The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Review is pu blished 10 tim es per year by the R esearch D epartm ent o f the F ed eral Reserve Bank o f St. Louis. Single-copy subscriptions are a v a ilab le to the public fr e e o f charge. Mail requ ests fo r subscriptions, b a c k issues, or address changes to: R esearch D epartm ent, F ed eral R eserve Bank o f St. Louis, P.O. Box 442, St. Louis, M issouri 63166. A rticles herein may be reprinted p rov id ed the source is credited. Please provide th e B an k’s R esearch D epartm ent w ith a copy o f reprinted m aterial. Will the Removal of Regulation Q Raise Mortgage Interest Rates? R. ALTON GILBERT EGISLATION passed in March 1980 calls for the gradual phase-out of interest rate ceilings on deposits by 1986. Some critics of this change have claimed that banks and thrift institutions will charge their borrowers higher interest rates once these deposit interest rate ceilings are removed. Accord ing to these critics, lenders will raise their lending rates to cover their increased deposit costs.1 Regulation Q of the Federal Reserve and, therefore, are commonly referred to as Regulation Q. One of the primary reasons for imposing ceilings on deposit interest rates was to reduce the number of failing banks by reducing their interest cost. Another ob jective was to reduce the incentives for rural banks to hold large interest-earning balances with their cor respondents in the financial centers.2 This article presents a brief history of deposit in terest rate ceilings in the United States and their effects. It then describes the process established by recent legislation for eliminating ceilings, and its likely impact on the interest rates that borrowers will pay. Finally, the analysis is extended to cover the effects of the All Savers Certificate program on in terest rates that depository institutions will charge on loans. Much of the concern in the early 1930s centered on interest payments on demand deposits. Interest payments on demand deposits were prohibited under the Banking Acts of 1933 and 1935. The max imum interest rate on all time and savings deposits was initially set at 3 percent, slightly below the average interest rate that commercial banks and thrift institutions had been paying on time and sav ings deposits, but abov e then-existing market yields on high-grade short-term securities.3 The choice of the initial ceiling rate on time and savings deposits indicates that the purpose of these ceiling rates on time and savings deposits was not to keep them below yields on alternative investments, but to re duce deposit rates slightly and thus lower the interest costs of depository institutions. W H Y HAS THE F E D E R A L G O V E R N M E N T R E G U L A T E D D E P O S IT IN T E R E ST RATES? Federal bank regulators received the legal author ity to regulate interest rates that commercial banks may pay depositors in the Banking Acts of 1933 and 1935. The interest ceilings have been set under ‘The view that the elim ination of ceilin g interest rates on deposits would cause interest rates paid by borrowers to rise appears in D ep o sit o r y Institu tions D ere g u latio n Act o f 1979, Hearing on S. 1.347 before the Subcom m ittee on Financial Institutions, Senate Committee on Banking, Housing, and Urban Affairs, Part II and Part I II , 96 Cong. 1 Sess. (Governm ent Printing Office, 1979). See comments by Ralph W. Pritchard, first vice president, Na tional Association o f Realtors (June 27, 1979); Thomas F. Bolger, first vice president, Independent Bankers Association (July 18, 1979); and Henry B. Schechter, director, D epartm ent of Urban Affairs, A FL -C IO (July 18, 1979). During the 20 years from the mid-1930s to the mid-1950s, the ceiling rates on time and savings deposits were above market interest rates. In 1957 and 1962, when market interest rates rose near or above the ceiling rates on savings deposits, these ceilings were raised (chart 1). 2Albert H. Cox, Jr., R egu lation o f I n te res t R a tes on B an k D e p o s its, M ichigan Business Studies, vol. X V II, no. 4 (Bujreau of Business Research, University of M ichigan, 1966). 3Charlotte E. Ruebling, “The Administration of Regulation Q ,” this R ev iew (February 1970), pp. 30-31. 3 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 C h art I 3 - M o n t h T re a su ry B ill R a te a n d C e ilin g R a te o n S a v i n g s D e p o s it s a t C o m m e rc ia l B a n k s [^ Q u a r t e r ly a v e r a g e s L a te st d a t a p lo t t e d :3 r d q u a r t e r 1981 In 1966, interest rate ceilings were imposed on deposits of thrift institutions. Sponsors of the en acting legislation asserted that interest rates were being driven up by competition for deposits among banks and thrifts, and that ceiling interest rates on deposits at thrift institutions would stop this escala tion. They assumed that by permitting slightly higher ceiling rates at thrift institutions specializing in resi dential mortgage lending, there would be an ade quate supply of credit for residential mortgages at reasonable mortgage interest rates.4 These controls on interest rates paid by thrift institutions were viewed initially as temporary mea sures to deal with “unusual circumstances.” Over time, however, thrift institutions have come to view the differentials between the ceiling interest rates on their deposits and those imposed on commercial banks as essential in attracting deposits to be used ‘T e m p o r a r y In te res t R a te Controls, Report No. 1777, House Com m ittee on Banking and Currency, 89 Cong. 2 Sess. (GPO, 1966); and In terest R ates a n d M ortgage C r e d i t , Hearing on S. 3687, S. 3627 and S. 3529 before the Senate Com mittee on Banking and Currency, 89 Cong. 2 Sess. (GPO, 1966). Digitized for 4 FRASER for residential mortgage lending. These differentials have been considered important elements ol a pub lic policy designed to expand the supply of mortgage credit and increase residential construction.5 If the differentials in ceiling rates between thrifts and commercial banks are to stimulate the flow of deposits to thrift institutions, ceiling interest rates on some categories of deposits at commercial banks must be below market interest rates. If all deposit interest rate ceilings were abov e market interest rates, the higher ceiling rates at thrift institutions would not induce individuals to hold their deposits there rather than at commercial banks. This would occur because both commercial banks and thrifts would be paying the lower market interest rate to depositors instead of the higher ceiling rates. Since 1966, the ceiling rate on savings deposits at com mercial banks has been below the three-month Treasury bill rate (a measure of market rates) except for only a few months in 1967, 1971, 1972 and 197677 (chart 1). 5Preston Martin, “A Case for Regulation Q ,” J o u r n a l o f the F e d e r a l H o m e L o a n B a n k B o a r d (O ctober 1970), pp. 1-6. FEDERAL RESERVE BANK OF ST. LOUIS T H E E F FE C T S O F D E P O S IT IN T E R E ST RATE C E IL IN G S If maintaining deposit interest rate ceilings below market interest rates, with slightly higher rates al lowed for thrift institutions, was intended to produce a stable supply of mortgage credit available to homebuyers at moderate interest rates, it has failed to do so. The growth of deposits at thrift institutions has slowed whenever market interest rates have risen above the deposit ceiling rates.6 These fluctu ations in the growth of deposits at thrift institutions may have contributed to the abrupt changes in the pace of residential construction activity in recent decades.7 Deposit interest rate ceilings have discriminated against the relatively less wealthy savers.8 There are no ceiling rates on deposits in denominations of $100,000 or more. The ceiling rate on money market certificates (time deposits with maturities of six months) fluctuates with market interest rates, but those require a minimum deposit of $10,000. Debt obligations of the U.S. Treasury, investments with risk characteristics most similar to deposits of fed erally insured institutions, are sold in minimum denominations that are substantially larger than the average time or savings deposits of individuals. 6Edward F. M cKelvey, I n te res t R a te C eilin g s a n d D is in ter m ed iatio n , Staff Econom ic Studies 99 (Board of Governors of the Fed eral Reserve System, 1978). 7Dwight M. Jaffee and Kenneth T. Rosen, “ Mortgage Credit Availability and Residential Construction,” B ro o k in g s P ap e rs on E c o n o m i c A ctivity (2: 1979), pp. 333-76; and Neil G. Berkman, “ Mortgage Finan ce and the Housing C y cle,” N ew E n g l a n d E c o n o m i c R ev iew (September/October 1979), pp. 54-76. Results of some studies, however, do not support the view that changes in the availability of mortgage credit through thrift institutions influence residential construction. See Francisco Arcelus and Allan H. M eltzer, “The Markets for Housing and Housing Ser vices, " J o u r n a l o f Money, C r e d it a n d B an kin g (February 1973), pp. 78-99; Allan H. M eltzer, “Credit Availability and Econom ic D ecisions: Some E vid ence from the Mortgage and Housing Markets, ’’J o u r n a l o f F in a n ce (June 1974), pp. 763-78; and Paul D e Rosa, “ Mortgage Rationing and Residential Investm ent: Some Results from a Brainard-Tobin Model " J o u r n a l o f Money, C r e d it a n d B a n k in g (February 1978), pp. 75-87. 8Edward J. Kane, “ Short-Changing the Small Saver: Federal Governm ent Discrim ination against Small Savers during the Vietnam W ar,” J o u r n a l o f Monet/, C r e d it a n d B an kin g (No vem ber 1970), pp. 513-22; Edward J. Kane, “Consequences of Contemporary C eilings on Mortgage and Deposit Interest Rates for H ouseholds in D ifferen t E cono m ic C ircu m stan ces,” in George M. von Furstenberg, ed.. T he G o v e r n m e n t a n d C a p ita l F o rm a tio n (Ballinger Publishing Company, 1980), pp. 401-41; Charles C lotfelter and Charles Lieberm an, “On the D istribu tional Impact o f Federal Interest Rate Restrictions,” J o u r n a l o f F in a n ce (March 1978), pp. 199-213; Edward C. Law rence and Gregory E. Elliehausen, “The Impact of Federal Interest Rate Regulations on the Small Saver: Further E vid en ce," J o u r n a l o f F in a n ce (June 1981), pp. 677-84. DECEMBER 1981 Consequently, savers with less than $10,000, who want an investment with risk and liquidity charac teristics similar to Treasury bills, are limited to savings deposits at federally insured institutions. Because of the interest rate ceilings on these de posits, the yield is generally less than that available on Treasury bills. Several studies have estimated that savers have “lost” several billion dollars in earnings as a result of the Regulation Q ceilings.9 E L IM IN A T IN G R E G U L A T IO N Q One of the most significant sections of the D e pository Institutions Deregulation Act of 1980 calls for the elimination of ceilings on deposit interest rates over a six-year period. The statement of find ings and purpose of that section of the act reads as follows: The Congress hereby finds that — (1) lim itations on the interest rates which are payable on deposits and accounts discourage persons from saving money, create inequities for depositors, impede the ability of depository institutions to com pete for funds, and have not achieved their purpose of providing an even flow of funds for home mortgage lending; and (2) all depositors, and particularly those with modest savings, are entitled to receive a market rate o f return on their savings as soon as it is econom ically feasible for depository institutions to pay such rate.10 The act does not specify a timetable for elim inating deposit interest rate ceilings, but delegates those decisions to a newly created committee: the Depository Institutions Deregulation Committee (DIDC). Voting members of the DIDC include: Sec retary of the Treasury; and chairmen of the Federal Reserve Board, Federal Deposit Insurance Corpora tion, Federal Home Loan Bank Board, and National Credit Union Administration. The Comptroller of the Currency is a non-voting member of the DIDC. 9Bruce W. Morgan, “Ceilings on Deposit Interest Rates, the Saving Public and Housing F in an ce,” E q u i t y f o r th e Sm all Saver, Hearings on S.Con.Res. 5 before the Subcom m ittee on Financial Institutions, Senate Com m ittee on Banking, Housing, and Urban Affairs, 96 Cong. 1 Sess. (GPO, 1979), p. 175; David H. Pyle, “T he Losses on Savings Deposits from Interest Rate Reg ulation,” B ell J o u r n a l o f E c o n o m i c a n d M an ag e m e n t S cience (Autumn 1974), pp. 614-22; David H. Pyle, “ In te re st Rate C eilings and Net W orth Losses by Savers,” in K enneth E . Boulding and Thomas Frederick Wilson, eds., R edistribution through the F in a n cia l System (Praeger Publishers, 1978), pp. 87-101; Robert A. Taggart, Jr., “E ffects o f Deposit Rate C eilings: The E vid ence from Massachusetts Savings Banks,” J o u r n a l o f Money, C r e d it a n d B a n k in g (May 1978), pp. 139-57. loD ep o sitor y Institutio ns D ere g u latio n a n d M onetary C ontrol A cto o f 1980, S. Rept. No. 96-640, 96 Cong. 2 Sess. (GPO, 1980), title 11, sec. 202(a). 5 FEDERAL RESERVE BANK OF ST. LOUIS The act directs the D ID C to provide for the orderly phase-out and ultimate elimination of maxi mum interest rates that may be paid on time and savings deposits as rapidly as economic conditions warrant. A primary consideration in determining when conditions warrant raising or eliminating these ceilings is the effect of such changes on the safety and soundness of depository institutions. The act lists the following methods the DIDC may use in phasing out ceiling interest rates on deposits: T he phase-out of' such limitations may be achieved by the D eregulation Com m ittee by the gradual increase in such lim itations applicable to all existing categories of accounts, the com plete elim ination of the limitations applicable to par ticular categories of accounts, the creation of new categories of accounts not subject to limitations or with lim itations set at current market rates, any com bination of the above methods, or any other method.11 One limitation imposed on the DIDC is that it may not raise interest rate ceilings on all deposit cate gories above market interest rates before March 1986. The DIDC has taken limited actions to raise or eliminate ceilings on deposit interest rates (see table 1). The first significant action was to lift caps on ceiling rates for time deposits with maturities of 2 ‘/2 years, which was effective August 1, 1981. The DID C has also created a new category of IRA/Keogh account (with minimum maturity of IV2 years) that will have no regulated interest rate ceiling as of January 1, 1982. THE E F FE C T S O F E L IM IN A T IN G R E G U L A T IO N Q O N IN T E R E ST RATES P A ID RY R O R R O W E R S The effects of eliminating ceiling rates on deposits cannot be determined by examining the effects of actions already taken by the DIDC, since few actions to eliminate the ceiling rates have been taken so far. Effects of eliminating deposit ceiling rates on the interest rates paid by borrowers must, therefore, be analyzed by considering the effects of eliminating Regulation Q in the context of a theory that describes how interest rates are determined. The Mark-up Theory vs. the Competitive Market Theory There are several competing theories of how depository institutions determine the interest rates they charge borrowers. The two theories discussed “ Ibid., title II, sec. 204(a). 6 DECEMBER 1981 in this section have different implications for the impact of eliminating the ceiling rates on time and savings deposits specified under Regulation Q. The M ark-up Theory — Those who assert that borrowers will be charged higher interest rates due to the elimination of Regulation Q are generally using a mark-up theory: Depository institutions are presumed to determine the interest rates they charge borrowers as a mark-up over the a verage interest rate they pay on deposits. The average interest rate on deposits will rise as Regulation Q is phased out, unless market interest rates should fortuitously fall below the Regulation Q ceilings currently in effect. The mark-up theory, therefore, predicts that bor rowers will pay higher interest rates as a conse quence of the elimination of Regulation Q. The C om petitive M arket Theory — Under this theory, the interaction of several factors influencing both supply and demand determine a market inter est rate, which all lenders charge on loans with similar characteristics. Lenders can make few loans at interest rates above the market rate, since bor rowers will search for the lowest rate available. Since lenders can make all the loans they wish at the market rate, they have no incentive to lend at in terest rates below the market rate. To describe this theory in more detail, consider the determinants of the market interest rate on a particular category of credit — residential mortgage loans. Demand for residential mortgage credit is determined by personal income and the preferences of individuals for housing and for home ownership. Several factors influence the supply of residential mortgage credit. One factor is the interest rates on in vestments other than residential mortgages. If, for instance, yields rise on U.S. Treasury securities with maturities similar to those of residential mortgages, depository institutions and other suppliers of resi dential mortage credit will supply less mortgage credit at each level of the mortgage interest rate. Another important determinant of supply is the interest rate on deposits not subject to Regulation Q ceilings. For example, depository institutions may pay whatever interest rate they wish on time de posits in denominations of $100,000 or more. In the competitive market, depository institutions will bid up the interest rates they are willing to pay on deposits fr e e o f Regulation Q ceilings until these rates are sufficiently close to their lending rates to eliminate the incentives to make additional loans. Consequently, it is the interest rate that depository DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS Table 1 DIDC Changes in Interest Rate Ceilings on Deposits Date of Meeting May 29, 1980 Effective Date of Change June 2, 1980 Type of Deposit Nature of Change Small savers certificates (time deposits with m aturities of 30 months or more, no m inimum denom ination) P rio r ceilings: Com mercial banks were perm itted to pay the yield on 21/2-year Treasury securities less 75 basis points, and th rift institutions were perm itted to pay 25 basis points more than com m ercial banks. The maximum interest rates perm issible, however, were 11.75 percent at com m ercial banks and 12 percent at th rift institutions. Changes: Ceiling rates relative to yield on 21/2-year Treasury securities raised 50 basis points. Ceiling rates w ill not fall below 9.25 percent at com m ercial banks or 9.50 percent at th rift in stitu tion s. The maximum ceiling rates of 11.75 and 12 percent were retained. May 29, 1980 June 5, 1980 Money market certificates (time deposits in denom inations of $10,000 or more with m aturities of six months) Raised the ceiling rate from the discount yield on sixmonth Treasury bills established at the most recent auction to that rate plus 25 basis points at both com m ercial banks and th rift in stitu tion s.1 O ctober 9, 1980 December 31, 1980 NOW accounts Set the ceiling rate on NOW accounts at 5.25 percent fo r com m e rcial banks, m utual savings banks, and savings and loan associations. The ceilin g rate on interest-bearing checkable deposits was 5 percent until December 31, 1980. June 25, 1981 August 1, 1981 Small savers certificates Eliminated caps on these ceiling rates of 11.75 percent at com m ercial banks and 12 percent at th rift institu tions. W ith the caps lifted, th rift institutions may pay the yield on 21/2-year Treasury securities, and com mercial banks may pay 25 basis points less. September 3, 1981 O ctober 1, 1981 All Savers Certificates Adopted rules fo r All Savers Certificates specified in the Econom ic Recovery Act of 1981. September 22,1981 November 1, 1981 Money market certificates Depository institutions are now perm itted to pay the higher of the discount rate on six-m onth Treasury bills at the most recent auction, plus 25 basis points, or the average auction rate in the past four weeks, plus 25 basis points. S ep tem be r22 ,1981 January 1, 1982 IRA/Keogh accounts Created a new category of IRA/Keogh account with m inim um m aturity of 11/2 years, no regulated interest rate ceiling, and no m inim um denom ination. ’ Other changes in the ceiling rate on money m arket certificates are relevant when the yield on six-m onth Treasury bills falls below 8.75 percent. institutions pay on deposits u n co n stra in ed by Regulation Q that influences the interest rates they charge on loans. rates for borrowers, if individuals are induced to save more of their income in response to the higher inter est rates available on deposits. Under the competitive market theory, a change in Regulation Q ceilings will affect interest rates on residential mortgages only if it affects interest rates on unregulated deposits or on alternative invest ments. One implication of this theory is that elim inating Regulation Q ceilings might reduce interest The effects of eliminating Regulation Q under the competitive market theory are in sharp contrast to the effects under the mark-up theory. The mark-up theory predicts that the elimination of Regulation Q would cause interest rates paid by borrowers to rise, while the competitive market theory suggests that 7 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 C h a rt 2 Comparison of Mortgage Interest Rate with Cost of Funds to S&Ls and 10-Year Treasury Bond Yield 11 Per c ent P e r c e nt [1 S e m ia n n u a l a v e ra g e s Latest d a ta plotted: 1st h a lf 1981 interest rates on loans would either be unaffected or would decline. W hat’s the E viden ce? — Chart 2 presents some evidence on whether U.S. interest rates on residen tial mortgages are determined according to the mark up or competitive market theory. The average cost of funds to savings and loan associations (S&Ls) over Digitized for 8 FRASER six-month intervals since 1966 is shown in conjunc tion with the average interest rate on conventional residential mortgages and the average yield on U.S. Treasury securities with maturities of 10 years over the same six-month periods. Chart 2 clearly indicates that there is no fixed mark-up between the average cost of funds to S&Ls FEDERAL RESERVE BANK OF ST. LOUIS and the average interest rate on residential mort gages. The difference between the average mort gage interest rate and the average cost of funds to S&Ls has varied widely, from 165 basis points in the first half of 1966 to 386 basis points in the first half of 1980. Chart 2 shows that there is a much closer rela tionship between the average mortgage interest rate and the average yield on U.S. Treasury securities with maturities of 10 years than the relationship between the mortgage interest rate and the average cost of funds.12 The difference between the mort gage interest rate and the yield on 10-year Treasury bonds has a standard deviation of 27 basis points, compared with a standard deviation of 59 basis points for the difference between the mortgage interest rate and the average cost of funds to S&Ls. These comparisons provide evidence that interest rates on residential mortgages are determined in a competitive credit market. Homebuyers must pay interest rates on mortgages that are competitive with yields on alternative investments in order to receive credit. Chart 3 presents additional evidence on whether interest rates are determined according to the mark up or the competitive market theory. The difference between the prime loan rate charged by commercial banks and the average interest rate they pay their depositors on total time and savings deposits is highly variable, ranging from 49 basis points in 1972 to 461 basis points in 1980. Thus, once again, there appears to be no fixed mark-up between the prime rate and the average interest rate paid on time and savings deposits. There is a much closer relationship, however, between the prime loan rate and the rate that com mercial banks pay on their three-month certificates of deposit, which are free of Regulation Q ceilings. The differential between the prime rate and the three-month certificate of deposit yield has a standard deviation of 73 basis points, compared with a standard deviation of 144 basis points for the differential between the prime rate and the average interest rate paid on time and savings deposits. Again, the interest rate relationships presented in 12T he conclusion that mortgage interest rates are more closely related to the yield on U.S. Treasury securities wit!) maturities of 10 years than to the average cost of funds to S& Ls has been confirmed using regression analysis. See Thomas Mayer and Harold Nathan, “ Mortgage Rates and Regulation Q ,” Working Paper S eries No. 171 (Departm ent of Econom ics, University o f California at Davis, Ju ly 1981). DECEMBER 1981 chart 3 are more consistent with the competitive market theory than with the mark-up theory. Is the Mortgage Market Separate From O ther Credit Markets? Despite the above evidence suggesting that in terest rates charged borrowers are more closely related to market interest rates uncontrolled by Regulation Q than to the average interest rates paid on deposits, the possibility that the elimination of Regulation Q would increase the interest rates paid by one class of borrowers — homebuyers — has not been ruled out. This possibility, produced by certain regulations and tax incentives affecting thrift insti tutions, is discussed in this section. Since 1966, the existence of higher ceilings on their deposit interest rates have given thrift insti tutions an advantage over commercial banks in at tracting deposits. At the same time, however, thrift institutions are faced with regulations that limit their investments in types of assets other than mortgages. In addition to these regulations, thrift institutions are also given tax incentives to specialize in resi dential mortgage lending: The deductions from gross income allocated to bad debt reserves, which are, therefore, not subject to income tax, are larger for institutions that invest more of their assets in mortgages. As a result of the higher ceiling interest rates allowable (which attract deposits) and the regula tions and tax incentives that favor mortgage lending, thrift institutions might charge residential mortgage lending rates that are below market interest rates (on securities with characteristics similar to residential mortgages). Eliminating Regulation Q would re move the advantage that thrift institutions have in attracting deposits. As a result, the share of credit channeled to residential mortgages would decline and interest rates on residential mortgages would rise relative to other interest rates. This result is unlikely for several reasons. First, the reactions by other suppliers of credit would tend to offset these effects, as long as non-thrift institu tions are making residential mortgage loans as well. If thrift institutions increase the amount of mortgage credit they offer at prevailing interest rates, other lenders will simply reduce the quantity of resi dential mortgage credit they supply, shifting their investments to other sectors of the credit market. The net result might be no change in mortgage in terest rates, but an increase in the proportion of 9 DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS C h art 3 Relationship B etw een the Commercial Bank Prime Rate and Selected Deposit Interest Rates Percent Commercial bank prime ra Market yield on 3-month C D s / Average interest rate on ^ time and savings de positsii 1972 1973 l i A t com m ercial b a n k s Latest d a ta plotted: 1 9 80 residential mortgage loans made by thrift institu tions relative to non-thrift institutions. Of course, it is possible that the increase in the supply of mortgage credit by thrifts may not be fu lly offset by reductions in supply by other lenders. Digitized for 10 FRASER Again, however, an increase in the net supply of residential mortgage credit would not necessarily depress mortgage interest rates relative to yields on alternative investments. The reason is that pre dictable adjustments in the demand for credit would FEDERAL RESERVE BANK OF ST. LOUIS tend to offset the effects of this shift in the supply on mortgage interest rates. Suppose that, initially, interest rates on residential mortgages are decreased relative to other market interest rates due to an increase in the supply of deposits and mortgage loans at thrift institutions. This triggers increases in the quantity of mortgage credit demanded at pre vailing mortgage interest rates until these rates are, once again, in line with other interest rates. There are a variety of reactions by individuals that would cause an increase in demand for mortgage credit. For example, those seeking to borrow to invest in business firms would take out second mortgages on their homes rather than seek business loans at commercial banks. Also, individuals buying homes would obtain mortgages with smaller percentage downpayments, and invest their wealth instead at interest rates higher than the rates they pay on mortgages. There is a simple method to test whether the residential mortgage market is truly separate from other credit markets. We can determine this by ex amining the correlation between the difference of the average mortgage interest rate and the yield on 10-year Treasury bonds with the rate of growth in time and savings deposits at mutual savings banks and savings and loan associations. If the correlation is significantly negative — if the spread between the mortgage interest rate and the 10-year bond rate tends to narrow when time and savings deposits at thrift institutions grow at a faster rate — the resi dential mortgage market is, to some extent, sepa rated from other credit markets. When their deposits increase rapidly, thrift institutions reduce the mortgage interest rate relative to other interest rates in order to acquire enough residential mortgages to retain the tax advantages from specializing in mortgage lending. In fact, the correlation between the interest rate spread and the growth rate of time and savings deposits at thrift institutions is positive. Using monthly observations from January 1968 through July 1981, the correlation coefficient is 0.234, which is statistically significant at the one percent level. Using quarterly averages for 1/1968 through 11/1981, the correlation coefficient is 0.262, which is not sta tistically significant at the five percent level. This result confirms the conclusion reached in the previous section. The competitive market theory is consistent with the actual behavior of interest rates. DECEMBER 1981 Therefore, eliminating Regulation Q would not affect mortgage interest rates adversely. IM P L IC A T IO N S F O R IN TEREST RATES O F ALL SAVERS C E R T IF IC A T E S The analysis presented above has implications for the effects of the All Savers Certificate (ASC) pro gram on interest rates paid by borrowers at depos itory institutions. ASCs are special time deposits with maturities of one year. The ceiling rate on ASCs is equal to 70 percent of the average yield set in the most recent auction of one-year Treasury securities.13 Individuals may declare up to $1,000 in interest on ASCs tax free (up to $2,000 on joint returns). Depository institutions issuing ASCs are receiv ing deposits at interest rates below market rates. For individuals subject to relatively high marginal tax rates, the tax-free yield on ASCs is greater than the after-tax return on many alternative investments. Depository institutions are required to invest 75 percent of the funds raised by issuing ASCs in housing and agricultural loans. Details of the legis lation and the regulations issued to implement the program provide depository institutions with a great deal of flexibility in meeting these investment re quirements. The objectives for establishing the ASC program, however, included increasing the amount of credit available to the housing and agricultural sectors of the credit market. The structure of regulations under the ASC pro gram is similar to that for promoting mortgage lending by thrift institutions. Differentials in Reg ulation Q ceilings have given thrift institutions advantages in attracting deposits, and thrifts have been given tax incentives to specialize in mortgage lending. All depository institutions that take ad vantage of the ASC program to attract deposits at interest rates below market rates are required to allocate increases in their assets to certain sectors of the credit market. The analysis developed earlier indicates that the inflow of deposits at institutions given inducements to specialize in mortgage lending has not lowered the level of mortgage interest rates relative to other 13Treasury secrities with maturities of one year are generally auctioned every four w eeks on a Thursday. T h e average yield on a Thursday auction determines the new ceilin g rate on All Savers Certificates beginning the following Monday. 11 DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS rates. The ASC program is therefore unlikely to have any appreciable impact on the interest rates charged on housing and agricultural loans relative to other interest rates. The ASC program may have some effect on the quantity of housing and agricultural loans, as depository institutions and their borrowers develop methods of classifying loans in the cate gories that will meet the investment requirements of the ASC program. The primary effects of the ASC program will be to reduce the interest costs of depository institutions and the income tax of investors. been the opening of branch offices. Since depository institutions have not been allowed to compete di rectly on the basis of interest rates they offer to pay on deposits, they have been competing indirectly by offering convenient locations for depository ser vices.16 Many branches that were profitable when Regulation Q ceilings were below market interest rates will become unprofitable when deposit inter est rate ceilings are lifted. THE E F FE C T S O F E L IM IN A T IN G R E G U L A T IO N Q O N PROFIT S O F D E P O S IT O R Y INSTITUTION S Under the directives of the Depository Institu tions Deregulation Act of 1980, the Depository In stitutions Deregulation Committee is in the process of lifting interest rate ceilings on time and savings deposits. That committee has taken some steps to raise the ceilings, but the most significant actions to eliminate the ceilings on deposit interest rates are yet to come. Eliminating Regulation Q will raise the interest rates paid to depositors relative to market interest rates. The evidence cited above indicates that bor rowers at depository institutions will not pay higher interest rates due to the elimination of Regulation Q. Decontrol of interest rates paid on deposits, there fore, will tend to reduce the net income of depository institutions. Several studies indicate, however, that the net income of depository institutions will not decline by the full amount of the increase in interest paid on deposits. Because interest rate ceilings on deposits have been below market interest rates, depository institutions have increased expenditures to attract deposits by means other than increasing interest payments on deposits.14 These non-interest expen ditures to attract deposits are estimated at between 40 and 50 percent of the direct interest expense depository institutions saved by paying only the ceiling interest rates on deposits rather than market interest rates.15 Although depository institutions can quickly eliminate some types of non-interest expenditures made to attract deposits, such as gifts of merchandise for depositors who open or add to accounts, they will incur losses in eliminating other expenditures. One major expenditure intended to attract deposits has 14Thomas E ric K ilcollin and Gerald A. Hanweck, “ Regulation Q and C om m ercial Bank P ro fitab ility ,” R e search Papers in Banking and Financial Econom ics (Board ot Governors of the Fed eral Reserve System, 1981). 15Taggart, “ E ffects o f D eposit Rate C eilin g s,” and Lew is J. Spellman, “ D eposit Ceilings and the Efficiency of Financial Interm ediation,” J o u r n a l o f Fin an ce (March 1980), pp. 129-36. Digitized for 12 FRASER C O N C L U S IO N S Some supporters of ceilings on deposit interest rates claim that eliminating the ceilings will cause depository institutions to raise the interest rates diey charge borrowers. An analysis of interest rates does not support this view. Interest rates paid by bor rowers are determined by market rates that are exempt from Regulation Q ceilings. Consequently, elimination of Regulation Q ceilings will not cause loan rates to rise, but may cause them to decline if depositors save more with higher deposit interest rates. Profits of depository institutions will not decline by the full amount of the increase in interest expense resulting from eliminating Regulation Q, since these institutions will eliminate some non interest costs that were incurred to attract deposits when Regulation Q ceilings were binding. Similar implications also hold for the effects of the All Savers Certificate program on interest rates paid by borrowers. Although depository institutions are required to invest at least 75 percent of funds raised by issuing All Savers Certificates in housing and agricultural loans, that requirement is unlikely to result in lower interest rates on such loans relative to other market interest rates. 16Law rence J. W hite, “ P rice Regulation and Quality Rivalry in a Profit-maximizing M odel: The Case o f Bank Branching,” J o u r nal o f M oney, C r e d it a n d B a n k i n g (February 1976), pp. 97-106; and W illiam M. Peterson, “ The E ffects o fln te re st Rate C eilings on the Number o f Banking Offices in the United States,” R e search Paper No. 8103 (Federal Reserve Bank o f New York, 1981). FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 A Comparison of the St. Louis Model and Two Variations: Predictive Performance and Policy Implications LAURENCE H. MEYER and CHRIS VARVARES T -M. HE St. Louis Model was first published in the Federal Reserve Bank of St. Louis Review in April 1970.1 This model, with modifications, has been used for years at the St. Louis Fed to provide alter nate scenarios for the response of inflation, output and the unemployment rate under different mone tary policy assumptions. In addition, it continues to be identified by those outside the St. Louis Federal Reserve as the model underlying the Bank’s policy prescriptions. performance and policy implications of two varia tions of the St. Louis Model — one incorporating a Phillips Curve, the other a monetarist reduced-form for inflation. Both versions outperform the St. Louis Model’s inflation predictions, and both yield nearly identical predictions and policy multipliers. This article is organized as follows: The first sec tion reviews the current version of the St. Louis Model. The second section introduces two alterna tive versions of the St. Louis-type model. The first version substitutes a simplified Phillips Curve for the St. Louis Model’s price-change equation; the second version introduces a simple reduced-form equation for inflation in place of the Phillips Curve. The third section compares the predictive perfor mance and policy implications of these three models. The final section summarizes our findings. This article has three basic themes. First, the struc ture of the St. Louis Model can be simplified and its predictive performance improved. Second, the St. Louis Model’s specification of the demand slack variable in its Phillips Curve may bias the equation’s estimate of inflation’s response to demand slack and, therefore, could yield an overly optimistic assess ment of the cost of reducing inflation in terms of the higher unemployment during the transition to a lower rate of inflation. Third, a monetarist reduced- THE C U R R E N T V E R S IO N O F THE form equation for inflation, in which inflation depends ST. L O U IS M O D E L directly on current and past monetary growth, is not The St. Louis Model consists of five estimated inconsistent with the existence of a Phillips Curve. This is demonstrated by comparing the predictive equations and a number of identities. The key equa tions are the Andersen-Jordan or St. Louis nominal The research reported here was begun w hile Laurence H. income reduced-form equation and the equation for M eyer was a Visiting Scholar at the Federal Reserve Bank of the change in the price level. There are also equa St. Louis. The results do not necessarily reflect the views of the St. Louis Federal Reserve Bank staff, nor should the models tions for the unemployment rate, the long- and short presented be view ed as new versions o f the St. Louis Model. term interest rates, the anticipated change in the The authors gratefully acknowledge financial support from the price level and the change in output. The only sig In stitu te o f Banking and F in an cial M arkets at W ashington University. nificant change since the model was introduced has been the substitution of a rate of change (or dot) 1LeonalI C. Andersen and Keith M. Carlson, “ A Monetarist Model for Econom ic Stabilization,” this R ev iew (April 1970), pp. 7-25. version of the Andersen-Jordan nominal income 13 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 reduced-form equation for the original first differ ence (or delta) version.2 The Andersen-Jordan Equation The Andersen-Jordan equation is currently speci fied in rate-of-change or dot form; compound annual rates of change are used for nominal income (Y), the money supply (M1B is the definition of money cur rently used with the St. Louis Model, M), and the high-employment level of government expenditures (G). Dots over a variable indicate compound annual rates of change. 4 (STL-1) V, = a„ + an M „ + i= () 4 i a2i G ,, i= () The parameter estimates (t-values in parentheses) for the equation estimated from 1/1955 through IV/ 1980 are as follows:3 an a io an a 12 a i3 an -an = = = = = = = R2 2.87 .46 .45 .24 .026 -.0 7 1 1.12 .44 (3.26) (4.32) (6.49) (2.51) ( .398) (-.1 2 ) (7.44) a 20 »21 a22 a23 a2j - « 2i = = = = = = .061 .048 -.0 0 1 -.0 5 -.0 6 -.0 0 3 ( ( 1.61) 1.66) (-.0 3 4 ) (-1 .9 4 ) (-1 .7 8 ) (-0 .0 3 8 ) SE = 3.6 DW = 2.04 The coefficients on the M variables approxi mately sum to unity while the coefficients on G sum approximately to zero. Thus, the estimated co2The Andersen-Jordan equation was initially reported in Leonall C. Andersen and Jerry L. Jordan, “ Monetary and Fiscal Actions: A T est o f Their Relative Importance in Econom ic Stabilization,” this Rev iew (November 1969), pp. 11-24. In that version, changes in the money supply had a strong and persistent influence on nominal income, while government expenditures had a weak initial impact that eroded to no effect at all after a single year, and tax changes had no effect at all. Benjam in Friedman noted subsequently that, when data were included through m id-1976, fiscal policy variables entered the reduced-form equation with larger, persistent effects. (Benjam in M. Friedman, “Even the St. Louis Model Now Believes In Fiscal Policy,” J o u r n a l o f M oney, C re d it, a n d B an kin g (May 1977), pp. 365-67.) In a reply, Carlson noted that the delta version o f the St. Louis equation, when estimated with data through m id-1976, suffered from heteroscedastieity. (Keith M. Carlson, “Does the St. Louis Equa tion Now Believe in Fiscal Policy?” this R eview (February 1978), pp. 13-19.) He therefore reestimated the equation in dot form, a standard approach to elim inating this problem. The dot version produced policy effects similar to the original delta version over the earlier time period: strong, persist monetary effects and weak, transitory fiscal effects. 3The equation is estimated using an Almon polynomial distrib uted lag (PD L) procedure with a fourth degree polynomial and with coefficients o f the lag distributions restricted to zero at both ends o f the lag distribution. 14 efficients support the general conclusions associated with a monetarist viewpoint: Monetary change is the key variable explaining nominal income movements while fiscal variables have at best a minor and transi tory effect.4 The Inflation Sector The inflation sector of the St. Louis Model includes three equations: a price-change version of a Phillips Curve, an identity defining the anticipated change in the price level and an equation for the long-term interest rate. The weights in the distributed lag of inflation in the long-term interest rate equation are used to construct the anticipated-price-change vari able; this variable, in turn, is included as an argu ment in the price-change equation. This structure is unnecessarily complicated. The predictive perfor mance of the model with respect to inflation can be improved with a simpler and more conventional specification of the Phillips Curve in which the weights on the distributed lag on past inflation are estimated as part of the estimation of the Phillips Curve.5 4Although the Andersen-Jordan equation has been controversial since it was first introduced, attempts to develop more eclectic versions allowing for a permanent effect o f fiscal variables on nominal income have generally been unsuccessful. For a survey o f em pirical evidence on the Andersen-Jordan equation, see Laurence H. Meyer and Robert H. Rasche, “ Em pirical E v i dence on Stabilization Policies,” in Sta b iliz a tio n Policies: L esson s f r o m the 1970s a n d I m p lic a t io n s f o r the 19H()s,” Pro ceedings of a Conference sponsored by the Center for the Study of American Business and the Federal Reserve Bank o f St. Louis, 1980, pp. 41-102. There is, on the other hand, considerable evi dence suggesting that simple reduced forms may yield unrelia ble estimates o f policy multipliers. See, for example, Franco Modigliani and Albert Ando, “ Im pact o f Fiscal Actions on Aggregate Incom e and the M onetarist Controversy,” in Jerom e L. Stein, ed., M on etar ism (Amsterdam, North Holland; 1976), pp. l? -4 2 ; and Stephen M. Goldfeld and Alan S. Blinder, “ Some Implications o f Endogenous Stabilization Policy',” B r o o k in g s P apers on E c o n o m i c A ctivity (3:1972), pp. 585-640. 5Many of the criticism s o f the St. Louis inflation sector discussed in this section were initially raised in comments by Nordhaus and Gordon at the time the St. Louis Model was presented at an N B E R conference on price determination in 1970. See Otto E ck stein, ed., T he E c o n o m e t r ic s o f P rice D eter m in ation , Proceed ings of a Conference sponsored by the Board o f Governors o f the Federal Reserve System and the Social Science Research Council, Federal Reserve System, June 1972. The St. Louis Model was described in the volume in Leonall C. Andersen and Keith M. Carlson, “An Econom etric Analysis o f the Relation o f Monetary Variables to the Behavior of Prices and Unemploy m ent,” pp. 166-183. Comments on the St. Louis Model’s model ling of inflation appear in W illiam D. Nordhaus, “Recent Devel opments of Price Dynam ics,” pp. 16-49; and in Robert J. Gordon’s discussion of the Andersen and Carlson paper, pp. 202 - 12. FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 The price-change equation — The price-change equation in the St. Louis Model is: produce an upward bias in the model’s response of inflation to monetary change. 5 A second unusual feature is that it uses a different demand slack variable than that used in most empiri cal Phillips Curves. Generally, either the unem ployment rate or the (percentage) GNP gap (poten tial or full-employment output minus actual output) is used as the measure of demand slack. The St. Louis demand slack variable (DSL), on the other hand, is defined as (STL-2) A P * t = bo + i hi, DSLt-i + i= 0 b2 AP-\ where A is the first difference operator, DSL is the demand slack variable (defined below), APA is the anticipated change in the price level (also defined below), and AP*t, the change in the price level, is specified as (STL-2a) A P * t = AP , • X M, where X is the level of real GNP. The explanation for this form of the price change variable will be given below. The parameter estimates when the equation is estimated over the period I/1955-IV/1980 are as follows: 1)2 = .65 (.77) = .012 (.53) = .028 (3.02) = .036 (5.59) = .038 (3.73) = .033 (2.97) = .019 (2.60) = .166 (5.93) = 1.29 (25.50) R2 = bo bio bn bi2 1)1.3 bu 1)15 Sb„ .88 SE = Although the price-change equation is, in essence, a Phillips Curve equation, it has several unusual features. First, it explains the first difference in the price level (the implicit GNP deflator), while Phillips Curves are typically specified in terms of the inflation rate or the rate of change in nominal wages.6 Its delta form reflects the now-abandoned delta specification of the Andersen-Jordan equation; it made the pricechange equation dimensionally compatible with the income-change equation, allowing the change in output to be solved for via a simple “identity.” Since the Andersen-Jordan equation is now used in dot form, the retention of the delta form for the pricechange equation is unnecessary. Moreover, the delta specification, due to the possibility of heteroscedasticity, could produce an upward bias on the coeffi cients in that equation, including the coefficients on both the demand slack variable and on the anticipated-price-change variable.7 These impacts would sMore precisely, the price change variable is the change in the price level times lagged real output. See equation STL-2a above. 7This possible source o f bias in the St. Louis price-change equa tion was noted by Gordon in his comments on the St. Louis (STL-2b) DSL, = AY, - (POTRT, - X M), where POTRT is the level of potential output as measured by the Rasche-Tatom series.8 This specifi cation of the demand slack variable may seriously bias upward the equation’s estimate of the response of inflation to demand slack, inasmuch as it allows changes in nominal income associated with changes in the price level to “explain” changes in the price level.9 The sum of the coefficients on the demand slack variable determines the degree to which decelera tions in monetary growth are initially reflected in declines in the rate of growth of output and hence increases in the unemployment rate. Meyer and Rasche report simulations of the St. Louis Model with different values of this parameter (its value based on a sample through 1/1975 and its value based on a sample through IV/1979 where the sum is three times larger) and demonstrate the dramatic differences in the implied responses of output and the unemployment rate to monetary decelerations.10 M odel in the E ckstein volume. Gordon argued that the results of the price-change equation “are plagued by heteroscedastieity” (p. 209). In response to the presence o f heteroscedasticity, the nominal incom e was changed from a delta to a dot specification, although the price-change equation, where heteroscedasticity may have been more of a problem,was left in first difference form. 8S ee Robert H. Rasche and John A. Tatom, “ Energy Resources and Potential Output,” this R ev iew (June 1977), pp. 10-24, for a discussion of this series. 9T he possibility that the St. Louis specification of the price-change equation yields an upward biased estim ate of the response of inflation to demand slack was suggested by the remarkable behavior o f the sum o f the estim ated coefficients on the demand slack variable as additional years o f data were added to the sample period during the 1970s. After 1975, the estimated coefficient begins to rise as more data is included; by the end of 1979, the coefficient is almost three times its value for the sample period ending before 1975. This pattern is consistent with what would be expected if the specification yielded biased estimates for the reason suggested above. This bias would be expected to becom e more serious during a period w here changes in nominal incom e w ere dominated by changes in the price level. 10See Laurence H. M eyer and Robert H. Rasche, “On the Costs and Benefits o f Anti-Inflation P olicies,” this R eview (February 1980), pp. 3-14. 15 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 The long-term interest rate equation —The equa tion for the long-term interest rate (RL) is: 16 (STL-.3) RLt — Co + ci \lt + c'2 Zt + — c.ii Xt-i i=0 16 + i! C4i (P,.i/(U^UFt.i)), i= 0 where X is the rate of change in real GNP, Z is a dummy variable, allowing for a shift in the constant term over the period, U is the unemployment rate, and U F is a measure of the rate of unemployment at “full employment.” The parameter estimates for the long-term interest rate equation are as follows: C'O Cl C'2 -C3i -C ji = .82 = .02 = .82 = .29 = 1.04 R2 = (1.42) (.65) (2.77) (1.69) (14.23) .89 SE The measure of expected inflation in the above equation is a distributed lag on past inflation adjusted for the level of demand pressure as proxied by the ratio of the unemployment rate to the full-employment rate of unemployment (UF) where the latter is measured by series developed at the Council of Economic Advisors. This equation not only provides predictions of the long-term interest rate, it also provides the weights, the C4i coefficients, used in the anticipated-price-change equation. There are a number of questionable features of this long-term interest rate equation, particularly related to its role in providing the weights for an expected price-change variable. First, the weighted sum of current and past inflation rates can be viewed as a measure of the expected inflation rate only if we assume that a one percentage point increase in the expected inflation rate increases the long-term interest rate by one percentage point. We cannot, however, separate out the weights that convert current and past inflation rates into the expected rate of inflation and the coefficient that translates an increase in the expected inflation rate into an increase in the long term interest rate. Recent work on the implications of specific tax structures for interest rate behavior in inflationary periods indicates that the simple Fisher ian view that a percentage point increase in the expected inflation rate raises the long-term interest rate by a percentage point is no longer so obvious.11 “ See, for exam ple, Martin Feld stein, “ Inflation, Incom e Taxes and the Rate o f Interest: A Theoretical Analysis,” A m erican Digitized for 16 FRASER Second, the expected-price-change variable that is derived from a long-term bond equation is likely to relate to a much longer horizon (the average term to maturity of long-term bonds) than is relevant to the formation of price expectations in the context of the Phillips Curve (the current period or at most an average of price change expected over the average length of contracts, implicit and explicit). This differ ence in horizon may affect the number of relevant lags and the weighting applied to past inflation. There is one additional question about the specifi cation of the long-term interest rate equation. One can derive a somewhat similar equation by begin ning with a money demand equation in which the demand for money depends on the long-term interest rate and current and past rates of inflation and by solving that equation for the long-term interest rate as a function of the level of real money balances, the level of real output, and current and past rates of inflation. However, the long-term rate would depend on the level of real money balances rather than the rate of change in nominal money balances and on the level of real income rather than the rate of change in real income. Finally, the Durbin-Watson statistic is very low, suggesting serious serial correlation ofthe residuals. Reestimating the equation using the CochraneOrcutt technique to correct for first-order serial cor relation yields quite different parameter estimates for the money and output variables and an unimpres sive equation in terms of the significance of key parameter values. The anticipated-price-change equation — The equation for anticipated price change (APA) is an ■identity, given by 17 (S T L -4 ) APA, = Y m ( [ 1 c 4i (P,-i/((Jt.i/UFt.,) ) .01 i= l + 1 ] 25 - 1 ). This seemingly complicated equation transforms the weighted distributed lag on current and past inflation into the first difference of the price variable used in the St. Louis Model. This price change variable is not the first difference ofthe implicit price deflator; it is, instead, the first difference in the implicit de flator multiplied by the lagged value of the level of real output. This particular form of the price change variable is necessary because ofthe way that output is determined in the model. E c o n o m i c R ev iew (D ecem ber 1976;, pp. 809-20; and John A. Tatom and Jam es E. Turley, “Inflation and Taxes: D isincentives for Capital Form ation,” this R e v ie w (January 1978), pp. 2-8. FEDERAL RESERVE BANK OF ST. LOUIS To determine the dynamics associated with the price-change equation and, in particular, whether the St. Louis price-change equation implies a long-run trade-off or a vertical Phillips Curve, one must solve for the implied sum of the coefficients on lagged price changes. The equation for the change in the price level can be expressed directly as a function of demand slack and a distributed lag on past price changes: 5 17 A P *, = ho + 1 hi, DSL,., + i =0 X b2i' AP,.i. i= l The sum of the coefficients on past price changes can in turn be related to the parameter on the antieipated-priee-change variable in the price-change equation, STL-2, and the coefficients on past inflation in the long-term interest rate equation. V l) 2i = h i ■ { 1 C4i / M ean (U /U F) } The 2b£i term equals 1.13, based on the St. Louis Model’s estimates for 1)2 and the C4i parameters. The fact that the sum of coefficients on past price changes exceeds unity results in a dynamic instability in long-run simulations with the Model and a more rapid response of inflation to monetary change than if this sum were constrained to unity. This feature of the price-change equation reinforces the influence of the upward bias in the coefficient on the demand slack variable. The Unemployment Gap Equation The unemployment rate (U) is determined by the following equation: (S T L - 5 ) ITC A P, = do G A P , + d i G A IV i, where UGAP = LI — UF and GAP is the percentage gap between potential output and actual output (GAP = ((POTRT - X)/POTRT)* 100). The unemployment rate is then calculated from the identity, (S T L -6 ) U, = U F, + U G A P,. The parameter estimates for STL-5, based on the sample period I/1955-IV/1980 are: do = .0 2 4 (.3 8 ) ill = .4 5 (7.2) R2 = .7 8 SE = .5 5 DW = .3 8 The pattern of coefficients on the gap variables in this equation are different from what might have been expected. The coefficient on the GAP variable in the contemporaneous period is essentially zero, DECEMBER 1981 implying that a change in the level of output relative to potential output has no impact on the unemploy ment rate in the same quarter. In addition, the Durbin-Watson statistic is low, suggesting the possible omission of other important explanatory variables. The Short-Term Interest Rate Equation We ignore the remaining equation in the St. Louis Model, the equation for the short-term interest rate (4- to 6-month commercial paper rate). This variable does not appear elsewhere in the model and we are not interested in the model’s predictions for interest rates. The Output Identity The change in output is determined in the St. Louis Model via an “identity.” Using first differ ences, AY can be expressed as AY, = P,_i AX, + X,.i AP, + AX, AP,. The price change variable in the St. Louis Model is thus not AP, but rather X-iAP, the dollar change in total spending due to price changes (ignoring the interaction term, AX AP). The “change-in-output” variable in the St. Louis Model is then determined by an approximation to the actual identity since the interaction term is excluded. Thus the change in output in the St. Louis Model, P-i AX, is defined by (STL-7) P,.| AX, = AY, - X,.i AP,. RE FR A IN S O N A ST. L O U IS M O D E L T H E M E : P H IL L IP S C U R V E A N D M O N E T A R IST R E D U C E D - F O R M A PPRO A CH ES TO T H E IN F L A T IO N RATE In this section, we present two variants of the St. Louis Model. The two versions differ from each other only in the equation used to explain the infla tion rate. The first version includes a fairly conven tional Phillips Curve and the second utilizes a monetarist reduced form instead. Thus, the inflation sector of the St. Louis Model is collapsed to a single equation in each of these two alternative models. Each of the revised versions includes the AndersenJordan equation and an unemployment equation. To avoid the appearance that either of these variants are intended to or actually have superceded the St. Louis Model previously presented, the two versions are labeled UCITYPC and UCITYRF, designating 17 FEDERAL RESERVE BANK OF ST. LOUIS that they were developed in University City (alias UCITY), a suburb immediately west of the city of St. Louis and adjacent to Washington University. This is intended to remind the reader that these versions are close to the St. Louis Model, but not identical o it. Of course, the PC and RF refer to the different] ting feature of the two versions, the Phillips C irve (PC) or the monetarist reduced-form (RF) equation used to explain the rate of inflation. First, we present the equations that the two versions have in common: the reduced-form equa tion for nominal income, the equation for the unem ployment gap, and the identity that converts pre dicted values for nominal income and price level into predictions for output. Then, we will detail the two alternative specifications of the inflation equation. The Identity Relating Nominal Income, Output and the Price Level The relation between nominal income (Y), output (X) and the price level (P) can be expressed by the identity, Y = FX. We wish to avoid the use of an approximation to solve for output, as the current version of the St. Louis Model does. The model will yield solutions for the rate of change in both nominal income and the price level. However, the equation, Y = X + P, is only an approximation when the dot variables are measured by compound annual rates of change, an approximation that becomes poorer as the size of the rates of change increases. To solve this problem, the rates of change are defined as changes in the logs of Y, X, and P (delta log specification). Taking logs and then first differences of the equation, Y = P X, yields the identity, (U C IT Y -7 ) A in Y = A in X + A in F . The Andersen-Jordan and inflation equations both will be specified in terms of delta logs. The identity above will then be used to determine the change in the log of output. The Andersen-Jordan Equation There is, of course, little difference between the dot and delta log specifications of an equation. The delta log specification is given by, Digitized for18 FRASER DECEMBER 1981 4 ( l !C I T Y - l ) A i n Y, = a<> + 1 a, , Ain M ti i =0 4 + i. a 2 i Al l ! Gt-i. i =0 The parameter estimates for the sample period I/1955-IV/1980 are:12 an = ai<> = an = = = = = a i2 a 13 a i4 —a i j R2 = 2 .69 (3.26) (4.34) .45 .44 (6.59) (2.6) .24 .032 (.50) - .0 6 6 (-.5 9 ) (7.19) 1.10 .45 SE = 3.3 3 = = = = = = a 2 <> 1*21 a22 <*23 a24 —a 2 i DW - .062 (1.59) .054 (1.81) .004 (.116) —,0 52( - 2 . 0 2 ) - . 0 6 9 ( - 2 .0 1 ) ~ .0 0 1 (-- 0 .0 1 ) 2.04 The Unemployment-Gap Equation The specification of the unemployment-gap equation is unchanged from the St. Louis Model (STL-5). The only modification is that it is estimated with a correction for second order autocorrelation. The estimates for this equation are: do di = = in [i2 = = R2 = .26 .17 1.17 -.3 8 .74 (14.3) (9.5) SE = .19 DW = 2.04 where pi and p2 are the values of the rho coefficients on the first and second lagged values of the residual. Note the dramatic decline in the standard error of this equation, compared with the one in the St. Louis Model.13 The two revised versions include equation (STL-5), as reestimated above, and the identity (STL-6). The Level o f Output Because the level of output is used in the GAP variable in the Phillips Curve, we must also include 12The change in log variables are all m ultiplied by 400 prior to estimation so that they approximate annual rates of change. 13The param eter estim ates o f the revised equation are quite close to those presented in John A. Tatom, “ Econom ic Growth and Unemployment: A Reappraisal o f the Conventional View ,” this R ev iew (O ctober 1978), pp. 16-22. Tatom corrects the levels equation for first-order serial correlation and also presents a first difference equation, also with a correction for first-order serial correlation. FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 an identity to determine the level of output fiom the predicted values of the change in the log of output and last period’s level of output. (UCITY-8) X, = exp (In X,., + Ain Xt) The Phillips C urve The Phillips Curve equation uses a delta log speci fication for the rate of change in the price level, measures the demand slack in the economy with the GNP gap variable, and proxies expected inflation with a distributed lag on past rates of inflation. The latter distributed lag can also be interpreted as capturing an element of inertia due, for example, to the existence of implicit or explicit contracts. The Phillips Curve also includes the differential in the rate of increase in the producers’ price of energy relative to the rate of increase in the implicit price deflator for GNP. This variable, labeled ENERGY, is intended to capture a major source of “supply shocks” that dramatically have affected the inflation rate over a couple of periods in the data sample, in particular, during the latter part of 1973, throughout 1974 and, more recently, in 1979 and early 1980. This variable is lagged two periods, reflecting some experimentation with other simple lag patterns. The Phillips Curve also includes a dummy vari able to capture the influence of the price controls during the period from 111/1971 through 1975. The variable, labeled CONTROLS, allows for a negative impact during the first part of the period and an off setting positive influence associated with “catch-up” effects during the period after which controls were relaxed and then removed. The sum of the values the dummy variable takes on over this period is con strained so that the net price control effect on infla tion is zero. Specifically, CONTROLS is 0 up to II/ 1971, 1 from III/1971-IV/1972, .2222 in 1/1973, —.7778 from II/1973-I/1975 and 0 thereafter.14 The estimated Phillips Curve equation is ( l ’ClTY-2) Ain P, - #> + GAPn + /S2 CONTROLS, + p3 ENERGY,., 20 + 1 /Si, Ain P,., i=l 14This specification of the controls variable was borrowed from John A. Tatom at the Federal Reserve Bank o f St. Louis. The distributed lag on inflation is estimated using a third degree polynomial with no end-point restric tions. We have employed the lagged GAP in this equation, preserving the simple recursive structure of the St. Louis Model. The empirical estimates with the contemporaneous GAP were almost indistin guishable from this equation. The parameter estimates for the period 1/1955IV/1980 are: .85 -.2 2 -1 .0 8 5 .044 0» 0i 02 03 .19 .14 .097 .066 .043 .028 .020 .017 .019 .023 011 042 043 044 045 /Sifi 047 04S 049 Hi w R2 = .827 (3.14) (-4 .3 ) (-2 .6 5 ) (3.82) .030 (4.32) /Sin .037 (5.19) 0412 .044 (6.39) 0413 (5.72) .045 0414 (3.24) .051 0415 .050 (1.83) 0416 (1.2.3) .043 0117 .030 (1.1) 04 IS (1.32) .010 /Si 19 -.0 1 9 (1.87) 0420 .96 9 (14.04) i-04, = SE = 1.166 DW = (2.63) (3.24) (3.54) (3.66) (3.77) (3.9) (3.8) (2.43) (.51) (-.5 6 ) 2.10 Note that the GAP variable is highly significant, that the sum of coefficients on the past inflation rates is not significantly different from unity, and that both the controls dummy and the energy differential variables are significant. The Inflation Reduced-Form Equation The inflation reduced-form equation explains the inflation rate in terms of current and lagged values of monetary growth and the energy inflation dif ferential and controls dummy variables discussed above.15 This equation is also specified in delta log form: ' “M onetarist reduced-form equations for inflation have been em ployed for some tim e at the Federal R eserve Bank o f St. Louis. The equation was initially reported in 1976 in D enis S. Karnosky, “T he Link Betw een M oney and Prices: 1971-76,” this R e v ie w (June 1976), pp. 17-23. Som e refinements have been presented in Keith M. Carlson, “T h e Lag from M oney to P rices,” this R ev iew (O ctober 1980), pp. 3-10, and in John A. Tatom, “ Energy Prices and Short-Run Econom ic Perfor m ance,” this R ev iew (January 1981), pp. 3-17. A sim ple annual version o f a monetarist reduced-form inflation equation was presented in Jerom e L. Stein, “Inflation, Em ploym ent and Stag flation, " J o u r n a l o f M on etary E c o n o m i e s (April 1978), pp. 193228. 19 3. The St. Louis Phillips Curve uses an unusual demand slack variable, the change in nominal in come minus the lagged real GNP gap; the UCITY Phillips Curve uses the GNP gap. 19 ( UC IT Y -2') A i n P, = i yi. A l n Mf_. i=0 + y2 C O N T RO LS , + y3 E N E R G Y ,.2 . This distributed lag in monetary change is estimated using a third degree polynomial with no end-point restrictions. The parameter estimates for the sample period I/1955-IV/1980 are: yio yu yi2 yi3 yi-t yis yi6 yi? yis yi9 .039 .047 .054 .058 .061 .063 .064 .063 .062 .060 y2 y.3 R2 yiio ym yii2 yii3 ym yii5 yii6 yin yn« yiw (1.34) (2.45) (4.02) (5.19) (5.38) (5.27) (5.25) (5.36) (5.50) (5.52) Vyii = l.,04 .057 .055 .052 .050 .046 .044 .042 .041 .041 .042 (5.28) (4.81) (4.30) (3.89) (3.66) (3.59) (3.12) (2.22) (1.11) (.35) .822 SE == 1.173 DW = 1.62 The parameter estimates on the controls dummy and the energy inflation variable are both significant, and the coefficients on the monetary change variable sum to unity. The two inflation equations, LJCITY-2 and UCITY-2', perform quite similarly with respect to in-sample error, with a very slight edge to the Phillips Curve. S ummary o f D ifferences o f U C I T Y Models and the St. Louis Model A summary of the St. Louis and UCITY models is given in table 1. The differences between the St. Louis Model and the UCITY models can be sum marized as follows: 1. The nominal income and inflation equations are both specified symmetrically in delta log form in the UCITY models, allowing the change in the log of output to be solved for via an identity. In the St. Louis Model, the nominal income equation is in a rate-of-change specification, the price equation in first difference, and the change in output is solved for via an approximation. 2. The St. Louis Model employs a three-equation inflation structure. The UCITY models employ alter native single equations for inflation. Digitized for20 FRASER 5. One of the UCITY models substitutes a mone tarist reduced-form equation for inflation for the Phillips Curve. The St. Louis inflation sector is built around a price-ehange version of a Phillips Curve. 6. The unemployment equation is estimated using a correction for second-order autocorrelation in the UCITY models. It is estimated using ordinary least squares in the St. Louis Model. (33.5) - 1 .9 4 ( -4 .7 2 ) .045 (4.12) = 4. The weights on past inflation in the St. Louis Phillips Curve are derived from an equation for the long-term interest rate. In the UCITY model, the weights are estimated directly during estimation of the Phillips Curve. C O M P A R IN G T H E T H R E E M O D E L S : P R E D IC T IV E P E R F O R M A N C E A N D P O L IC Y IM P L IC A T IO N S This section compares the predictive performance and the policy implications of the three models. The results reported here bear directly on the three themes outlined at the beginning of the paper. First, in-sample and out-of-sample static simulations are used to compare the predictive performances of the St. Louis Model and the two UCITY models. Second, the responses of output, unemployment and inflation in the models to a deceleration in monetary growth are compared. Third, the two UCITY models are compared to determine whether any differences exist in their predictive performance or policy multipliers. Predictive P erform ance o f the Three Models Because the two UCITY models include two significant variables not included in the St. Louis Model — the controls dummy and the energy infla tion differential — it would not be surprising if they perform better than the St. Louis model. In order to determine the degree to which differences in predic tive performance were due to the addition of these variables, two additional versions of each UCITY model were estimated: one without the controls dummy, the other without the controls dummy and the energy-inflation differential. DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS Table 1 Summary of St. Louis and UCITY Models St. Louis Model UCITY Models P hillips Curve (1) Yt = ao + 4 ai, Mt-j i=0 + 4 4 2 a2i Gt-i i=0 A in Yt = «o + a ii A in Mt-i Same as UCITYPC model i =0 + (2) A P *, = b0 + M onetarist Reduced Form 4 2 i= 0 « 2j A in Gt-i 5 £ bii DSLh + b2 APA, i=0 (2a) AP*, = X,.t AP, (2b) DSLt = AY, - A in Pt = /3o + /3i GAPm (POTRTt - Xm ) (3) RLt = Co + c i Mt + C2 Zt 16 16 + 2 c3i Xm + S c4j (Pt-i/(U,-i/UF,-,)) i=0 i=0 A in Pt = 19 £ y ii A in Mt-i i=0 + /82 CONTROLSt + 72 CONTROLSt + (3s ENERGY,-2 + y3 ENERGYt-2 + 20 2 i= 1 4i A in Pt-i 17 (4) A P At - Y,-i( [ 2 c4i( Pt-i/{U,-,/UFt.,)).01 + 1]25 _ -,) i =1 (5) UGAPt = do GAPt + d i GAP m Same as St. Louis Model but corrected for 2nd-order auto correlation Same as UCITYPC model (6) U, = UF, + UGAP, Same as St. Louis Model Same as St. Louis Model A in Y = A in X + A in P Same as UCITYPC model Xt = exp(ln Xm + A in Xt) Same as UCITY PC model (7) Pm A X t = A Y t (8) X, - Xm + XM AP, p t-i A x t Pt-1 In-sam ple static sim ulation results — The insample root-mean-square errors (RMSEs) for inflation (P), rate of change in nominal GNP (Y), rate of change in real GNP (X), level of real GNP (X), GNP Gap (GAP), and unemployment rate (U) for the various versions of the UCITY models and for the St. Louis Model are presented in table 2. Table 3 reports the percentage declines in RMSEs in the two UCITY models compared with the St. Louis Model. The two UCITY models uniformly predict more accurately than the St. Louis Model (the sole exception being the rate of change in nominal GNP for which the equations and hence predictions are virtually identical). large and, surprisingly, is accounted for to only a minor degree by the addition of the two new vari ables, although each does marginally improve the inflation predictions. The inflation RMSEs for the St. Louis Model, UCITYPC, and UCITYRF were 2.11, 1.12 and 1.14, respectively. This translates into a reduction in the RMSE for inflation of 47 percent and 46 percent in the UCITYPC and UCITYRF models, relative to the St. Louis Model. When hath the con trols dummy and energy inflation differential vari ables were excluded, the RMSEs in the UCITYPC model increased to 1.29 and UCITYRF to 1.42, still dramatically below the RMSE in the St. Louis Model. The improvement in the inflation forecast is quite These results indicate that: (1) the inflation predie- 21 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 Table 2 Full In-Sample Static Stimulations, I/1955-IV/1980 (root-mean-square errors) St. Louis UCITYPC UCITYPC w/o ENERGY UCITYPC w/o CONTROLS w/o ENERGY UCITYRF UCITYRF w/o ENERGY UCITVRF w/o CONTROLS w/o ENERGY P 2.11 1.12 1.20 1.29 1.14 1.23 1.42 Y 3.23 3.22 3.22 3.22 3.22 3.22 3.22 X 3.24 2.98 3.06 3.08 3.07 3.13 3.14 X 8.80 8.09 8.29 8.62 8.32 8.54 8.58 GAP .79 .72 .74 .75 .75 .76 .76 U .55 .26 .27 .27 .26 .26 .27 tion in the St. Louis Model can be improved by substituting either a more traditional Phillips Curve or a monetarist reduced-form for the St. Louis Model’s price-change equation; and (2) inflation predictions with the two versions of the UCITY model are very close, not surprising given the small differences in the standard errors in the two inflation equations. The UCITY models also outperformed the St. Louis Model for the rate of change in output, the level of output, the GNP gap, and the unemployment rate, although the degree of improvement is smaller for the two output variables and GAP than for the inflation rate and the unemployment rate. For the rate of change in output, the RMSE in the St. Louis Model was 3.24, compared with 2.98 in the UCITYPC model and 3.07 in the UCITYRF model. This repre sents an improvement in the RMSEs of 5.5 percent to 8 percent in the two UCITY models, a much small er improvement than might have been expected given the margin of improvement for inflation. As in the case ofthe inflation predictions, eliminating the controls dummy and inflation differential vari ables results in a small deterioration in the quality ofthe predictions from the UCITY models, but still leaves those predictions superior to those from the St. Louis Model. Interestingly, the improvement in the predictions for the unemployment rate was about as great as for the inflation rate, surprising in compari son with the much smaller improvement in predic tions ofthe GAP, but less surprising in light ofthe particularly poor statistical quality o fth e St. Louis unemployment equation. O ut-of-sam ple static foreca sts — The three models were re-estimated over the shorter period, 1/1955-1V/1976, and static forecasts were made for Digitized for22 FRASER Table 3 Percent Reduction in In-Sample UCITY RMSEs (relative to St. Louis Model) UCITYPC UCITYRF 46.0% p 46.9% X 8.0 5.2 X 8.1 5.5 GAP U 8.9 5.1 49.1 49.1 the period I/1977-IV71980. The results of the outof-sample static forecasts were consistent with the in-sample results. The two UCITY models again outperformed the St. Louis Model for all variables (except nominal income, of course). The improve ment for inflation was somewhat smaller (33 percent and 17 percent for UCITYPC and UCITYRF, respec tively) while the improvement for the output and GAP variables was somewhat larger (9 percent to 15 percent in the UCITY models) than in the case of the in-sample results. Once again, the unemploy ment rate predictions for the St. Louis Model were poor compared with the UCITY results. The out-ofsample RMSEs for the various variables are reported in table 4, and the percent improvement in RMSEs in the UCITY models is given in table 5. The Response o f Output, Unemployment and Inflation to Monetary C hange in the T hree Models The UCITY models were developed, in part, to improve the predictions of inflation, output and FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 Table 4 Out-of-Sample Static Simulations: 1/1977-IV/1980 (root mean square errors) St. Louis UCITYPC UCITYPC w/o CONTROLS UCITYPC w/o CONTROLS w/o ENERGY UCITYRF UCITYRF w/o CONTROLS UCITYRF w/o CONTROLS w/o ENERGY P 2.12 1.43 1.51 1.67 1.77 1.93 1.81 S' 3.78 3.80 3.80 3.80 3.80 3.80 3.80 X 3.92 3.34 3.58 3.47 3.58 3.84 3.55 X 14.36 12.12 13.00 12.59 13.04 14.04 12.92 GAP .95 .81 .87 .84 .87 .93 .86 U .59 .28 .29 .29 .30 .32 .30 Table 5 Percent Reduction in Out-of-Sample UCITY RMSEs (relative to St. Louis Model) UCITYPC UCITYRF 16.5% P 32.5% X 14.8 8.7 X 15.6 9.2 GAP 14.7 8.4 U 52.5 49.2 unemployment from those available using the St. Louis Model presented here. Also of interest are the differences in the policy simulations obtained using the three models. For the policy simulations, CEA projections for potential output and for high employment govern ment expenditures were used for the period from I/1981-IV/1984. Two alternative monetary growth rates were used: as the “base” series, we used a constant rate of 5 percent per year, for the “policy” series we used 2 percent per year. We then com puted the differences in the rates of change of nominal and real income, and differences in the level of real GNP, in the GNP gap, in the unemploy ment rate, and in inflation between the base and policy simulations. The results are reported in tables 6, 7 and 8 for the inflation rate, the rate of change in real output and the unemployment rate. The figures reported in each case are the values in the policy run minus the values in the base run. The results confirmed our expectations about the direction of the differences, but the magnitude of the differences between the St. Louis and UCITY simulations were somewhat smaller than expected. Inflation declines more rapidly in the St. Louis Model and, as a consequence, the decline in the rate of growth of output and the increase in unemploy ment are smaller in the St. Louis Model. For the inflation rate, all three models’ projections are close during the first year, with inflation falling about 0.4 percentage points. By the end of the second year, inflation has fallen 1.8 percentage points in the St. Louis Model, compared with only 1.2 in the two UCITY models. By the end of the fourth year, inflation has fallen by 4.0 percentage points in the St. Louis Model compared with 2.8 and 2.9 per centage points in the UCITY models. Thus, while inflation has fallen more rapidly in the St. Louis Model, the decline by the end of the fourth year exceeds the equilibrium response, implying a ten dency to overshoot. In the St. Louis Model, the rate of increase in out put declines for the first 12 quarters, the decline exceeding 2 percent per year for the first 6 quarters. In the two UCITY models, the rate of increase in out put is lower throughout the 16-quarter simulation horizon, by 2 percent per year or more for eight quarters. The unemployment results indicate that the monetary deceleration raises the unemployment rate for 16 consecutive quarters for each model, but that unemployment is about 0.6 percentage points higher in the two UCITY models at the end of the simulation horizon compared with the St. Louis Model. 23 Table 6 Table 7 Dynamic Simulation Results: Inflation Rate (difference between the 2 percent and 5 percent simulations) Dynamic Simulation Results: Rate of Change in Real GNP (difference between the 2 percent and 5 percent simulations) Date St. Louis UCITYPC UCITYRF 1981 I -.0 1 -.0 4 -.0 4 Date II -.0 7 -.0 6 -.1 2 St. Louis UCITYPC III -.2 0 -.2 0 - .2 4 IV -.4 3 -.3 8 1982 I -.7 5 II UCITYRF 1981 I -1 .1 7 -1 .5 7 -1 .3 3 II -2.51 -2 .5 5 -2 .5 7 -.4 0 III -3 .2 2 -3 .0 8 -3 .1 5 -.5 9 -.5 7 IV -3 .0 2 -3 .0 3 -3 .0 8 -1.11 -.8 0 - .7 7 1982 I -2 .4 4 -2 .6 3 -2 .7 2 III -1 .4 8 -1 .0 2 -.9 8 II -2 .0 5 -2 .4 3 -2 .5 3 IV -1 .8 4 -1 .2 4 -1 .2 0 III -1 .6 6 -2.21 -2 .3 2 -1 .2 9 -1 .9 8 -2 .1 0 -1 .8 8 1983 I -2 .1 7 -1 .4 7 -1 .4 2 IV II -2 .4 9 -1 .6 9 -1 .6 4 1983 I -.9 7 -1 .7 5 III -2 .7 9 -1 .9 0 -1 .8 6 II -.6 5 -1 .5 3 -1 .6 5 IV -3 .0 6 —2.11 -2 .0 8 III -.3 4 -1.31 -1 .4 4 1984 I -3 .3 2 -2 .3 2 -2 .2 8 IV -.0 5 -1 .1 2 -1 .2 2 II -3 .5 5 -2 .5 2 -2 .4 8 1984 I .02 -.9 0 -1 .0 2 III -3 .6 7 -2 .7 2 -2 .6 6 II .41 -.7 0 -.8 2 IV -3 .9 5 -2 .9 2 -2 .8 2 III .63 -.5 1 -.6 4 IV .83 -.3 1 -.4 8 C om paring the Predictive Perform ance and Policy Implications o f the Phillips C urve and Monetarist Reduced-Form Inflation Equations There is a considerable literature that views the Phillips Curve and the monetarist reduced form for inflation as mutually exclusive, alternative inflation equations.16 Generally, these “competitive” alterna tive approaches are tested by investigating the con sequences of adding monetary change to a Phillips Curve or introducing the unemployment rate into a monetarist reduced form.17 l6See, for exam ple, Keith M. Carlson, “ Inflation, Unemployment, and M oney: C om paring the E v id e n ce from Tw o Sim p le M odels,” this R ev iew (Septem ber 1978), pp. 2-6; and John A. Tatom, “ Does the Stage of the Business Cycle Affect the Infla tion R ate?” this R eview (Septem ber 1978), pp. 7-15; and Stein, “ Inflation, Em ploym ent, and Stagflation.” 17Tests o f this kind have been reported by Franco M odigliani and Lucas Papademos, “Targets for M onetary Policy in the Coming Years,” B r o o k in g s P apers on E c o n o m i c Activity (1:1975), pp. 141-63; George L. Perry, “ Slowing the W age-Price Spiral: The M acroeconom ics V iew ,” B r o o k in g s P a p e rs on E c o n o m i c A ctiv ity (2:1978), pp. 259-91; Stein, “ Inflation, Em ploym ent and Stag- Digitized for24 FRASER We do not view the Phillips Curve and monetarist reduced form as mutually exclusive, alternative models of the inflation process but rather as structural vs. reduced form approaches to explaining inflation. Because we view the two inflation equations as reasonable alternative specifications, we find no value in tests that add monetary change variables to the Phillips Curve or unemployment rates to the monetarist reduced form. Such experiments mix structural and reduced-form equations. We would not expect to be able to obtain significant coefficients on both monetary change and unemployment rates in an inflation equation, and, consequently, no such experiments were conducted. Instead, we compared the two inflation equations individually and as alter native components of a St. Louis-type model; we found that the two inflation equations were virtually indistinguishable in predictive performance and policy implications. First, when the single equation performance of the Phillips Curve and monetarist reduced form flation;” and John A. Tatom, “What E v er Happened to the Phillips Curve?” Federal Reserve Bank o f St. Louis R es ea rch Papers, No. 81-008. DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS Table 8 Dynamic Simulation Results: Unemployment Rate (difference between the 2 percent and 5 percent simulations) Date St. Louis 1981 I 0 UCITYPC UCITYRF 0.15 0.08 II .15 0.40 0.30 III .41 0.67 0.59 IV .78 0.94 0.90 1982 I .91 1.19 1.17 II 1.34 1.42 1.41 III 1.54 1.63 1.63 IV 1.70 1.81 1.82 1983 I 1.83 1.97 1.99 II 1.93 2.11 2.13 III 1.99 2.23 2.26 IV 2.03 2.33 2.36 2.44 1984 I 2.04 2.41 II 2.02 2.48 2.50 III 2.00 2.53 2.56 IV 1.94 2.56 2.60 were compared, the Phillips Curve and monetarist reduced forms yielded standard errors of 1.166 and 1.173, respectively. Thus the two equations fit the data almost equally well. Note the high level of significance of the key variables in both equations — the gap and the sum of the coefficients of past infla tion in the Phillips Curve and on the sum of the coefficients of monetary change in the monetarist reduced form. Second, the in-sample and out-of-sample static forecasts of the two UCITY models were compared, the only difference beingthatone includes a Phillips Curve while the other includes a monetarist reduced form. Looking at tables 2 and 4, we observe that the performance of the two models is very close, with a small but consistent edge to the Phillips Curve for virtually all variables in both in- and out-of-sample results. Finally, policy simulations with the two UCITY models yielded remarkably similar results. Looking at tables 6, 7 and 8, we observe that the policy multi pliers are nearly equal in both cases for inflation, output and unemployment. Rounded off to the first decimal point, they are almost identical, particularly after the first four quarters. C O N C L U S IO N S In this paper, we reviewed a current version of the St. Louis Model and presented two alternative ver sions, referred to as UCITY models, that retain the Andersen-Jordan nominal income reduced form but simplify the inflation sector and improve the estima tion of the unemployment rate. In the UCITYPC version, we replaced the St. Louis Model’s inflation sector with a more conventional Phillips Curve. In the UCITYRF version, we substituted a monetarist reduced form for inflation for the Phillips Curve. We demonstrated that the UCITY versions yield improved predictive performance of the major economic variables of interest to policymakers when compared with the St. Louis Model. In addition, the St. Louis Model yields more rapid deceleration of inflation and a smaller temporary rise in unemploy ment in response to a deceleration monetary growth than in the UCITY models. Finally, the UCITY models yield very similar predictive performances and virtually identical policy multipliers, suggesting that the Phi Hips Curve and monetarist reduced form are both reasonable, alternative equations for ex plaining inflation and correspond to structural vs. reduced-form approaches to modeling inflation. 25 FEDERAL RESERVE BANK OF ST. LOUIS DECEMBER 1981 FEDERAL RESERVE BANK OF ST. LOUIS REVIEW INDEX 1981 JANUARY JUNE/JULY John A. Tatom, “ Energy Prices and Short-Run Economic Perform ance” Neil A. Stevens, “ Indexation of Social Security Benefits — A Reform in Need of Reform ” W. \N. Brown and G. J. Santoni, “ Unreal Estimates of the Real Rate of Interest Scott E. Hein and James C. Lamb, Jr., “ Why the Median-Priced Home Costs So M uch” Neil A. Stevens, “ O utlook for Food and A griculture in 1981” Dallas S. Batten, “ Inflation: The Cost-Push M yth” FEBRUARY AUGUST/SEPTEMBER Michael E. Trebing, “ The New Bank-Thrift Competi tion: Will It Affect Bank Acquisition and Merger Analysis?” R. Alton G ilbert and Michael E. Trebing, “ The FOMC in 1980: A Year of Reserve Targeting” R. \N. Hafer, “ Selecting a Monetary Indicator: A Test of the New Monetary Aggregates” Clifton B. Luttrell, “ Grain Export Agreem ents— No Gains, No Losses” MARCH Scott E. Hein, “ Deficits and Inflation” G. J. Santoni and Courtenay C. Stone, “ Navigating Through The Interest Rate Morass: Some Basic Principles” R. IV. Hafer, “ The Impact of Energy Prices and Money Growth on Five Industrial Countries” OCTOBER Lawrence K. Roos, “ Lessons We Can Learn” Dallas S. Batten, “ Money Growth Stability and Infla tion: An International Com parison” R. W. Hafer, “ Much Ado About M2” Keith M. Carlson, “ Recent Revisions of GNP” NOVEMBER APRIL Anatol B. Balbach, “ How C ontrollable is Money G row th?” Clifton B. Luttrell, “ A Bushel of Wheat for a Barrel of Oil: Can We Offset OPEC’s Gains With a Grain Cartel?” Dallas S. Batten, “ Foreign Exchange Markets: The Dollar in 1980” MAY Michael David Bordo, “ The Classical Gold Stan dard: Some Lessons fo r Today” John A. Tatom, “ We Are All Supply-Siders Now !” Keith M. Carlson, “ Trends in Federal Revenues: 1955-86” G. J. Santoni and Courtenay C. Stone, “ What Really Happened to Interest Rates?: A Longer-Run Analysis” Keith M. Carlson, “ Trends in Federal Spending: 1955-86” Clifton B. Luttrell, “ The Voluntary Autom obile Import Agreement with Japan — More Protectionism” DECEMBER R. Alton Gilbert, “ Will the Removal of Regulation Q Raise Mortgage Interest Rates?” Laurence H. Meyer and Chris Varvares, “ A Com pari son of the St. Louis Model and Two Variations: Predictive Performance and Policy Im plications” 27