh ow n in chart 2, three broad grow th patterns em erge in the M2 and 49 Chart 2 M2 and M3 Multipliers Ratio 15 Ratio 15 Annual Data 14 14 13 13 M3 Multiplier 12 12 / 11 11 s ' 10 10 '' M2 Multiplier I ~ 11 1 1 1 1 II1 1 w 1960 ^ 1965 1970 M3 m ultipliers. From the early 1960s through early 1982, they in creased on average at about a 2 p e r cent rate. In early 1983, th ey cam e to a halt, and for the next tw o years, they sh ow ed essentially no grow th . In early 1986, how ever-, the M2 and M3 m u ltipliers began a d eclin e that has lasted into 1988. A Model o f the Money Multipliers Th e substantial break in the usual beh avior o f the m o n ey m u ltipliers in the 1980s w as reflected in the unusual beh avior o f the m on etary aggre gates relative to the grow th o f the m on etary base, and to each other. T o exam in e w h y this was the case, on e m ust d ev elo p explicit form s o f the re spective m u ltipliers to analyze h o w the ch anging po rtfo lio preferen ces o f the pu blic have affected them. 1975 1985 1980 1990 m3 = M3/B, R t3 = reserves o f d ep o sito ry institutions adjusted fo r reserve requ irem en t changes, = cu rren cy h eld by the public, = m on etary base = R + C, = checkable deposits, = the public's desired cu rren cy ratio = C/D, = the p u b lic’s desired nontransactions bal ance ratio = IM2 — M l I/D, = IM 3-M 2I/ D , and r = reserve ratio = R/D, C B D k t2 the fo llo w in g explicit form s o f the m u ltipliers can be d erived (see a p p en d ix 1 fo r this derivation): ml = i ± k r+ k m , = 1 + k + t2 r+ k m3 = 1 + k + 13 r+ k In this fram ew ork, a distin ction can be m ade Given the fo llo w in g definitions, am ong three m a jor classes o f assets. As table 2 shows, M l represents transaction balances, m l = Ml/13, (M2 — M l) represents liqu id savings balances, and m2 = M2/B, (M3 — M2) represents m an aged liabilities o f de- SEPTEMBER/OCTOBER 1988 50 ratio exert a dom in ant in flu ence on m ovem en ts in Table 2 Components of the Monetary Aggregates: December 1987 (not seasonally adjusted)________________ Monthly Average M1 M2-M1 M3-M2 J Currency ( Total checkable deposits 1 Savings deposits Small time deposits MMDA Money market mutual funds , Overnight RP and Eurodollars Large time deposits Term RP and Eurodollars I Institution-only MMMF $199.4 560.1 410.0 914.6 525.2 221.1 78.1 485.4 196.3 89.6 pository financial institutions. W h en eith er the specific characteristics or the relative yield s o f these assets change, the pu blic respon ds b y alter ing the am ounts o f these assets th ey w ish to hold. T h e k, t2 and t3 ratios capture the effects o f the pu blic's shifting p referen ces am on g these assets on the g row th rates o f M l, M2 and M3. A rise in the r-ratio reflects an increase in d ep o sito ry in sti tu tion s’ d esired h oldin gs o f reserves relative to deposits; h ence, a rise in this ratio redu ces all three m ultipliers. Given this fram ew ork, w e can n o w exam in e the beh avior o f these ratios and d eterm in e th eir c o n tribution to the m o n ey m u ltip lier m ovem ents, esp ecially in recent years. The Currency Ratio A rise in the k-ratio reflects an increase in the p u b lic’s desired h old in gs o f cu rren cy relative to checkable deposits. For a given am ount o f m o n e tary base, this m eans a redu ction in the portion o f base held by d e p o s ito iv institutions (reserves) and, consequently, a redu ction in checkable deposits. Th erefore, a rise in the k-ratio redu ces all three m o n ey m ultipliers. It has been lo n g re co g n ized that, given the grow th o f the m on etary base, variations in the k- 2See Cagan (1958). 3See Gutmann (1977). 4See Garcia (1978) and Dotsey (1988). FEDERAL RESERVE BANK OF ST. LOUIS M l and a strong in flu ence on m ovem en ts in oth er m o n eta iy aggregates.- As chart 3 illustrates, m o ve m ents in the M l m u ltip lier are essentially the m irror im age o f m ovem en ts in the k-ratio. Thus, deviations o f M l grow th from base grow th are pred om in an tly due to sharp changes in the grow th o f the cu rren cy ratio (the quantitative ef fects o f these changes are d erived in a p p en d ix II). Chart 3 show s that the cu rren cy ratio in creased from the early 1960s until the early 1980s. On an annual basis, the k-ratio sh o w ed no n oticeable declin e in this 21-year p eriod ; in deed, there w ere fe w years w h e n it d id not increase b y at least 1 percent. D uring the early 1980s, the cu rren cy ratio sh ow ed little grow th . Then, in early 1985, instead o f the public in creasin g its cu rren cy h oldin gs rela tive to checkable deposits, as had been its lo n g term pattern, the pu blic began to do just the o p posite. C onsequently, there w as a m ajor change in the beh avior o f the k-ratio. D uring 1985, the k-ratio fell 2.8 percent; in 1986, it d ec lin ed 7.7 percent; and, in 1987, it d ro p p e d an oth er 4.1 percent. Studies in dicate that m a jor changes in the grow th o f the k-ratio are related prim arily to fac tors that affect the checkable dep osit co m p o n en t o f this ratio. A lthou gh attem pts have been m ade to trace the rise in the k-ratio in the 1970s to a sharp increase in cu rren cy d em an d alon g w ith the rise o f the "u n d ergrou n d eco n o m y ,”3 cu rren cy d e m an d has b een fou n d to be stable over lo n g p e ri ods o f tim e.4 T h e am ou nt o f transaction balances that in d i viduals and firms desire to h old relative to oth er assets is in flu en ced by such factors as current and e x p ected rates o f inflation, relative yield s on oth er assets and available alternative assets. In the 1970s, inflation accelerated, interest rates rose, n ew form s o f savings accounts w ere o ffered to the public and n ew cash m an agem ent techniques becam e available to business. Unlike the dem an d for currency, the d em a n d fo r checkable deposits was substantially affected by these developm en ts, particularly the financial innovations. For exam ple, business h oldin gs o f transaction balances relative to financial assets d ec lin ed from about 74 p ercen t in 1970 to about 38 p ercen t in 1981. This d eclin e w as m ost closely related to the rise o f cash 51 Chart 3 Currency Ratio and M1 Multiplier k-ratio M1 Multiplier Annual Data m anagem ent techniqu es.1T h e m a jor effect of these develop m en ts fell on the checkable deposit co m p o n en t o f transaction balances, resulting in an accelerated rise in the cu rren cy ratio from 1972 su per-N O W accounts (NOW' a ccou nts w ith no m in im u m m aturity and no ceilin g on yield s) w ere perm itted. throu gh the rest o f the decade. This deregu lation blu rred the sharp distinction b etw een transaction and savings accounts that had existed fo r nearly 50 years. Th e Banking A ct o f 1933 had p roh ib ited the paym en t o f interest on d em a n d deposits, m aking the checkable c o m p o n ent o f M l a relatively unattractive sav ings vehicle, esp ecia lly in tim es o f rising interest rates. Som e changes to this situation took place in the 1970s, but d id not have a m a jor effect on the unique transaction characteristics o f M l. Then, in the 1980s, checkable dep osits that y ie ld e d explicit interest and had m an y o f the characteristics o f In 1978 and 1979, sm all-d enom in ation tim e d e posits o f varying m aturities, w ith interest rates tied to Treasu ry certificates o f com parab le maturities, w e re au th orized. In 1980, w ith the passage o f the D ep ository Institutions D eregu lation and M o n e tary C ontrol Act, a six-year phase-out o f interest rate ceilin gs on tim e dep osits w as established11; m oreover, n a tio n w id e N O W a ccou nts w e re autho rized at the en d o f 1980. In 1982, n ew types o f time deposits that paid market interest rates w ere in trod u ced and the Garn-St. Germ ain Act was passed w h ich a u th o rized m o n ey market deposit accounts. By the en d o f 1983, alm ost all interest savings deposits w ere in trodu ced. T h e yie ld s on these n ew checkable deposits rates on tim e dep osits w ere deregu lated and adju sted very sluggishly to changes in market 5From 1972 to 1980, the demand deposit share of liquid assets fell at about a 6 percent annual rate. The decline in house holds’ holdings of transaction balances as a proportion of liquid assets was relatively minor. The rate of decline of neither household nor business holdings of transaction balances seems closely tied to interest rate fluctuations in the 1970s (Kopcke, 1987). 6See Gilbert (1986). SEPTEMBER/OCTOBER 1988 52 interest rates.7C onsequently, as market interest rates fell sharply in the 1980s, the spread b etw een the rates offered on checkable deposits and m ar ket interest rates on o th er short-term liqu id assets clo sed rapidly. Th e public resp o n d ed bv h old in g m ore checkable deposits." T h e dem and for cu r rency, how ever, w as m u ch less affected by these developm en ts, causing the cu rren cy ratio to flat ten out from 1980 to 1984, then decrease sharply in 1985. In addition to its dom in ant effect on the M I m ultiplier, the k-ratio also exerts a strong in flu ence on the m ovem ents o f the o th er m o n eta iy aggregates. A com parison o f charts 2 and 3, h o w ever, show s that the M2 and M3 m u ltipliers w ere rising w h e n the k-ratio rose then flattened out in recent years w h e n the k-ratio fell sharply. Clearly, for the M2 and M3 m ultipliers, the in flu ence o f o th er factors d o m in a ted the effect o f the k-ratio. The t2-Ratio A rise in the t2-ratio reflects the p u b lic’s desire to h old m o re savings-tvpe deposits (M2 — M l) relative to checkable deposits. Since the t2-ratio enters d irectly into the n um erator o f the M2 and M3 m ultipliers, a rise in this ratio increases these m u ltipliers.” Chart 4 show s the dom in an t in flu en ce o f the t2-ratio on the M2 and M3 m u ltip li ers. A lthou gh the rising k-ratio exerted a negative in flu ence on these m u ltipliers fo r m ost o f the p e rio d sh ow n in the chart, its in flu ence w as offset by the m o vem en t o f the t2-ratio. (A p p en d ix II qu an tifies the in flu en ce o f each o f these ratios on the M2 and M3 m ultipliers.) T h e greater d isparity b e tw een the m ean grow th rate o f these m u ltipliers and that o f the base (than that b etw een M l and the base) d u rin g m ost o f the 1960s and 1970s was the result o f the 4 p ercen t annual rate o f grow th o f the t2-ratio. T h e 1985-87 period stands out in contrast to previous periods. A lthou gh the t2-ratio declined , as sh ow n in chart 4, the M2 and M3 m u ltipliers did not d eclin e as m u ch as on e w o u ld have ex pected, given the d eclin e in the t2-ratio alone. In 7See Wenninger (1986) and Roth (1987). 8A Federal Reserve survey of changes in the use of cash and transaction accounts from 1984 to 1986 found that individuals consolidated their accounts, increased their use of checking accounts as a family savings vehicle and diminished their use as a media for transactions. The study also found that average cash balances increased with the decline in interest rates, while portfolio considerations became more important and transaction motives less important in how people managed cash and transaction accounts between 1984 and 1986 (Avery et. a l„ 1987). FEDERAL RESERVE BANK OF ST. LOUIS this period, how ever, the falling k-ratio, as show n in chart 3, partly offset the t2-ratio's negative effect on these m ultipliers. As chart 5 shows, m ovem en ts in the t2-ralio have been dom in ated by relative m ovem en ts o f savings (SVC) and small tim e dep osits (S I D). Dur ing the 1970s, the sharply rising proportion o f small tim e deposits relative to checkable deposits ISTD/D) p rovid ed the m a jor im petus fo r the rise in the t2-ratio. T h e stron g n egative in flu ence o f the savings co m p o n en t in the late 1970s and early 1980s w as further offset by a sharp rise in o th er liquid savings instrum ents such as MMDAs, M M M Fs and overnight RPs relative to checkable deposits (OL/D). W h en the t2-ratio d ec lin ed in late 1985 through mid-1987, it w as p red o m in an tly b e cause o f a sharp fall in the ratio o f sm all tim e d e posits to checkable deposits. The t3-Ratio In recent years (1983-87), the spread b etw een the grow th rates o f M3 and M 2 has been m u ch n arrow er than it w as in the 1970s and early 1980s. This change can be explain ed b y the beh avior o f the t3-ratio. This ratio, w h ic h captu res the public's d esired h oldin gs o f assets in clu d ed so lely in M3 co m p a red w ith checkable dep osits d eterm in es the spread b etw een the M3 and M2 m u ltipliers. Chart 6 show s that, as this ratio rose sh arply from the early 1970s to the early 1980s, the spread b etw een the M3 and M2 m u ltipliers rose steadily. A fter 1982, h ow ever, as the t3-ratio fell, the spread b e tw een the M3 and M2 m u ltipliers stabilized. M ovem en ts o f large tim e dep osits have d o m i nated m ovem en ts o f the t3-ratio. T h e o th er c o m p on en ts o f (M3 — M2) con stitu ted n o m o re than 20 p ercen t o f the total until 1977. A lth ou gh these o th er m an aged liabilities (term RPs and E u rod ol lars and in stitu tion-only M M M Fs) rose rapid ly enou gh to account fo r 36 p ercen t o f the total by 1987, fluctuations in large tim e deposits con tin u ed to be the dom in an t cause o f t3-ratio fluctuations. The sharp break in this ra tio ’s long-run pattern that o ccu rred in late 1984 and con tin u ed o ver the 9To the extent that (M2 - M1) contains reservable liabilities, an increase in time and savings deposits absorbs reserves and reduces the multipliers. In previous formulations of the multi plier, a t-ratio appears in the denominator of all the multipliers (see Burger, 1971). In the multipliers presented in this paper, this effect is not separated out in the denomination of the multipliers, but its effect is reflected in movements in the r-ratio. This influence varies between the period before 1980 and after 1980, because of the definition of adjusted reserves that ap pears in the r-ratio. The exact nature of this influence is shown in Gilbert (1987). 53 Chart 4 M2 and M3 Multipliers and t2-Ratio 1960 1965 1970 1975 1980 1985 1990 next n ine quarters reflected a slo w in g o f the grow th o f large tim e deposits relative to the grow th o f checkable deposits. A lthou gh the affected the various m o n e ta iy aggregates in d is parate w ays. T h e fram ew ork p resen ted in this article is o n e w a y to isolate the shifts that in grow th o f o th er m an aged liabilities slo w ed in 1985, it resu m ed its previous pace in 1986 and 1987. flu en ced the m on etary aggregates and illustrate th eir effects on the grow th rates o f the aggregates. SUMMARY Lookin g at past relationships, on e m ight be tem p ted to con jectu re that, in the 1980s, the m o n eta iy aggregates b ecam e totally d iscon n ected from Federal Reserve actions as su m m arized in the m o n eta iy base. Bv presen tin g a fram ew ork that can be u sed to explain the m ovem en ts o f the ag REFERENCES Avery, Robert B., George E. Elliehausen, Arthur B. Kennickell, and Paul A. Spindt. "Changes in the Use of Transaction Accounts and Cash from 1984 to 1986,” Federal Reserve Bulletin (March 1987), pp. 179-96. gregates both relative to each o th er and relative to the grow th o f the m on etary base, this article has Brunner, Karl and Meltzer, Allan H. “ Liquidity Traps for Money, Bank Credit and Interest Rates,” Journal of Political Economy (January/February 1968), pp. 1-37. sh ow n this not to be the case. D uring the 1980s, n ew financial assets w e re in trod u ced and m ajor Burger, Albert E. The Money Supply Process, (Wadsworth Publishing Co., 1971). changes o ccu rred in inflation, interest rates and tilt; basic characteristics o f m ost o f the traditional m o n eta iy assets. In response to these events, the public m ade sizable shifts in its portfolio, w h ich _________“ Alternative Measures of the Monetary Base,” this Review (June 1979), pp. 3-8. Burger, Albert E., and Anatol B. Balbach. “ Derivation of the Monetary Base,” this Review (November 1976), pp. 2-8. SEPTEMBER/OCTOBER 1988 54 Chart 5 Components of the t2-Ratio Burger, Albert E. and Rasche, Robert H. “ Revision of the Monetary Base,” this Review (July 1977), pp. 13-23. Gutmann, Peter. “The Subterranean Economy," Financial Analysts Journal (November/December 1977), pp. 26-28. Cagan, Phillip. “The Demand for Money Relative to the Total Money Supply,” Journal of Political Economy (August 1958), pp. 303-28. Hess, Alan C. “ An Explanation of Short-Run Fluctuations in the Ratio of Currency to Demand Deposits,” Journal of Money, Credit and Banking (August 1971), pp. 666-79. Dotsey, Michael. “The Demand for Currency in the United States," Journal of Money, Credit and Banking (February 1988), pp. 22-40. Judd, John P., and Bharat Trehan. “ Portfolio Substitution and the Reliability of M1, M2 and M3 as Monetary Policy Indica tors,” Federal Reserve Bank of San Francisco Economic Review (Summer 1987), pp. 5-29. Gillian, Garcia. “The Currency Ratio and the Subterranean Economy," Financial Analysts Journal (November/December 1978), pp. 64-69. Gavin, William, and Pakko, Michael. “ M1-M1A?" Federal Reserve Bank of Cleveland Economic Review (July 1,1987). Gilbert, R. Alton. “ Requiem for Regulation Q: What It Did and Why It Passed Away," this Review (February 1986), pp. 2 2 37. _________“ A Revision in the Monetary Base,” this Review (August/September 1987), pp. 24-29. FEDERAL RESERVE BANK OF ST. LOUIS Kopcke, Richard W. “ Financial Assets, Interest Rates and Money Growth,” New England Economic Review Federal Resen/e Bank of Boston (March/April 1987), pp. 17-30. Motley, Brian. “ Should M2 be Redefined?” Federal Reserve Bank of San Francisco Economic Review (Winter 1988), pp. 33-51. Rasche, Robert H., and James M. Johannes. Controlling the Growth of the Monetary Aggregates (Kluwer, 1987). 55 Chart 6 Spread Between M3 and M2 Multipliers and t3-Ratio t3-ratio A nnual D ata Multipliers S** -------- A M3 Multiplier less *t * M2 Multiplier * scaie^ t * ** *■+ • ^lrl _.■------- " — / * * — /r y /✓ / / v / v“ v • • t3-ratio < scale —''' 1960 1965 1970 1975 1980 1985 1990 Roth, Howard. “ Has Deregulation Ruined M1 as a Policy Guide?” Federal Reserve Bank of Kansas City Economic Review (June 1987), pp. 24-37. Tatom, John A. “ Recent Financial Innovations: Have They Distorted the Meaning of M1 ?” this Review (April 1982), pp. 23-35. Simpson, Tom. “ Changes in the Financial System: Implications for Monetary Policy,” Brookings Economic Papers (Volume 1, 1984), pp. 249-65. Wenninger, John. “ Responsiveness of Interest Rate Spreads and Deposit Flows to Changes in Market Interest Rates,” Federal Reserve Bank of New York Quarterly Review (Au tumn 1986), pp. 1-10. SEPTEMBER/OCTOBER 1988 56 Appendix I Derivation of Multipliers M l m ultiplier (m l) M3 m ultiplier (m3) AM B = R + HAM + C m3 = C + D + M 2 - M l + M3 - M2 ^ + C ml = Ml C + D AM B R + RAM + C K) ( ™ > ( e) ml = 1 + k r + k M2 m ultiplier (m2) m2 = C + D + M2 - M l R + RAM + C ( 1 + cK m ^ ) /R + R A M \ V C \ {— ET-A d) m2 = y R + RAM + C Ml = C + D 1 + k + t2 r + k FEDERAL RESERVE BANK OF ST. LOUIS ^M2 - M1^ D( ™ m3 = )+ (C 1 + k + t2 + t3 r + k M3 - M2^ ) D 57 Appendix II Magnitude of the Influence of the Component Ratios on the Multipliers T h e size o f the effect that each o f the ratios ex e(m2,t2l = 12/I1+k + t2) > 0 elm3,t2l = t2/( 1 + k + t2 + t3l > 0 e(m3,t3) = t3/H + k + t2 + t3l > 0 e(m l,r), e(m2,r), e(m3,r) = - r / (r + k ) < 0 erts on the grow th o f the m o n ey m u ltipliers d e pends hoth on the grow th rate o f each ratio and the responsiveness o f the m u ltip lier to a ch ange in the ratio. T h is responsiveness can be qu an tified by calcu lating the partial elasticities o f each o f the m u ltipliers w ith respect to its co m p o n en t ratios, as show n below . T h ese results sh ow that, in this form ulation, although the respon se o f all the m u l tipliers to a ch ange in the r-ratio are the same, there are differen ces in the response o f the m u lti pliers to the o th er ratios. Table A1 presents the co m p u ted annual averages o f these elasticities. T h e values o f these elasticities ch ange over tim e as the ratios change. For exam ple, the rise in t2-ratio has affected the relationship betw een the response o f m2 and m3 to a change in the t2-ratio. In the early 1960s, e(m2,t2) and e(m3,t2) w ere both about the same. By the early 1980s, the e(m2,t2) h ad risen to about .76 w h ile e(m3,t2) w as still about .62. In 1985-87, these elasticities fell as the t2- and t3-ratios declined . ELASTICITIES OF THE MULTIPLIERS WITH RESPECT TO THEIR COMPONENT RATIOS T h e m agnitu de o f the in flu ence o f the p o rtfolio shifts e m b ed d e d in the k-, t2-, and t3-ratios on the grow th o f the m u ltipliers can be isolated using the fo llo w in g form ula: e(m l,k ) = k(r — 1)/(r + k)(1 + k) < 0 e(m2,k) = k ( r - 1 - t2)/(r + k )(l + k +12) < 0 e(m3,k) = k ( r - 1 - 12- t3)/(r + k ) ( l + k +12 +13) < 0 Table A1 Elasticities of the Multipliers with Respect to Their Component Ratios Year e(m1,k) e(m2,k) e(m3,k) e(m2,t2) e(m3,t2) e(m3,t3) e(m,r) Year 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 5 -0 .4 4 -0 .4 4 -0 .4 4 - 0 .4 3 -0 .4 3 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 4 -0 .4 5 -0 .4 6 -0 .4 6 -0 .4 6 -0 .4 6 -0 .4 7 -0 .4 7 -0 .5 7 -0 .5 8 -0 .5 8 -0 .5 8 -0 .5 9 -0 .5 9 -0 .5 9 -0 .5 9 -0 .6 0 -0 .6 0 -0 .6 1 -0 .6 3 - 0 .6 4 -0 .6 4 - 0 .6 5 -0 .6 5 -0 .6 6 -0 .6 7 -0 .6 8 -0 .6 8 -0 .6 8 -0 .6 7 -0 .6 6 -0 .5 8 -0 .5 8 -0 .5 8 -0 .5 8 -0 .5 9 -0 .6 0 -0 .6 0 - 0 .6 0 -0 .6 0 -0 .6 1 -0 .6 2 -0 .6 3 -0 .6 4 -0 .6 5 -0 .6 6 -0 .6 6 -0 .6 8 -0 .6 9 -0 .6 9 -0 .6 9 - 0 .6 9 -0 .6 8 - 0 .6 7 0.63 0.64 0.65 0.65 0.65 0.65 0.67 0.69 0.69 0.70 0.71 0.73 0.74 0.74 0.74 0.75 0.75 0.76 0.76 0.76 0.76 0.75 0.74 0.60 0.61 0.61 0.61 0.62 0.62 0.62 0.63 0.62 0.60 0.61 0.64 0.66 0.64 0.62 0.62 0.61 0.61 0.62 0.61 0.61 0.60 0.59 0.04 0.05 0.06 0.06 0.05 0.05 0.08 0.09 0.11 0.14 0.14 0.12 0.12 0.14 0.16 0.17 0.19 0.20 0.18 0.20 0.20 0.20 0.20 -0 .3 5 -0 .3 4 -0 .3 4 -0 .3 4 -0 .3 4 -0 .3 3 -0 .3 3 -0 .3 3 -0 .3 3 -0 .3 3 -0 .3 2 -0 .3 0 -0 .3 0 -0 .2 9 -0 .2 8 -0 .2 8 -0 .2 7 - 0 .2 6 -0 .2 6 -0 .2 6 - 0 .2 6 -0 .2 7 - 0 .2 8 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 SEPTEMBER/OCTOBER 1988 58 m = e(m,k)(k) + e(m,t2)(t2) + e(m,t3)(t3) + e( m,rllr). In the above form ula, e lm ___I represents the partial elasticity o f the respective m u ltip lier w ith respect to the specified ratio. For exam ple, elm l.ki w o u ld represent the partial elasticity o f the M l m u ltiplier (m l) w ith respect to the k-ratio. Th e dots above the ratios d en ote g ro w th rates. Th e results o f this d ec om p os itio n o f the grow th rates ot the respective m u ltipliers are sh ow n in tables A2, A3 and A4. Th e results through 1984 w ere co m p u ted using annual g ro w th rates o f the co m p o n en t ratios that a p p ea r in the m ultipliers, and the elasticities are the ones rep orted in table A l. Q uarterly data fo r I/1985-I/1988 w ere co m p u ted using quarterly grow th rates and quarterly elasticity measures. O ver the three years en d in g in 1984, the k-ratio, on average, sh ow ed essentially no grow th . Then, from fourth quarter 1984 to first qu arter o f 1986, it fell at an annual rate o f about 5 percent; over the next fou r quarters, it fell 10 percent. This effect is sh ow n in tables A2, A3 and A4, as the negative contribu tions o f the k-ratio to the g ro w th rates o f the m u ltipliers becam e sm aller in the early 1980s and then tu rned into large positive effects beginn in g in 1985. This effect d om in a ted the grow th o f m l, lea d in g to a p ro n o u n ced change in the relationship b etw een the g ro w th o f M l and the m on etary base. From fourth qu arter 1985 to first qu arter 1987, the grow th o f M l ex c e e d e d the grow th o f the m o n e ta iy base b y about 7 percen tage points. In the 1985-87 period, the effect o f the d eclin in g k-ratio on the relationships b etw een the grow th o f M2 and M3 and the grow th o f the m on etary base was not nearly as m arked as w as the case w ith M l. Tables A3 and A4 sh o w that the ch anged beh avior o f t2- and t3-ratios acted to offset the ch anged beh avior o f the k-ratio on these m ultipliers. Table A2 Contribution of the Component Ratios to the Growth Rate of ml Year EEMK EER MULX 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 -0 .5 9 -0 .9 4 -0 .6 5 0.27 -0 .4 8 -1 .4 3 -0 .2 1 0.11 - 0 .6 7 -2 .3 1 - 2 .6 8 -2 .1 0 -0 .8 2 -0 .9 3 -1 .2 2 -1 .8 3 -0 .2 5 -0 .4 8 0.86 -0 .9 1 -1 .3 0 3.59 1.93 -0 .4 6 0.07 -0 .4 1 -0 .0 4 0.75 0.37 -0 .9 1 - 0 .3 4 -0 .3 7 -0 .6 3 -0 .4 1 0.18 0.43 -0 .1 1 0.51 -0 .1 6 -1 .3 6 0.83 0.37 -0 .1 8 0.43 0.51 0.29 -1 .0 2 -0 .8 5 -1 .0 5 -0 .2 6 0.30 -1 .0 5 - 1 .0 6 -0 .2 0 -1 .0 0 - 2 .9 0 -2 .9 9 -1 .7 9 -0 .3 1 -1 .0 2 -0 .6 7 -1 .9 9 1.16 0.33 1.17 -1 .0 8 1.74 4.04 2.19 Quarter EEMK EER MULX 1985.1 1985.2 1985.3 1985.4 1986.1 1986.2 1986.3 1986.4 1987.1 1987.2 1987.3 1987.4 1988.1 2.60 1.97 3.50 2.55 1.50 5.27 6.15 5.93 2.47 - 0 .4 6 -4 .1 2 -3 .6 6 -3 .5 9 0.53 0.69 0.67 0.03 0.29 0.80 0.75 0.72 -0 .4 3 0.22 0.53 0.32 -0 .7 7 3.20 2.70 4.28 2.48 1.77 6.01 6.87 6.49 2.06 - 0 .2 8 -3 .4 8 -3 .3 4 -4 .2 8 EEMK = contribution of k-ratio to growth of m l Since early to mid-1987, the k-, t2- and t3-ratios all have risen, resu m in g patterns that are m ore in EER = contribution of r-ratio to growth of m l MULX = actual growth rate of m l line w ith th eir historical behavior. Since the relative grow th rates o f the aggregates d ep en d on the in flu en ce o f each o f these ratios on the respective m ultipliers, the rise in the k-ratio, w h ic h has been esp ecially strong relative to its historical pattern (from 11/1987 to 1/1988, the k-ratio rose at an 8 percen t rate), has d om in a ted the grow th o f all three m ultipliers, as sh ow n in tables A2, A3 and A4. C onsequently, the M l m u ltip lier has fallen and the grow th o f the FEDERAL RESERVE BANK OF ST. LOUIS m o n eta iy base has e x ceed ed the grow th o f M l, as w as gen erally the case before 1985. Th e m u ltipliers associated w ith M2 and M3, how ever, have fallen since early 1987; as a result, the grow th rate o f the m on etary base also has ex ce e d e d the grow th o f these aggregates. This pattern is qu ite different from that exp erien ced before 1985. 59 Table A3 Contribution of the Component Ratios to the Growth Rate of m2 Year EEM2K EEM2T2 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 -0 .7 7 -1 .2 4 -0 .8 6 0.36 -0 .6 4 -1 .9 1 -0 .2 8 0.15 -0 .9 2 -3 .2 0 -3 .7 9 -3 .0 1 -1 .1 8 -1 .3 6 - 1 .7 7 -2 .7 0 - 0 .3 7 -0 .7 1 1.26 -1 .3 4 1.89 5.12 2.71 3.86 2.15 3.00 1.13 0.41 0.78 5.13 4.95 2.75 1.94 5.71 7.80 5.00 0.72 1.12 2.59 2.06 2.76 0.86 1.22 -0 .6 1 -6 .2 2 -5 .4 5 Quarter EEM2K EEM2T2 1985.1 1985.2 1985.3 1985.4 1986.1 1986.2 1986.3 1986.4 1987.1 1987.2 1987.3 1987.4 1988.1 3.83 2.89 5.11 3.71 2.17 7.56 8.74 8.34 3.46 - 0 .6 4 - 5 .8 0 -5 .1 6 -5 .1 0 0.49 -4 .5 8 -6 .3 2 -5 .7 8 -3 .4 4 -7 .4 8 -8 .9 1 -10.51 -7 .5 7 - 3 .7 9 3.75 1.54 -4 .3 9 EER -0 .4 6 0.07 -0 .4 1 -0 .0 4 0.75 0.37 -0 .9 1 - 0 .3 4 -0 .3 7 - 0 .6 3 -0 .4 1 0.18 0.43 -0 .1 1 0.51 -0 .1 6 1.36 0.83 0.37 -0 .1 8 0.43 0.51 0.29 EER 0.53 0.69 0.67 0.03 0.29 0.80 0.75 0.72 -0 .4 3 0.22 0.53 0.32 -0 .7 7 MUL2X 2.59 0.98 1.71 1.45 0.53 -0 .7 4 3.90 4.71 1.45 -1 .8 8 1.51 4.92 4.22 -0 .7 5 -0 .1 2 -0 .2 6 3.07 2.87 2.47 -0 .3 1 1.72 -0 .6 0 -2 .4 7 MUL2X 4.87 -0 .9 9 -0 .5 2 -2 .0 5 -0 .9 8 0.88 0.59 -1 .4 8 -4 .5 3 -4 .2 2 -1 .4 8 -3 .2 9 -1 .4 6 EEM2K = contribution of k-ratio to growth of m2 EEM2T2 = contribution of t2-ratio to growth of m2 EER = contribution of r-ratio to growth of m l MUL2X = actual growth rate of m2 SEPTEMBER/OCTOBER 1988 60 Table A4 Contribution of the Component Ratios to the Growth Rate of m3 Year EEM3K EEM3T2 EEM3T3 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 -0 .7 7 -1 .2 5 - 0 .8 7 0.36 - 0 .6 5 -1 .9 2 -0 .2 8 0.15 -0 .9 3 -3 .2 6 - 3 .8 5 -3 .0 5 - 1 .2 0 -1 .3 8 -1 .8 1 - 2 .7 5 -0 .3 7 -0 .7 2 1.28 -1 .3 7 1.93 5.22 2.77 3.69 2.04 2.83 1.06 0.39 0.74 4.74 4.52 2.44 1.66 4.93 6.87 4.42 0.61 0.94 2.14 1.66 2.20 0.71 0.97 -0 .4 9 -4 .9 7 -4 .3 5 1.10 0.87 0.99 0.41 -0 .9 0 0.43 3.17 1.47 3.82 4.20 -0 .0 7 -0 .8 4 0.21 3.65 2.86 1.66 2.74 1.98 -2 .0 8 2.40 -0 .2 0 -1 .3 3 -0 .9 1 Quarter EEM3K EEM3T2 EEM3T3 1985.1 1985.2 1985.3 1985.4 1986.1 1986.2 1986.3 1986.4 1987.1 1987.2 1987.3 1987.4 1988.1 3.91 2.95 5.20 3.78 2.22 7.71 8.91 8.50 3.52 -0 .6 6 -5 .9 2 - 5 .2 7 -5 .2 1 0.39 -3 .6 6 -5 .0 7 -4 .6 3 -2 .7 4 -5 .9 7 -7 .1 3 -8 .4 4 -6 .0 8 -3 .0 3 2.98 1.22 3.48 -1 .8 9 -1 .7 0 -3 .3 6 -0 .7 3 1.75 -2 .5 6 -2 .9 1 -3 .7 2 -1 .5 2 1.33 2.58 1.95 0.85 EEM3K = contribution of k-ratio to growth of m3 EEM3T2 = contribution of t2-ratio to growth of m3 EEM3T3 = contribution of t3-ratio to growth of m3 EER = contribution of r-ratio to growth of m3 MUL3X = actual growth rate of m3 FEDERAL RESERVE BANK OF ST. LOUIS EER - 0 .4 6 0.07 -0 .4 1 -0 .0 4 0.75 0.37 -0 .9 1 -0 .3 4 - 0 .3 7 -0 .6 3 -0 .4 1 0.18 0.43 -0 .1 1 0.51 -0 .1 6 1.36 -0 .8 3 0.37 -0 .1 8 0.43 0.51 0.29 EER 0.53 0.69 0.67 0.03 0.29 0.80 0.75 0.72 -0 .4 3 0.22 0.53 0.32 - 0 .7 7 MUL3X 3.43 1.68 2.48 1.79 -0 .4 7 -0 .3 7 6.26 5.70 4.52 1.57 0.58 3.00 3.84 2.42 2.34 0.87 5.29 4.24 0.15 1.73 1.68 - 0 .5 8 -2 .2 2 MUL3X 2.94 -1 .7 1 -2 .5 5 -1 .5 6 1.48 -0 .0 2 -0 .3 7 -2 .9 6 - 4 .5 0 -2 .1 6 0.20 -1 .7 8 - 1 .6 4 Federal Reserve Bank o f St. Louis Post O ffice Box 442 St. Louis, M issou ri 63166 The Review is published six 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 free 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. The views expressed are those o f the individual authors and do not necessarily reflect official positions o f the Federal Reserve Bank o f St. Louis or the Federal Reserve System. 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.