The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
FEDERAL RESERVE BANK OCTO B ER 1971 OF ST. LOUIS Slowing in Money Growth: The Key to Success in Curbing In flatio n .... c i r * L I T LI t lu r i I n LITTLE*ROCK Money Stock Control and Its Implications for Monetary Policy Slowing in Money Growth: The Key to Success in Curbing Inflation by KEITH M. CARLSON T HE ROLE of monetary actions in the administra tion’s new economic program has received little em phasis. Controlling inflation is an important objective of the program, yet a course for monetary actions has not been proposed. Historical experience in the United States and other countries indicates that the rate of inflation is related very closely to the trend rate of monetary expansion. Inflation has been a major economic problem since 1965, but experience since 1969 has been especially frustrating to the nation’s economic policymakers. In flation continued unabated in 1970, while unemploy ment rose sharply, and has since persisted in the face of relatively high unemployment. Conventional stabili zation actions taken in the last three years to reduce the growth of money and Federal expenditures have had their major impact on output rather than on the price level. This development has led some analysts to become skeptical about the usefulness of traditional monetary and fiscal actions in coping with the current economic situation. The new economic program was announced on August 15, outlining a set of measures designed to deal with the problem of simultaneous inflation and relatively high unemployment.1 A temporary wageprice freeze was implemented, and a number of fiscal actions were recommended with the objective of stim ulating output and employment. The second phase of the program takes the form of Government inter vention in private wage and price decisions for pur poses of making certain that the rate of advance of consumer prices is reduced to below a 3 per cent annual rate by late 1972. There are two major views prevailing with regard to the current problem of simultaneous occurrence of high unemployment and rapid inflation. One view stresses the role of “market power” as a factor influ!The program also included important provisions relating to the international system of exchange and the U. S. balance of payments. These provisions are not discussed here. Page 2 encing recent price level movements.- The other view is a part of the monetarist explanation of eco nomic processes, where the rate of inflation is de termined by the trend of monetary expansion. This latter view attributes the simultaneous occurrence of inflation and high and/or rising unemployment to uncertainty regarding the state of total demand and the costs of information and adjustment involved in making price and output decisions. These two views relating to the current economic situation are dis cussed along with their policy implications. Preceding the discussion of these two views is a brief survey of economic developments from 1969 to the present. Economic Developments From Late 1969 to the Present Monetary and fiscal actions turned restrictive in late 1968 and early 1969, and this policy stance was maintained until early 1971. The money stock rose at a 4 per cent average annual rate from early 1969 to late 1970, following an 8 per cent increase in the year ending first quarter 1969. So far this year monetary actions have been very stimulative, with the money stock rising at about a 9 per cent annual rate from January to September. Fiscal actions became less expansionary after pas sage of the Revenue and Expenditure Control Act of 1968, but more recently have turned stimulative. Fed eral expenditures rose at a 6 per cent average annual rate from mid-1968 to late 1970, compared with a 15 per cent average rate of advance in the previous three years. Since the fourth quarter of 1970, growth of Federal expenditures has accelerated to more than a 10 per cent rate. The income tax surcharge was im plemented in mid-1968, then allowed to expire in mid-1970. These stabilization actions, while having limited success in slowing inflation, have had their primary impact in reducing output growth and increasing un2The term “market power” is ambiguous in its meaning. As used here it is meant to encompass several views, including “cost-push,” “administered price,” and “sellers’ ” inflation. FED ERAL R ESER V E BANK OF ST. LOUIS OCTOBER 1971 Fiscal Measures Dem and and Production { + ) S u r p lu s ; (-)D e fic it R a tio S c a le T r i ll i o n s o f D o l l a r s R c t io S c a l e T r i ll i o n s o f D o l l a r s Quarterly Totals at A n n u a l Rates i.i 1.0 .9 1963 Sources: U.S. Department of Commerce, Council ol Economic Advisers, ond Federal Reserve Bonk of St. Louis latest doto plotted: 2nd quarter employment. The average level of prices advanced 5.7 per cent from fourth quarter 1969 to fourth quarter 1970, then slowed slightly to a 4.7 per cent annual rate. Unemployment increased from 3.6 per cent of the labor force in December 1969 to 6.2 per cent a year later, and has remained near this level, averaging 6 per cent in the first nine months of 1971. A year before unemployment started rising, unit labor costs, which had been rising since 1965, acceler ated further, reflecting a slowdown in productivity. Compensation per man-hour rose at a 7 per cent aver age rate from late 1968 to second quarter 1971, the same as from late 1965 to late 1968. In contrast, output per man-hour slowed to a 1.2 per cent rate of increase after late 1968, well below the 2 per cent average rate of advance from late 1965 to late 1968. As a result, 1964 1965 1966 1967 1968 i_[G NP in current dollars. [2 G N P in 1958 dollars. Percentages ore an nu al rates of c h o n ge for p eriod s indicated Latest d o to plotted: 2n d quarter 1969 1970 1971 Source: U.S. Department of Commerce unit labor costs have increased at a 5.8 per cent aver age rate since late 1968, compared to a 4.3 per cent rate from late 1965 to late 1968. Simultaneous Occurrence of Inflation and Unemployment: Two Views Inflation is a common economic problem, but infla tion in conjunction with high and/or rising unemploy ment is not so common. In the United States since Prices R a t io S c a le R a tio S c a le Source: U.S. Department of Labor Percentages are annual rotes of chan ge for p eriod s indicated. Latest d ata plotted:Consum er • August; W h o le sa le - September Page 3 OCTOBER 1971 FED ERA L R ESER V E BANK OF ST. LOUIS World War II, there have been few instances when rapid inflation and high unemployment have per sisted together for a significant period of time. Any such experiences lead some observers to develop the opinion that conventional monetary and fiscal actions are not adequate for the task of achieving what seem to be conflicting social goals. Explanations of the simultaneous occurrence of high unemployment and inflation can be divided into two categories. One attributes the source of the dilemma to an “excess of market power” among certain eco nomic units, and advocates Government intervention in market pricing to supplement conventional mone tary and fiscal policies. The other views the inflationunemployment dilemma as a result of rational eco nomic behavior in the face of uncertainty, and stresses monetary and fiscal actions as the means of restoring high employment and relative price stability.3 Market Power View According to the “market power” view, economic units, most commonly labor or businesses, are said to have excess market power when the prices of the goods or services they sell can be pushed up despite stable or falling demand. For example, because of substantial bargaining power, union wages may be pushed up faster than productivity advances. Firms respond to the increase in unit labor costs by increas ing their prices so as to protect their profit margins. According to this illustration, possession of market power by either labor or firms (or both) causes prices to increase in the face of stable or declining demand. Little evidence has been offered to support the market power view as it applies to the conjuncture of inflation and unemployment. For inflation to occur ( not just a once-and-for-all increase in the price level) when demand is stable, market power must be grow ing, or the use of existing market power must be in creasing. The existence of market power, however defined, probably means only that the price level is higher than it would be if no such market power prevailed. If one is to argue consistently within this view, a continuing rise in the price level can persevere despite stable demand only if market power or its use continues to grow. 3Those who demonstrate impatience with conventional mone tary and fiscal actions are not prepared to abandon such policies. Rather, these analysts feel that conventional stabili zation actions need to be supplemented at certain times to improve their effectiveness. On the other hand, those who oppose a supplement in the form of Government interven tion do so because they feel that the effectiveness of tradi tional actions would be hindered by such intervention. Page 4 Cost of Information View The cost of information view concerning the simul taneous occurrence of inflation and unemployment emphasizes the lags in adjustment of prices to an excess of total demand relative to total supply.4 A slowing of growth in total demand following a long period of excess demand and inflation, according to this view, can produce continuing inflation along with declining output and rising unemployment. Such a situation occurs, and is to be expected, because in formation about changes in total demand is not avail able without cost to decision-making units. Information about demand can be acquired only at a cost, and that cost varies with the speed at which it is obtained. Firms and households must incur a high cost per unit of information if they are to determine quickly that a change in total demand has taken place. There are also costs involved in adjusting prices and output, so prices can be expected to respond only slowly to a change in total demand. Under such cir cumstances of uncertainty, past price and demand changes frequently are used to form expectations about most likely future changes, and these expecta tions about demand will lag behind the actual facts of the demand situation. Thus, following a period of accelerating inflation, as from 1965 to 1969, there is a tendency for inflation to continue until changed econ omic facts are recognized and eventually lead to ex pectations of a slowing in inflation. Policy Implications of the Two Views Evidence that clearly supports one view as opposed to the other is difficult to develop. Nevertheless, the implications of the two views for monetary policy are of great significance and require further discussion. Monetary Policy and the Market Power View Since the problem of inflation in conjunction with unemployment is viewed by some observers as one of excessive market power, the natural policy implica tion is to curb the growth of such power or the exer cise of existing power. A policy of Government inter vention to guide prices and wages in line with pro ductivity would supposedly lower the rate of inflation sooner than otherwise.5 Furthermore, the implication is that with prices and wages directly controlled in 4This view does not deny the existence of market power, but it does deny the growth or growing use of that power as a major factor explaining inflation. 5Note that attention is focused on the inflation aspect of the problem. There is no disagreement between the two views on improving the operation of free markets. FED ER A L R ESER V E BANK OF ST. LOUIS OCTOBER 1971 this way, monetary and fiscal actions can properly be conducted towards stimulating total demand. An in crease in total demand with prices “directly” con trolled would be reflected in increased output and employment. Monetary Policy and the Cost of Information View The information-cost view of the inflation-unemployment dilemma is that the current situation is “temporary.” Following four years of accelerating in flation, the length of this period of adjustment to slower growth in total demand is probably quite sub stantial. However, according to this view, it is desir able for the economic system to complete the adjust ment without direct Government intervention in the price and wage policies of private economic units. Direct intervention in the form of wage-price-productivity formulae runs the risk of being viewed as a substitute for anti-inflationary monetary and fiscal actions; the return of monetary growth to a sustain able noninflationary rate consequently may be delayed. Wage-price standards, when accompanied by sus tained rates of monetary expansion in excess of the growth rate of potential output, will not prevent the creation of pressures which eventually result in the abandonment of any system of guidelines.“ Further more, controls hamper the operation of free markets, which are required to allocate resources according to consumer wants. Conclusions Given that the new economic program is concerned ultimately with achieving high employment with rela tively stable prices, historical experience indicates that monetary expansion should be maintained at a rate 6The experience with wage-price guideposts during the 1962-66 period is a case in point. See the 1970 Annual Report o f the Council o f Econom ic Advisers (Washington, D.C.: U. S. Government Printing Office, February 1970), pp. 23-25. about equal to the growth of the economy’s produc tive potential, or about 4 per cent per year under present circumstances. Whether a policy of moderate monetary growth supplemented with Government in tervention in private wage and price policy can re store high employment with price stability more quickly than without s u c h a supplement is problematical. Recent monetary developments, if continued, would provide the underlying basis for the control of inflation. After increasing at a 12.5 per cent annual rate from January to June, the money stock has slowed to about a 3 per cent annual rate of increase from June to September. If monetary actions are to contribute towards the battle against inflation, moderate mone tary expansion must be continued until a trend rate of money growth is established more in line with the growth of productive potential. Page 5 Money Stock Control and Its Implications for Monetary Policy by ALBERT E. BURGER, LIONEL KALISH III, and CHRISTOPHER T. BABB In the last two years there has been an increased concern within the F ederal Reserve Sys tem about the role of monetary aggregates in policy, and about the operating procedures impli cit in the policy directive o f the F ederal Open M arket Com mittee. A collection o f studies on these subjects, O p e n M a r k e t P o l i c i e s a n d O p e r a t i n g P r o c e d u r e s - S t a f f S t u d i e s , was pu b lished by the Board of Governors in 1971. Other economists have presented m ethods for analyz ing the effects of different growth rates o f m oney on policy objectives. An equally important subject is the controllability of different aggregates and the effect this controllability would have on the ability o f policym akers to achieve policy objectives. This article presents a procedure that could be used by the F ederal Reserve System to con trol the growth o f the money stock and a m ethod for evaluating the effect o f this control pro cedure on the ability o f policym akers to achieve their policy objectives. growing volume of research has demonstrated that changes in the money stock are a reliable sum mary measure of the effect of monetary policy actions on economic activity. One result of this research has been the suggestion that the monetary authorities could best achieve ultimate policy objectives, such as full employment and stable prices, by controlling the growth rate of the money stock. Such a suggestion re quires (1 ) an operational procedure for controlling money, and (2 ) a means of assessing the implica tions of such a procedure for the ability of policy makers to achieve their policy objectives. A possible procedure for monetary policy includes: ( 1 ) T h e Federal Open M arket Committee (F O M C ) decides upon the ultimate objectives of monetary policy, such as desired growth rates for real output and prices, and a desired level of employment. ( 2 ) These ultim ate objectives are related to a growth rate of money, and the FO M C issues a “direc °The authors wish to express their thanks to the many people who read earlier drafts of this article. A special obligation is due the following economists who, in working sessions or otherwise, offered specific comments: Professors Milton Friedman, Arnold Zellner, Robert Gordon, Richard Zecher, Stanley Fisher, Allan Meltzer, Michele Fratianni, William Yohe, David Fand, and Messrs. Paul Meek and Wolfgang Gebauer. As always, we benefited from comments and criti cism of the research staff at the Federal Reserve Bank of St. Louis. The procedures and conclusions are the responsi bility of the authors and do not necessarily reflect the views of any of the commentators on the article or the Federal Reserve System. Page 6 tive” to the Trading Desk to obtain this growth rate for money.1 ( 3 ) T he Trading Desk uses open market operations to achieve the growth rate of money which is consistent with the policymaker’s objectives. This article is concerned with the actual implemen tation of policy decisions. It is not concerned with how the policymakers decide upon their ultimate ob jectives, or with the specific way in which these ob jectives are related to a growth rate for money. The policy objectives are taken as given. Converting policy objectives into a desired growth rate of money re quires information on the linkage between changes in the growth rate of the money stock and the ultimate objectives. Such information can be derived from com peting models of income or spending determination. This article presents a procedure the Federal Re serve could use to control money and a method for evaluating the effect of this control on the policy maker’s ability to achieve GNP objectives. The money stock control procedure requires only that the Federal Reserve has information about the previous three 'The FOMC issues a policy directive to the New York Fed eral Reserve Bank. The Trading Desk at the New York Bank carries out day-to-day open market transactions (purchase and sale of Government securities) for the System. The text of each policy directive issued by the FOMC is made public about 90 days after each FOMC meeting and published in the Federal Reserve Bulletin. FED ER A L R ESER V E BANK OF ST. LOUIS month’s values of the money multiplier and the effect of reserve requirement changes on member bank re serves. Using a simulation technique, some empirical evidence is presented on the control the Federal Re serve could expect to exercise, using this procedure, over the growth of the money stock, and the effect of such control on the Federal Reserve’s ability to attain its policy objectives. The technical details of the money stock control procedure, the simulation proced ure, and the development of the statistic for assess ing the influence of money stock control on achieving policy objectives, are discussed in the Appendix to the article and in a working paper of technical ap pendices available upon request from this Bank.- Money Stock Control Procedure There are two major ways in which the Federal Reserve might operate to control the growth of money. One way is to estimate the money market conditions that would be consistent with the growth rate of money stated in the directive, and then operate to achieve these conditions in the money market. This approach might involve choosing bounds for the Fed eral funds rate and free reserves and then operating on a day-to-day basis to maintain money market con ditions within these bounds. A second method of money stock control, the one discussed in this article, involves estimating the changes in the source base ( or some other reserve aggregate) required to achieve the policy determined growth path for money. The Federal Reserve would then operate on a day-to-day basis to determine the growth of the source base.:! The money stock control procedure used in this article is developed from a multiplier-base framework, within which the money stock (M ) is expressed as: M = mB. In this expression B denotes the net source base and m represents the money multiplier. An increase in Federal Reserve holdings of securities, float, the gold stock, and Treasury currency outstanding will increase the net source base. An increase in Treasury deposits 2Albert E. Burger, Lionel Kalish III, and Christopher T. Babb, “Money Stock Control and Its Implications for Mone tary Policy: Technical Appendices,” Working Paper No. 14, Federal Reserve Bank of St. Louis, October 1971. 3These two methods of money stock control are not inde pendent of each other. Open market actions taken to deter mine money market conditions will influence the growth of the base, and actions taken to influence the base will affect short-term money market conditions. See, Albert E. Burger, “The Implementation Problem of Monetary Policy,” this Review (March 1971). OCTOBER 1971 at the Federal Reserve, Treasury cash holdings, and other deposits and other Federal Reserve accounts will decrease the net source base. A complete listing of the sources and uses of base money and the rela tionships between the net source base, source base, and monetary base are given in Table I. The net source base is taken as the control variable for the process , 4 From the sources side, the major component of the net source base ( about 75 per cent) is Federal Reserve holdings of Government securities. The Federal Reserve is assumed to be able to accur ately measure and determine the magnitude of the base within a monthly period. Evidence on the accu racy with which the Federal Reserve has been able to forecast and measure the net source base is presented in the working paper of technical appendices.5 The money multiplier (m ) summarizes all other factors involved in the money supply process. The money multiplier responds to portfolio decisions by the commercial banks, the Treasury, and the public. Also included in this formulation of the multiplier are the influences of reserve requirement changes, the discount rate, and Regulation Q.° In our money stock control procedure the Federal Reserve decides upon the desired growth rate of !The data requirements for controlling the net source base are as small or smaller than any of the other major aggregates commonly suggested as operating targets for the Federal Reserve. Richard Davis has shown that out of a wide range of possible aggregate targets the nonborrowed base and nonborrowed reserves would be the easiest targets for the Desk to hit. These two targets are entirely exogeneous with respect to open market operations. Contrary to other pro posed targets, success in hitting these two targets does not depend upon the Desk offsetting items whose movements are functionally related to open market operations. See Richard G. Davis, “Short-Run Targets For Open Market Operations,” Open M arket Policies and Operating Procedures-Staff Studies, Board of Governors of the Federal Reserve System, July 1971, pp. 37-70. 5Burger, Kalish, Babb, Working Paper No. 14. BThe money multiplier associated with the net source base is: 1+ k “ ( r - b ) (1 + t+ d ) + k where k and d, respectively, are the ratios of currency held by the public and U.S. Government deposits at commercial banks to the demand deposit component of the money stock. r, b, and t, respectively, are the ratios of bank reserves, member bank borrowings, and time deposits to commercial bank deposit liabilities (excluding interbank deposits). The reserve ratio (through the dependence of banks’ de sired excess reserves), the borrowing ratio and the time deposit ratio are all dependent upon credit market interest rates. This formulation of the money multiplier is taken from the Brunner-Meltzer nonlinear money supply hypothesis. Karl Brunner and Allan H. Meltzer, “Liquidity Traps for Money, Bank Credit and Interest Rates,” Journal o f Political E co nomy (January/February 1968), pp. 1-37. Page 7 O CTOBER 1971 FED ER AL R ESE R V E BANK OF ST. LOUIS Table I Sources and Uses o f the N et Source Base, the Source Base, and the M o n etary Base, January 1971 (millions of dollars) Uses Sources Federal Reserve holdings of Governm ent securities Federal Reserve float G old stock plus special draw ing rights Treasury currency outstanding O ther Federal Reserve Assets Less: Treasury cash holdings Treasury deposits at Fed eral Reserve Banks Foreign deposits at Fed eral Reserve Banks O ther deposits at F.R. plus F.R. liabilities and capital Equals: Net source base Plus: Federal Reserve discounts and advances Equals: Source base Plus: Reserve adjustment Equals: M onetary base $62,1 41 3,6 3 6 1 1,132 7 ,1 5 7 1,216 M em ber bank deposits at Federal Reserve Banks less discounts and advances Currency held by banks Currency held by the public 445 1,028 Forecasting the Money Multiplier 155 2,894 $ 8 0 ,7 6 0 370 $ 8 1 ,1 3 0 3 ,8 2 6 $ 8 4 ,9 5 6 Equals: Net source base Plus: Federal Reserve dis counts and advances Equals: Source base Plus: Reserve adjustment Equals: M onetary base ♦ D ata n o t seasonally ad ju sted . money, converts this growth rate into desired money stock levels for the control periods, and forecasts the money multiplier (m ) for the control periods. Then during the control periods, the Federal Reserve uses open market operations to attain the net source base ( B ) such that the product ( m B) equals the desired money stock levels. Implementing monetary policy under such a money stock control procedure requires three considerations: (1 ) the length of the control period; (2 ) a procedure for forecasting the money multiplier; and (3 ) the response to previous errors in money stock control. Control Period The maximum acceptable time period for forecasts of the multiplier depends upon the relationship be tween changes in money and changes in economic activity. Empirical evidence indicates that quarter-toquarter changes in the growth rate of money influ ence economic activity. Therefore, the maximum time period over which the Federal Reserve would aim to control the money stock would be a quarterly period. Such an assumption, however, leaves open the possi bility of sharp fluctuations in the growth of money over the quarter. Therefore, it is further assumed that as an operating strategy, it is preferable to minimize Page 8 $ 2 4 ,5 6 8 7,092 4 9 ,1 0 0 the expected squared deviation of the monthly value of money from its de sired growth path. The net source base is assumed to be controllable on a dailyaverage monthly basis; therefore, with in our control procedure monthly aver age multipliers are forecast. Having predicted the value for the month’s money multiplier, and given the desired level for the money stock in that month, the average monthly value for the net source base necessary to achieve the desired growth of money is determined. $ 8 0 ,7 6 0 Next period’s multiplier might be forecast by any one of the following methods: 370 (1 ) D efin ition al m e th o d multiplier-base framework is treated as an accounting iden 3,826 tity. Some of the ratios of the $ 8 4 ,9 5 6 multiplier are forecast using in formation about the various components (for example, Treas ury deposits) acquired by the Desk in its daily operations. O ther elements of the ratios are treated as being equal to their previous values with some adjustment for trend or seasonal variation.7 $ 8 1 ,1 3 0 ( 2 ) R eg ression m e th o d — T he money multiplier is expressed as a function of variables that are known or are under the policy control of the Federal Reserve at the time each forecast is made. This relationship is estimated each period by multiple regression analysis. ( 3 ) B eh a v io ra l m e th o d — E a ch of the ratios of the multiplier is expressed as being dependent upon other variables such as interest rates, policy instruments, and other factors influencing the deposit behavior of the banks and the public. This procedure requires predicting these other variables. In this article, the second method is used. Each month’s multiplier is forecast using the three-month moving average of past values of the multiplier, re serve adjustment magnitude in the forecast month, 7See Leonall C. Andersen, “A Study of Factors Affecting the Money Stock: Phase I,” Federal Reserve Bulletin (October 1965), p. 1379; and William G. DeWald, “Monetary Control and the Distribution of Money,” unpublished Ph.D. thesis, University of Minnesota, 1963. — T FED ER A L R ESE R V E BANK OF ST. LOUIS dummy variables to account for seasonal factors, and an adjustment for autocorrelation. The values of these independent variables are known to the Federal Reserve.8 OCTOBER 1971 M oney Stock Control Process G R O W T H RATE OF M O N E Y CO N SISTEN T W IT H THE P O LIC Y O B JEC T IV E S OF THE FEDERAL RESERVE Response to Previous Errors in Money Stock Control If there are errors in the forecasts of the money multiplier, the desired growth of money and the con trolled growth of money will not be the same in every period. Under these conditions, further information is required to determine the optimal setting for the net source base. Suppose in period tj the money man agers over-predict the money multiplier. Consequently, the achieved growth of money is less than the desired growth rate. What is the optimal setting for the net source base in period t2? Should the money managers ignore the shortfall of money in t 3? Should they try and make up the shortfall of money in t, by setting the net source base in t2 so that the growth of money is above the desired growth path? If they try and make up the shortfall, should they operate to make up all of the gap in t2, or only part of the gap in t2 and the remainder in succeeding periods? There are many possible error-response mecha nisms. Our procedure assumes that the money man agers assign proportionally more weight to large errors in money stock control than small errors. Therefore, the error-response mechanism is designed to minimize the expected value of the squared deviations of con trolled money from its policy chosen growth path.9 At the end of each control period, the money man agers compute their error in money stock control. During the next control period the net source base is set approximately to make up last period’s error.10 The money stock control procedure is illustrated in the following exhibit. 8The Federal Reserve sets member bank reserve require ments. Since, under the current lagged reserve requirement procedure the effect on member bank required reserves of a change in reserve requirements effective this week depends upon member bank deposits subject to reserve requirements two weeks earlier, the Federal Reserve can accurately de termine the effect of a change in reserve requirements on the reserve adjustment magnitude. 9This error-response mechanism assumes a quadratic loss function for the money managers. The control periods’ base values are determined by minimizing the expected value of the loss function with respect to B. For a discussion of this procedure see the Appendix at the end of this article. 10If only the growth rate of money mattered, then the money managers would not attempt to make up last period’s error in the level of money. Each period the money man gers would try to move along the desired growth path from where they were last period. G R O W T H RATE FOR M O N E Y CONVERTED TO M ON TH LY M O N E Y S T O C K LEVELS M O N EY STOCK -------- ^ 7 -------- E R R O R -R E S P O N S E Com pute the difference betw een the desired level of m o ney and actual m oney achieved by the control process in the p re v io u s month M o n th ly ch an ge in b a se e q u a ls am o un t necessary to achieve current m o n th 's d e sire d m o n e y sto ck level, takin g into account the error in last m onth’s money stock level C O N VERTED TO M O N T H LY D A IL Y -A V E R A G E BASE LEVEL BY FOR EC A ST IN G THE M ON TH LY M O N EY M U LT IP LIER <3> B A SE TARGET EQUALS A M O U N T OF B A SE N ECESSA RY TO ACH IEV E DESIRED M O N EY ST O C K LEVEL O NET SO U R C E BASE CONVERTED TO O PEN M A R K E T O PERA TIO N S B Y CO M PU T IN G THE P R E V IO U S D A Y ’S BASE A N D FORECASTING THE U N C ON TRO LLED CO M PONENTS OF THE BASE FOR EACH D A Y O O PEN M A R K E T O P ER A T IO N S Simulation of Money Stock Control and GNP How would the money stock control achieved by this procedure affect the ability of the Federal Re serve to achieve policy objectives? To gain some in formation on this question, the money stock control procedure was simulated over two sample periods, Page 9 OCTOBER 1971 FED ER AL R ESE R V E BANK OF ST. LOUIS and the effects of these simulation re sults on GNP were analyzed. First, the method and results of simulating the money stock control procedure are pre sented. Then, the method of relating changes in money to GNP is discussed, and the results of simulating GNP when money is controlled without er ror are compared to the case where money is controlled by our procedure. Table II Exam ple of M o n e y Stock Control Desired M o ne y Actual Multiplier Con trolled M o ne y $ 2 0 0 .0 $8 0 .00 2.50 January 201 2.50 $.40 80.40 2.50 201.0 February 202 2.50 .40 80.80 2.50 202.0 March 203 2.51 .08 80.88 2.50 202.2 April 204 2.50 .72 9 1.6 0 2.50 204 .0 December The following method was used to simulate money stock control: ( 1 ) It was assumed that the policymakers choose a constant 4 per cent seasonally adjusted annual growth rate for money over the control period.11 (2 ) The Federal Reserve adjusts the net base in the current month to minimize pected value of the squared deviation achieved monthly stock level from the 4 per cent growth line.12 $200 the one that actually prevailed in that month, the achieved level of money is different from the desired. Simulation of the Money Stock Control Procedure source the ex of the original ( 3 ) Two control periods were chosen, 1962 through 1965 and 1966 through 1969. The base periods were chosen as fourth quarter 1961 and fourth quarter 1965. ( 4 ) E ach month, the money stock achieved by the control process (controlled money) is computed by taking the level of the net source base de termined by our operating strategy and multi plying it by the value of the multiplier that actually prevailed in that m onth.13 To the ex tent that the forecast multiplier is different from n The desired growth rate of money was converted into desired monthly levels in the following manner: (1 ) the averages of money in IV/61 and IV/65 were taken as the base period; (2 ) these base values were placed on Decem ber of 1961 and December 1965; (3 ) to compute the con version factor for a 4 per cent growth rate we divided .04 by 12 to yield .00333; and (4 ) each month’s desired money stock level was equal to (base month) + (base month) X (number of months out from base month) X (.00333). For example, December 1966 desired level equals (167.0999) + (167.0999) (1 2 ) (.00333) = $173.78 bil lion, where 167.0999 equals the average of the last three months of 1965. This procedure yields a simple 4 per cent growth rate of money that appears as a straight line on an arithmetic scale. When computing quarter-to-quarter growth rates of money, however, the desired rate will be below 4 per cent near the end of the period. The results of our procedure would not have been altered if we had used a compounded annual rate for money. 12This implies the Desk aims slightly below the desired growth path. See the Appendix at the end of this article. http://fraser.stlouisfed.org/ Page 10 Federal Reserve Bank of St. Louis Control Forecast C hange in Actual Multiplier Base Base level The example in Table II, which may be used to illustrate this procedure, should be taken only as an illustration. For the technical aspects of the proce dure, especially die error-response mechanism, con sult the Appendix at the end of this article. The first column in Table II gives the monthly money stock levels consistent with the growth rate of money that the policymakers are advised will give them their de sired policy objectives. The second column gives the forecast of the multiplier and the fifth column gives the money multiplier that actually prevailed in each month. It is assumed that the control procedure be gins in January. For the first two months the Federal Reserve forecasts the multiplier with complete ac curacy, the net source base is changed by $.4 billion, and the money stock achieved by the control proce dure equals the desired. In March, however, there is an error in the forecast of the multiplier. The multiplier is forecast to be 2.51, when it actually (the historical value) is 2.50. Conse quently, the net source base is increased by only $.08 billion. Based on a forecast of 2.51 for the multiplier, the Federal Reserve expects that it would only have to supply $.08 billion of base, compared to $.40 bil lion in the previous two months. The result is an error in money stock control, controlled money is less than desired ($202.2 billion compared with a desired level of $203 billion). In April, the Federal Reserve again forecasts the multiplier correctly. In this month the net source base is increased enough to make up last month’s money stock error, and to hit the target of $204 billion.14 13This procedure assumes the independence of changes in the net source base and the multiplier. If m and B are not independent, then the actual multiplier might not be the one that prevailed, given a different change in B. For a discussion of this condition see Lionel Kalish, “A Study of Money Stock Control,” Journal o f Finance (September 1970), pp. 761-776. i^This example uses an absolute loss function. To minimize the expected value of the squared deviations of money, the FED ER A L RESERV E BANK OF ST. LOUIS There are two prevalent views among economists concerning the constancy of the desired rate of change of the money stock.1-' One view is that the desired rate of change should never be altered ( seasonally or cyclically). An alternative view holds that the mone tary authorities have enough knowledge to alter the monetary growth rate seasonally and cyclically so that economic goals can be achieved better than if the rate were held constant. This particular issue does not affect our control procedure. The choice of a constant 4 per cent growth rate for money does not necessarily imply that a 4 per cent rate was a desirable monetary growth path for this period. Different desired rates of change mean only that the monetary authorities aim for different money stock levels, and combined with the same forecasted multiplier, the only difference in the operating strategy would be a different change in the net source base. Com parison of the sam ple periods —In order to gain information about the stability and robustness of the money stock control procedure, it was simulated during two historical periods which were markedly different with respect to the stability of the money multiplier. A change in any of the ratios appearing in the money multiplier ( see footnote 6) can alter the value of the multiplier. These ratios are influenced by a number of factors such as market interest rates, the relationship between Regulation Q ceiling rates and market rates, Treasury deposit decisions and chang ing patterns of tax payment dates, and introduction of changes in reserve requirements such as lagged requirements in September of 1968. Therefore, in periods where there are sharp or er ratic changes in the factors influencing the multiplier, one might expect the errors in predicting the multi plier to be larger than in periods where these factors remain constant or follow a steady trend.10 The fol actual change in base would be slightly less than $.72 billion. 15See Milton Friedman, A Program For Monetary Stability (New York: Fordham University Press, 1959), and Franco Modigliani, “Some Empirical Tests of Monetary Manage ment and of Rules Versus Discretion,” Journal o f Political Economy, June 1964, pp. 211-245. 1,:The three interest rate series, commercial paper rates, mar ket yields on Treasury bills, and long-term corporate bond rates, all exhibited much greater variation in the 1966-69 period. Examination of the t, k, and r-ratios also reveals a pattern of increased variability and sharp erratic move ments in these ratios in the latter period. Of special interest is the behavior of the t-ratio (time deposits/demand de posits of money) in the two periods. In the 1962-65 period the t-ratio follows a steady upward trend with only a small amount of variation about the trend. In contrast, the t-ratio during the 1966-69 period exhibits wide and erratic fluctuations about its trend line. In the 1966-69 period the contribution of the t-ratio to the month-to-month percentage change in the historical money stock had a mean of —.57 per OCTOBER 1971 lowing chart shows the variation of the money multi plier about its trend during both sample periods. Com paring the behavior of the historical money multi plier, it can be seen that it exhibited much greater variability in the 1966-69 sample period than in the 1962-65 period. Em pirical results —The results of simulating money stock control over the two sample periods are illus trated in the following chart.17 Table III presents several alternative ways of evaluating the results of our control procedure. This table presents controlled and desired levels, controlled and desired growth rates, and includes the mean, variance, mean square, and median of the errors. Although the underlying conditions for money stock control are quite different in the two sample periods, the mean value of differences between controlled and desired growth rates is approximately the same in both periods. The mean value of deviations of con trolled and desired levels is somewhat larger in the 1966-69 period. However, relative to the levels in volved, the average percentage errors these devia tions represent is approximately the same for the 1966-69 period as for the 1962-65 period. The major difference between the results of the control procedure in the two periods is the occurrence of somewhat more frequent large deviations in the 1966-69 period. One indication of this difference is that the mean squared error for differences in the levels for the latter sample period is $.62 billion, com pared to $.36 billion in the earlier period. Also, the average for the five largest percentage errors in the levels is 0.68 per cent in the latter period, compared to 0.58 per cent in the earlier period. Projections of GNP Policymakers are primarily concerned with attain ing ultimate policy objectives, not just with controlling the growth of money. Controlling money is a means to an end, not the end in itself. In this section, the growth of GNP implied by a constant 4 per cent growth rate of money is chosen as the policy objec tive. This policy objective path for GNP (desired GNP) is then compared to the growth of GNP atcent and a variance of 4.48, compared to a mean of —2.29 per cent and a variance of .76 in the 1962-65 period. There are pronounced changes in the pattern of the t-ratio in the last half of 1966, in 1968, and during 1969. These changes reflect primarily the constraint of Regulation Q, which was an additional factor influencing the money supply process in the latter period. 17Charts plotting monthly values of controlled money are given in Working Paper No. 14. Page 11 FED ERAL R ESE R V E BANK OF ST. LOUIS tained with controlled money. The following proce dure was used: (1) A model linking changes in the money stock to changes in nominal GNP was chosen. The model used was the Andersen-Jordan (A-J) spending equation which relates changes in GNP to current and lagged changes in the money stock and high employment government expenditures.18 18Leonall C. Andersen and Jerry L. Jordan, “Monetary and Fiscal Actions: A Test or Their Relative Importance in Economic Stabilization,” this Review (November 1968), pp. 11-24. Page 12 O CTOBER 1971 (2) Actual high-employment government expendi tures were used in both simulations. This pro cedure assumes that forecast high-employment government expenditures are always equal to the actual. (3) The A-J equation was used to project quarterly GNP with a constant 4 per cent growth of money. This projected GNP path is the policy objective. Then, the A-J equation was used to project GNP for the same period, with the growth pattern of money as generated by our control procedure when aiming at a constant 4 per cent money stock growth. This is the GNP FED ERA L R ESER V E BANK OF ST. LOUIS Table III OCTOBER 1971 Controlled C o m p are d to Desired Q uarterly A v e ra g e s of the M o n e y Stock1 (Billions of Dollars) 1 9 6 2 -1 9 6 5 Controlled Level Desired Level Difference Between Controlled and Desired Levels 1962 1 II III IV $146.1 1 147.12 148.22 150.94 $ 1 4 6 .5 7 148.03 149.48 150.94 $ - .46 — .91 -1 .2 6 0 196 3 1 II III IV 151.65 153.99 155.63 155.90 152.40 153.85 155.31 156.76 1964 1 II III IV 157.50 159.02 161.54 162.06 158.22 159.68 161.13 162.59 — — 1965 1 II III IV 163.68 165 .37 167.24 168.61 164.04 165.50 166.95 168.41 - Quarter M e an value of deviations: Absolute W ith Sign Variance of deviations: Absolute W ith Sign M ean squared deviations: — — — Per Cent Error Between Controlled and Desired Levels — — .3 % .6 — .8 0 .75 .14 .32 .86 — — .72 .66 .41 .53 — — .36 .13 .29 .20 — — — .50 -.3 3 Difference in Annual Q uarter-to-Quarter Growth Rates (Controlled M inus Desired) — 1 .3 % — 1.2 -0 .9 3.4 .5 .1 .2 .5 — 2.0 2.4 0.5 -3 .0 .5 .4 .3 .3 0.4 0.2 2.7 — 2.3 .2 .1 .2 .1 0.4 0.5 1.0 -0 .2 M ean of Per Cent Error (Absolute) .32 .1 1 .25 M ean of Difference in Growth Rates .04 Per Cent Error Between Controlled and Desired Levels Difference in A nnual Quarter-to-Quarter Growth Rates (Controlled M inus Desired) .36 1 9 6 6 -1 9 6 9 Controlled Level Desired Level 1966 1 II III IV $168.71 170.28 173 .06 174.22 $168.21 169.89 171.56 173.23 1967 1 II III IV 1968 1 II III IV 174.82 177.13 177.73 178 .96 Difference Between Controlled and Desired Levels 174.90 176 .57 178 .24 179.91 181.58 183.25 184.92 18 6 .6 0 Quarter 1969 1 II III IV 181 .56 185.14 185.13 187.26 188.54 189 .24 192.08 193 .37 $ 188 .27 189.94 191.61 193.28 M e an value of deviations: Absolute W ith Sign — .08 .56 — .51 — .95 — .02 1.89 .21 .66 — .61 .33 Variance of deviations: .50 .39 1.50 .99 .25 .51 Absolute W ith Sign .27 .70 .47 .09 .3 % .2 .9 .6 1 .2 % -0 .3 2.6 — 1.2 .0 .3 — .3 — .5 .0 1.0 .1 .4 — 2.5 1.5 -2 .4 — 0.9 2.1 4.2 -3 .6 1.0 .1 .4 .2 .0 -0 .9 — 2.0 2.5 — 0.8 — M ean of Per Cent Error (Absolute) .33 M ean of Difference in Growth Rates .03 M ean squared deviations: .62 Q u a r te r ly av erag e s o f controlled m oney w ere com puted by a v e ra g in g m o n th ly values achieved by the m oney stock control procedure. M onthly, controlled m oney stock d a ta is given in W o rk in g P a p e r No. 14. that would have actually resulted from policy actions.11 1 Several important points about this procedure should be emphasized. It is assumed that the GNP 19The desired quarterly growth rate of money was computed from the desired quarterly average money stock levels re ported in Table III. that would have resulted from a constant 4 per cent growth rate of money and from controlled money would have been the GNP projected by the equation relating changes in money to changes in GNP. There fore, the only source of error between the policy ob jective GNP and the GNP resulting from money stock control is the error in money stock control. Page 13 FED ERA L R ESER V E BANK OF ST. LOUIS Controlled and Desired M oney Stock a O CTOBER 1971 employment and prices and then convert these into a desired growth rate of nominal GNP. An analysis of what a given growth of GNP implies for prices and employment could be carried out by using a larger model. Em pirical results —The results of the GNP simula tions are presented in Table IV and the following chart. The monetary policymakers are assumed to have chosen objectives for GNP, and then, based on the information they have about the relationship be tween money and GNP, they have decided that a 4 per cent monetary growth rate will best achieve these GNP objectives. The second column of Table IV contains the growth path of GNP the policymakers desire to achieve ( quarterly averages of nominal GNP projected by the A-J equation when a constant 4 per cent growth rate of money is assumed). The first column of this table shows the quarterly averages of nominal GNP projected by the A-J equation when the money stock resulting from our operating proce dure for those years is read into the A-J equation. □ .T h e d e s ir e d g r o w t h p a t h fo r m o n e y is a s s u m e d to b e a c o n st a n t 4 p e r cent s e a s o n a l ly a d ju s t e d g r o w t h rate. Q u a r t e r l y d a t a fo r c o n t ro lle d m o n e y a r e c o m p u t e d b y a v e r a g i n g m o n t h ly v a lu e s of m o n e y re su lt in g fro m the m o n e y st o c k co n tro l p r o c e d u r e . Second, instead of the A-J equation, any other econ ometric model that relates changes in money to changes in nominal GNP could have been used. Other economists might work out the implications of this and other money control procedures for the ability of pol icymakers to hit a desired value of GNP, using al ternative forecasting and structural models of the econ omy. Such results would provide valuable supple mental evidence on the adequacy of proposed money stock control procedures. A third point is that only the influence of money stock control on nominal GNP was considered. The ultimate objectives of monetary policy are variables such as employment and prices. However, this article is not concerned with the influence of different GNP growth rates on employment and prices. It is as sumed that the policymakers pick desired values for Page 14 At an operational level, the Trading Desk is di rected to follow an open market policy to achieve the 4 per cent growth rate of money. To carry out its “directive,” the Trading Desk forecasts the money multiplier by our procedure, and then supplies the amount of net source base each month that is required to achieve the level of the money stock consistent with the 4 per cent growth of money. Since there are deviations between the quarter-to-quarter growth rate of money achieved by the control procedure and the desired 4 per cent rate, there are deviations of achieved GNP from the policy objective. Under the simulation exercise, the success of the policymakers in achieving their desired GNP objec tives on average is approximately the same in both sample periods. The largest percentage error in the levels is seven-tenths of one per cent, and in both periods 10 of the 16 quarterly misses are three-tenths of one per cent or less. The mean difference between money stock control and policy objective (desired) quarter-to-quarter growth rates of GNP is .01 per cent in the 1962-65 period and .02 per cent in the 1966-69 period. The simulations indicate that the Federal Reserve would have been about equally successful in achiev ing its GNP objectives in both periods. This result follows from two conditions. First, although there are more frequent large deviations in the achieved money stock in the 1966-69 period, they are not maintained for a long period. On average the degree of control is about the same in both sample periods; deviations FEDERAL RESERV E BANK OF ST. LOUIS OCTOBER 1971 GN P Projections: Policy Objective GNP Compared With M oney Stock Control G N P 1 C o m p a riso n of N o m in al G N P Levels B il l i o n s o f D o l l a r s B il l i o n s o f D o l l a r s B illions o f D o l l a r s B illions of D o ll a rs 680 660 640 620 600 580 560 C o m p a riso n of Q uarter to Q uarter A n n u a l Rates of C h a n ge of GNP Per C en t Per C e n t Per C e n t Per C e n t 12 12 10 10 L]_The A n d e r s e n - J o r d a n (A-J) s p e n d i n g e q u a t io n w a s u s e d to p ro je c t q u a r t e r ly G N P w ith a c o n st a n t 4 p e r c e n t g r o w t h o f m o n e y . T h is p ro je c t e d G N P p a th is the p o lic y o b je ctiv e . Th e n , the A -J e q u a t io n w a s u s e d to pro je ct G N P , w ith the g ro w t h p attern o f m o n e y g e n e r a t e d b y the m o n e y sto c k c o n tro l p r o c e d u r e w h e n a im in g a t a c o n st a n t 4 p e r cent m o n e y st o c k g ro w th . A c t u a l h i g h - e m p lo y m e n t g o v e r n m e n t e x p e n d it u r e s w e re u s e d in b o th s im u la tio n s , a s s u m i n g th a t p roje cted h ig h - e m p lo y m e n t e x p e n d it u r e s a r e a l w a y s e q u a l to the ac tu al. F o r a d is c u s s io n o f the A -J e q u a tio n , se e L e o n a ll C . A n d e r s e n a n d J e rry L. J o r d a n , " M o n e t a r y a n d F isc a l A c tio n s : A Test of T h e ir R e la t iv e Im p o rt a n c e in E c o n o m ic S t a b iliz a t io n , " F e d e r a l R e s e r v e B a n k o f St. L o u is R e v ie w { N o v e m b e r 1968), p p. 11-24. above the desired growth path are followed by devia tions below the growth path. Second, in the GNP equation, the influence of changes in the growth rate of money are distributed over time. The whole impact of a change in money on GNP does not occur in the same quarter. The influence of money on income in cludes the growth of money over the preceding four quarters. Assessing the Effect of Money Stock Control on Policy Objectives In the previous section, the growth path of GNP projected assuming no errors in money stock control was compared to GNP projected with money stock control using our control procedure. The comparisons were made for two sample periods. However, even Page 15 FED ER AL R ESE R V E BANK OF ST. LOUIS O CTOBER 1971 Table IV N om inal G N P Levels and C o m po un ded Annual Rates o f C h a n g e G enerated by the Andersen-Jordan Equation (Billions of Dollars) 1 96 2 -1 96 5 Difference Between Controlled and Desired Levels Per Cent Error Between Controlled and Desired Levels Difference in A nnual Quarter-to-Quarter Growth Rates (Controlled M inu s Desired)^ Quarter Controlled Level Desired Level1 1962 1 II III IV $ 5 4 4 .5 4 5 5 3 .5 4 5 6 0 .8 9 5 6 8 .7 4 $ 5 4 5 .2 0 5 5 5 .6 0 56 4 .7 8 57 2 .3 3 $ — .66 1963 1 II III IV 5 7 5 .8 6 5 8 5 .9 2 5 9 6 .7 4 60 4 .8 5 5 7 9 .3 7 588.21 5 9 7 .4 2 6 0 5 .95 — 3.51 — 2.29 — .68 — 1.10 — .2 — .3 1964 1 II III IV 615.61 6 2 8 .44 6 3 8 .3 7 64 6 .2 3 6 1 7 .7 9 63 1 .38 6 4 0 .35 647 .98 -2 .1 8 -2 .9 4 — 1.98 — 1.75 — — — .4 .5 .3 .3 — .7 .5 .7 .2 1965 1 II III IV 6 5 3 .9 7 6 6 5 .0 6 6 8 0 .3 0 6 9 5 .9 9 6 5 5 .5 7 66 6 .44 6 8 0 .85 6 9 5 .8 7 — 1.60 — 1.38 — .55 — — .2 .2 M ean value of deviations: Absolute W ith Sign Variance of deviations: Absolute W ith Sign M e a n squared deviations: -2 .0 6 — 3.89 — 3.59 — — — — . 1% .4 .7 .6 — — .6 .4 .1 — .1 1.2 .5 % — 1.1 — 1.4 .3 .9 .1 .2 .5 .4 — .1 .12 1.89 — 1.88 - .0 M e an of Per Cent Error (Absolute) .32 1.24 1.29 M e an of Difference in Growth Rates .01 Per Cent Error Between Controlled and Desired Levels Difference in Annual Quarter-to-Quarter Growth Rates (Controlled M inus D e sire d )- 4.82 1 9 6 6 -1 96 9 Difference Between Controlled and Desired Levels Controlled Level Desired Level1 1 96 6 1 II III IV $ 7 2 8 .1 3 7 4 4 .0 3 7 6 0 .7 9 7 7 7 .4 3 $ 7 2 7 .4 7 742 .58 7 5 7 .2 4 7 7 2 .19 1 96 7 1 II 79 1 .13 80 2 .19 80 8 .86 8 1 6 .4 7 7 8 6 .43 7 9 8 .4 7 8 0 7 .44 8 1 7 .6 7 4.70 3.72 1.42 — 1.20 II III IV 827 .63 8 4 6 .6 7 8 6 2 .9 7 8 7 5 .86 829 .78 8 4 6 .2 0 860.21 8 7 1 .6 6 — 2.15 .47 2.76 4.20 1 96 9 1 II III IV 883 .42 888 .92 8 9 9 .46 9 1 3 .0 4 879.71 8 8 7 .7 2 899 .03 9 1 2 .83 3.71 1.20 .43 .21 Quarter 11 1 IV 1968 1 M ean value of deviations: Absolute W ith Sign 2.32 1.90 Variance of deviations: 2.60 4 .3 7 M e an squared deviations: Absolute W ith Sign 7.97 1A ssu m in g a sim ple 4 p e r c e n t g ro w th ra te fo r m oney. 2C om pounded a n n u a l ra te s. Page 16 $ . 1% .2 .5 .7 .66 1.45 3.55 5.24 .6 .5 .2 .4 % .5 1.2 .9 — .1 — .3 — .6 — 1.2 — 1.3 — — .3 .1 .3 .5 .4 .1 .0 .0 M e an of Per Cent Error (Absolute) .29 .5 1.4 1.1 .7 — .3 — 1.2 — .4 — .1 M ean of Difference in Growth Rates .02 FED ER AL R ESER V E BANK OF ST. LOUIS O CTOBER 1971 though the control procedure worked reasonably well during the sample periods, it will not necessarily do as well in some time interval outside the sample periods.-" In an actual policy application, the procedure would be used outside the sample period. Therefore, policymakers must have some means of assessing what a suggested control procedure implies for their ability to achieve policy objectives in a forecasting situation. This criterion can be a comparison of the ability to attain policy objectives when there are no errors in the control procedure, with the case where there are errors in the control procedure. In this ar ticle GNP was chosen as the policy objective, and policy was implemented using a money stock control procedure. Therefore, the basis for judging the control procedure is the amount by which the errors in money stock control a d d to errors in forecasting the GNP that would result from a desired growth rate of money. In this section, representative GNP prediction con fidence intervals are presented. In arriving at these confidence intervals, allowance is made for the relia bility of the sample estimates of the model’s para meters and the multiplier forecasts. A modified stand ard error of forecast statistic is used to specify con fidence intervals for GNP projections when money is controlled. These confidence intervals are then com pared with confidence intervals for GNP projections when there are no errors in money stock control.21 GNP being within ±$8.64 billion of the projected level. On average, over the four quarters in 1970, money stock control would have reduced the proba bility of the actual value of GNP falling within the given confidence interval from 95 per cent to 93.3 per cent. These results indicate that, for 95 per cent confidence intervals, the errors in money stock con trol implied by our control procedure would have had only a very small effect on the policymaker’s ability to forecast GNP. Conclusions The implementation procedure for monetary policy developed in this article provides the basis for a welldefined operational procedure for controlling money. The money stock control procedure does not require the use of any information which the Federal Reserve does not already have available. In fact, it greatly simplifies the operating instructions which would be issued to the Trading Desk. The FOMC would issue a directive to the Trading Desk stated in terms of a growth rate for money. The Desk would convert this growth rate of money into a monthly daily-average net source base figure by using the procedure de veloped in this article to forecast the monthly money multiplier. During each month, the Desk would use open market operations to set the net source base at the level consistent with the growth rate of money stated in the directive. The Desk would not have to interpret the “consensus of the members of the FOMC.” Each month the Desk would have a precise monthly daily-average net source base figure to attain. Table V presents 95 per cent confidence intervals for GNP projections, assuming no errors in money stock control for the four quarters of 1970. The final column in Table V presents the probability of the Using a simulation technique, this article presented actual value of GNP falling within these same con evidence on the effect this money stock control profidence intervals, given that our control procedure is used to control money. For example, there would have been a 95 per cent Table V probability of actual GNP being within Quarter to Q uarter G N P Projections for 19 7 0 1 ±$8.64 billion of the projected level in Probability of Actual Value 11/1970, assuming there was no possi Falling W ithin Confidence Interval bility of errors in money stock control. Confidence Interval Error in about the N o Error in If money had been controlled by our M o n e y Stock Control2 Projected Level M o n e y Stock Control procedure, there would have been a (Billions of Dollars) 93-94 per cent probability of actual 1/1970 20This result can occur because the point estimates of the parameters of the model differ from their unobservable population values. 21The deviation of the SE F statistic and the technical aspects of specifying these confi dence intervals are discussed in the Ap pendix at the end of this article, and in the technical appendices available in Work ing Paper No. 14. ± $ 8 ,1 2 0 95% (8 9 .5 -9 4 .3 )% 11/1970 ± $8,641 95 (9 3 .3 -9 4 .4 ) 111/1970 ± $8,2 2 8 95 (9 3 .3 -9 4 .3 ) IV / 1 9 7 0 ± $ 8 ,3 4 7 95 (9 3 .3 -9 4 .4 ) 1A n d e rse n -Jo rd a n equation (a ) sam p le p erio d 1/1953 to IV /1969. (b) A lm on specification: both la g d istrib u tio n s em ploy 4th degree polynom ial w ith lags t - f l an d t —5 co n strain ed to equal zero. 2T he first n u m b e r in each p a ir is associated w ith a q u a rte rly m u ltip lier e quation and th e second w ith a m on th ly m u ltip lie r e quation. F o r a fu r th e r discussion see p ag e 21. Page 17 FED ERAL R ESER V E BANK OF ST. LOUIS cedure would have had on the ability of policymakers to achieve GNP objectives. In both of the four-year sample periods the largest percentage error in GNP levels was less than one per cent, and in each period 10 of the 16 quarterly GNP errors were three-tenths of one per cent or less. To assess the effect of money stock control, moving outside the sample periods, the standard error of forecast statistic was developed to permit the construction of appropriate confidence in tervals for GNP projections. For the four quarters of 1970, the money stock control procedure only reduced the probability associated with the 95 per cent con fidence interval to 93.3 per cent. The final judgement on airy monetary policy pro cedure ultimately rests with the members of the Fed eral Open Market Committee. As an ideal situation the FOMC would want no errors in achieving their policy objectives. However, this ideal cannot be real ized. Therefore, the policymakers must have some means of comparing the effects of different control procedures on their ability to achieve their policy ob jectives. This article presented some information on these matters for a money stock control procedure. The Federal Reserve in operating such a money stock control procedure would have additional infor mation that could be used to more closely monitor its control process. The multiplier-base framework used in this article is taken from a fully developed specifi cation of the money supply process, within which the influence of changing economic conditions on the money supply process may be analyzed. Also, a per centage change in the money stock may be decom posed into the percentage changes due to the net source base and the multiplier. The percentage change in the multiplier may then be broken down into the percentage change due to each of its com ponents. For example, at times when large inflows and outflows of time deposits are induced by changes in market rates relative to Regulation Q ceilings, this factor may exert an important influence on the money multiplier. Using this additional information, the Federal Reserve should be able to improve its control of the money stock.22 When a money stock control procedure is suggested, a question that is frequently raised is “What does such 22Also, this procedure does not imply that the value of the multiplier forecast for the coming month at the end of this month should be the one used throughout the month. Each week of the month, as new data on the money stock in the preceding period becomes available, a new forecast of the monthly multiplier could be prepared. Based on this addi tional information, the net source base target for the month might be altered. Page 18 OCTOBER 1971 a procedure imply for the stability of the money market?” The Federal funds rate is commonly used as a summary measure of short-run (daily or weekly) conditions in the money market. If, as implied in our simulations, the Desk had exactly achieved its tar geted net source base level each month, would there have been significantly greater fluctuations in the Federal funds rate? The answer to this question re quired the use of a tested, very short-term, money market model that relates daily or weekly fluctua tions in the Federal funds rate to changes in the net source base. Unfortunately, such a model is not avail able. Using quarterly models, some evidence can be gained on the quarterly average results of money stock control on interest rates. However, these results are not satisfactory to individuals interested in daily or weekly fluctuations. The money stock control procedure in this article does not necessarily require that the Desk hit the targeted level for the net source base each day or week of the month. The Desk is to attain a dailyaverage monthly net source base target. Therefore, as a practical operating strategy, the Desk would have some latitude to offset short-term shocks to the money market within each month. However, the Desk would have to give primary consideration to achieving the net source base target. The Desk would have to guard against allowing one short-term special situation to be followed by another, resulting in a deviation of the target base level from the one necessary to achieve the desired monetary growth path. One tentative piece of evidence related to the problem of aggregate control versus money market stability has been presented by Richard Davis.23 Davis analyzed the effect that control of nonborrowed reserves would have had on short-term money market rates for a sample period in 1967. He concluded that Having said that certain features of the experiment tend to overstate the degree of potential money m ar ket instability, however, the writer is inclined to the view that the degree of instability indicated is never theless rather surprisingly mild. T h e computed aver age absolute weekly change in the Federal funds rate tends to be only around 5 0 basis points, cer tainly substantially larger than the average changes 23Davis’ method consisted of computing the weekly levels of free reserves that would have resulted during an historical time period if the System had provided a constant weekby-week growth in nonborrowed reserves during that pe riod, given the historical pattern of actual changes in re quired reserves. An equation relating the Federal funds rate to free reserves and the discount rate is then used to estimate what the funds rate would have been had the System followed the quantity target. The computed funds rate was then compared to the actual rate for the period. FED ER AL R ESER V E BANK OF ST. LOUIS OCTOBER 1971 that actually occurred (around 17 basis points), but not more than the market would seem able to handle without undue stress.24 It is important to the well-being of the whole econ omy that monetary policy be implemented using the procedure that offers the highest probability of policymakers achieving their policy objectives. Other economists have proposed alternative strategies for implementing policy. However, a useful comparison of our procedure with these alternative procedures 24Davis, p. 61. can only take place when these alternatives are ex plicitly formulated and the effects on GNP of using these procedures is illustrated. Criticism of this money stock control procedure is welcomed. Propo nents of other policy procedures are challenged to explicitly formulate their proposals so the effects of these procedures on attaining policy objectives can be analyzed. In this manner monetary policy can be implemented and improved on the basis of empirical evidence, rather than implemented on the basis of conjecture, personal belief, and tradition. APPENDIX In this appendix the technical aspects of the forecast ing equation for the money multiplier, the error response mechanism, and the development of the standard error of forecast statistic are presented. Forecasting Equation for the Money Multiplier as a seasonal dummy variable. The influence of the d, is to shift the intercept from period to period. (3) The reserve adjustment magnitude is introduced to capture the effects of reserve requirement changes. Reserve adjustments are expressed in dollar amounts which are positive when average reserve requirements fall and are negative when reserve requirements rise.27 Xi = three month moving average of past value of the multiplier, (4) The regression equation’s Durbin-Watson (D-W) statistic indicates the existence of significant autocorrela tion in the equation’s errors. With this condition it is possible to get improved estimates of the money multi plier over time by including an additional variable in the prediction equation. This variable which “allows” for the autocorrelation is the lagged value of the error (Ht-i) in the estimate of the money multiplier times the correla tion coefficient Rho (p) for consecutive-error terms in the equation during the sample period.28 Xa = reserve adjustment magnitude in the forecast month, One means of judging the forecasting ability of the multiplier equation is to compare the root mean square di = dummy variables to account for seasonal factors,20 Each month’s multiplier is forecast using the following equation:25 nit = bo + b iX it + b ^ X 2t + 2 bj_j_2di + PHt—i i= l where: ( 1 ) 27Shifts of deposits between banks with different legal re serve requirements and between different deposit categories ( demand to tim e) also exert a slight influence on the month-to-month changes in the reserve magnitude. The variance of the monthly first differences of the reserve ad justment magnitude during 1963-69 was six times greater (-it— = lagged value of the error in the estimate i of the money multiplier. when all months were included, than when months in which reserve requirement changes took place ( and adjacent months) were excluded. For the 1962-69 period, except for (2) The coefficients bi are estimated by least squares months where reserve requirements were changed, the Fed using the previous 36 months’ observations. Each month eral Reserve could have assumed the forecast month’s re serve adjustment magnitude would be the same as the cur the coefficients are re-estimated by adding the most re rent month’s value without an appreciable error over the cent month and dropping the first month of the previous period. p = the correlation coefficient for consecutive error terms in the equation during the sample period, 36 observations. b0 is an intercept term which also acts 25Parameter estimates of the monthly forecasting equations are included in the technical appendices in Working Paper No. 14. 26Similar results were obtained by omitting the seasonal dummy variables and instead adjusting not seasonally ad justed money by the seasonal factors used by the Federal Reserve. For an explanation of the method by which the reserve adjustment magnitude is computed, see Leonall C. Ander sen and Jerry L. Jordan “The Monetary Base — Explana tion and Analytical Use,” this Review, August 1968, p. 8. 28This technique assumed first order autocorrelation. Rho is estimated as 1- DW 2 Page 19 OCTOBER 1971 F ED ERA L R ESE R V E BANK OF ST. LOUIS error (R M S E ) of the forecasting procedure with the R M S E of a no change extrapolation. T h e R M S E of monthly predictions of the money multiplier in the 196265 period was .0151, which was 5 3 per cent as large as the R M SE for a no change extrapolation. In the 1966-69 period the R M SE was .0200, which was 64 per cent as large as the R M SE for a no change extrapolation. then it is necessary to modify the standard error of fore cast statistic associated with the forecasting equation. T he im portance of errors in money stock control can then be assessed by comparing the S E F of the GNP forecasting equation when there are no errors in money stock control with the standard error of forecast statistic ( S E F ° ) of GN P estimates when there are errors in money stock control. Error-Response Mechanism One statistic frequently used by economists to assess the forecasting ability of an econom ic model is the stand ard error of estimate ( S E E ) . However, a more appro priate measure of the forecasting ability of a model is the correctly specified standard error of forecast ( S E F ) statistic.29 T he value of the S E F statistic depends upon the values assumed by the independent variables during the forecast period, and upon the variances and covari ances of the parameters of the forecasting equation, as well as upon the S E E statistic. O nce there is a possibility of error in money stock control, specification of an optimal operating strategy for changing the net source base requires that the money managers’ loss function b e specified. There are many possible loss functions, each one representing somewhat different preferences by the policymakers. In our pro cedure a quadratic loss function of the following form is specified: L = (M — Md)2 money manager’s loss function M = mtB, actual money where: mt = money multiplier in period t B = net source base Md = desired money stock in period t. In the above expression the product of the money multiplier (m ) and the net source base ( B ) gives the level of money achieved in period t by our operating procedure. M d is the level of money consistent with a given desired growth rate of money. This type of loss function assigns proportionally more weight to large de viations of controlled money from desired than to smaller deviations. O nce the money manager’s loss function has been specified, the optimal strategy w ith respect to the net source base is the one that minimizes the expected value of the loss function. T h e expected value (w here E is used to denote an expected value) for this loss func tion may be written: E (L ) = B2 var (mt ) + [Md - B E (m ,)]2 Minimizing with respect to B , gives the following ex pression for the optimal net source base ( B ° ): Md DO E v ------ \ f --------- ----- j---- rl I, I B = E< mi) I 1 + var (m w l [EOM p I T o calculate the value for B ° in any period t, ( 1 ) we used M d in period t, which is determined by the desired growth rate; ( 2 ) we used E ( m t ) = predicted multiplier in period t; and ( 3 ) var (m t ) was approximated by taking the sum of squared residuals for the multiplier equation and dividing by 3 6 - K - 1, where K equals 14, the number of independent variables in the forecasting equation for the multiplier. Confidence Intervals for GNP Projections with Money Stock Control If a forecasting equation is used in which the money stock is assumed to be controlled by some procedure, Page 20 In this section, the S E F statistic which is appropriate for our policy procedure is presented and a comparison of it with the S E F which is associated w ith the A-J equation when money is assumed to be perfectly con trolled is presented.30 T h e equation used in this article to project G N P spe cifies the quarterly change in GN P as dependent upon current and lagged values of changes in money arid government expenditures. T o simplify the exposition, and to focus on the effect of errors in money stock con trol, the change in government expenditures is assumed to b e predicted without error. Therefore, errors are postulated to exist only in the GN P forecasting equation and in the control of money. T h e predicted change in money ( AM?-) n) in the forecast period and the actual change in money (AM, n) are related in the following manner: A M t-fn = A M t+ n + 6 t+ n where 0t+n is an error term. W ith errors in the money stock control procedure, an S E F which assumes perfect money stock control is no longer appropriate. This S E F statistic could be too large or too small.31 In other words, policymakers should not use this statistic to construct the confidence interval for their GN P forecasts. T h e probability of over-or underes tim ating GN P could be greater or smaller than that in dicated by this S E F statistic. 29For a discussion of this subject, see Carl Christ, E cono metric M ethods and Models, John Wiley Co., 1968, pp. 549-564. 30The derivation of the SE F statistic is presented in the technical appendices available as Working Paper No. 14. 31For example, if the forecasting equation for the money multiplier results in an overestimate of the money multi plier (which results in the actual money stock being less than the desired), and if, in addition, the A-J equation overestimates the effect of a change in the money stock on GNP, then the influences of the two errors (negative cor relation) tend to offset one another. However, when the errors in predicting the money stock and forecasting GNP are in the same direction (positive correlation), then the errors reinforce one another, and the error in GNP forecasts is increased. OCTOBER 1971 FED ER A L R ESER V E BANK OF ST. LOUIS The standard error of forecast statistic ( S E F ° ) for the A-J equation, assuming errors in money stock control, is given by: Table VI S ta n d a rd Errors o f Forecast For the G N P Equation for 1 9 7 0 1 SE F ° = [( S E F )2 + PI var ( 0 t + n ) + 2 |3l COV ( 0 t + n , E t + n ) ] 1/2 where: S E F denotes the standard error of forecast sta tistic when there are no errors in money stock control, and [3, is the coefficient for contemporaneous changes in the money stock in the GNP equation. Exam ining the expression for S E F ° , it can be seen that the existence of errors in money stock control in troduces two extra terms in the standard error of fore cast statistic, involving the error in the money stock con trol procedure (0tH n) and the error in the G N P equa tion ( e t + n ) . Term s with the variance ( 0 i : n ) and the covariance ( 0 t + n , £ t + n ) are introduced. Since in general these terms are not equal to zero, S E F ° is unequal to S E F . As remarked earlier, the cov ( 0 t + n , £ t + n ) may be either positive or negative. If it is positive, then this factor would increase the S E F 0; if negative, it might be large enough to make S E F “ less than S E F .32 Standard error of forecast statistics are dependent upon the particular values of the independent variables which apply to the prediction period. In particular, an S E F statistic assumes its absolute minimum value when the respective independent variables which enter into it take on values that equal their sample means. All other sets of values of the independent variables will generate larger values of the S E F statistic. T he reason for this result resides in the statistical uncertainty surrounding the regression estimates of the coefficients in the fore casting equation. For the period 1/1953 -11/ 1969, the minimum S E F of the GN P equation is $3.87 billion. As an illustration of how the S E F statistic actually differs in practice from its minimum value, let us consider the hypothetical problem of predicting GNP for each of the four quarters of 1970. Under conditions of perfect money stock control, the S E F statistic would assume the values given in row A of T ab le V I which are from approximately 7 to 14 per cent larger (row B ) than the minimum value of the S E F statistic. For the money stock control procedure, the S E F " sta tistics are given in rows C and F of Table V I. T he S E F ° statistics in row C correspond to the quarterly average performance of the monthly money stock control proce dure in the simulations with the GNP equation. The SEF® statistics in row F correspond to a quarterly money stock control procedure, and are rigorous “outer bounds” to the S E F ° statistics given in row C .33 32For any given quarter, it is possible for the S E F 8 statistic to be smaller than the traditional S E F statistic if certain conditions are met. Let p be the correlation between the error in money stock control (0 t) and the error in predict ing GNP (£t). The following can be shown to hold. If < Pi [var 9t ]i/2 holds, var £t then S E F * < SEF. 33The simulation value of v a r(0 t), or simulation var(0t), is (Dollar Am ounts in Billions) 1/1970 SEF with no error in money stock control 11/1970 111/1970 IV / 1 9 7 0 4.259 A $ 4 ,143 4.409 4.198 B 7 .1 % 13 . 9 9 % 8.5 % C $ 4 ,2 6 7 4.5 2 6 4.3 2 0 4.379 Percentage increase of C over A D 3.0 % 2.7 % 2 .9 % 2 .8 % Ratio of B to D E 2.4 5.2 2.9 3.6 F $ 4 ,9 8 6 4.7 1 4 4.503 4 .5 5 0 Percentage increase of F over A G 20 .3 % 6.9 % 7.3 % 6 .8 % Ratio of B to G H .35 2.02 1.2 1.5 Percent increase of A over minimum SEF of $3,868. 1 0 .1 % SEF* with error in money stock control: simu. var. ( 0 t ) and simu. cov. (0t,£t) SEF* with error in money stock control: equa. var. (0t) and simu. cov. (0t,£t) where: 1) variance (E t ) — 2) simulation variance (0 t) — 0.6 7 4 14.7 4 9 (corresponds to SEE of 3 .840) 3) simulation covariance (0 t,E t) = 0 . 0 1 1 4) equation variance (0i t)> 1/1970; (quarterly multiplier equation R2 = 0.2 7 ) 11/1970; 1.835, (quarterly multiplier equation R2 — 0.5 0 ) 111/1970; 1.751, (quarterly multiplier equation R2 = 0.5 9 ) IV/1 970 ; 5) 5.1 19, 1.696, (quarterly multiplier equation R2 = 0.7 7 ) G N P equation is the A-J 1/1953 to IV/1 9 6 9 ). equation (sam ple period irThe S E F sta tistic s in each q u a rte r a re com puted u sin g actual changes in m oney and g o v e rn m e n t e x p en d itu re s, except th a t th e contem po raneous change in m oney is given as th e sim ple 4 p e rc e n t increase over th e lagged level o f th e m oney stock. T h e S E F ° statistics associated with our monthly money stock control procedure are only 3 per cent larger than the S E F statistics assuming no errors in money stock control (see row D ). T o further understand the implicaprobably the most reasonable estimate for the variance in money stock control, given that a monthly money stock control model is used in conjunction with a quarterly fore casting model of GNP. Unfortunately, the statistic simula tion var(0t) makes no allowance for the imprecision in the coefficients of the multiplier forecasting equation. However, as is shown in the technical appendices, this shortcoming can be overcome if the quarterly money stock control pro cedure is used. In that case, the equation var (0 t) equals B r-i X ( S E F )2 (quarterly multiplier equation) which is the most appropriate estimate of var (0 t). This variance is just the square of the lagged level of the base times the squared standard error of forecast of the multiplier forecast ing equation. Page 21 FED ER AL R ESE R V E BANK OF ST. LOUIS tions of the standard errors of forecast in Table V I, it is helpful to refer back to Table V on page 17. The data in Table V show that the errors in money stock control have a negligible effect on the policymaker’s ability to forecast GNP, if 95 per cent confidence intervals are used as standards of comparison.-14 :!4An illustration of how the confidence intervals in Table V were computed is given below. Consider the perfect money stock control case given in the first quarter of 1970 entry in row A of Table VI, SEF = $4,143 billion. Because the errors in predicting changes in GNP can be shown to be normally distributed for large samples ( see the technical appendices in Working Paper No. 14), it is appropriate to set up confidence intervals, using a table of the standard OCTOBER 1971 normal distribution. A range of ± 1 .9 6 standard deviations gives a 95 per cent confidence interval for the standard normal distribution, whose standard deviation is unity by definition. Consequently, the distribution of normally distributed errors with an SE F value of $4,143 has a proportionately larger 95 per cent confidence interval of ± (1.96) X (4.143) equals ± $8,120. When errors in money stock con trol raise the value of the SE F statistic to $4,267 (row C ), the probability of achieving the same confidence interval of ± $8,120 billion is reduced to 94.26 per cent, since now only ± 1.90 standard deviations of the standard normal distribution will give that same confidence interval, [i.e., ± (1.90) X (4.267) equals ± 8.120] For a comparison of 90 and 80 per cent confidence intervals see the technical appendices available in Working Paper No. 14. This article is available as Reprint No. 72. FED ERA L R ESER V E BANK OF ST. LOUIS OCTOBER 1971 WORKING PAPERS S IN G L E COPIES of the following working papers are available to persons with a special interest in these research areas, and any discussion or comment would be welcomed by each author. For copies write: Research Department, Federal Reserve Bank of St. Louis, P. O. Box 442, St. Louis, Missouri 63166. Title of Working Paper Release Date The Three Approaches to Money Stock Analysis (Now available in our Reprint Series as No. 24) July 1967 Chapter on Agribusiness Prepared for American Institute of Banking Textbook Agricultural Credit (50 pages) Aug. 1967 Monetary Policy and the Business Cycle in Post war Japan (108 pages) Revised April 1968 The Influence of Fiscal and Monetary Actions on Aggregate Demand: A Quantitative Appraisal (53 pages) Revised March 1969 The Development of Explanatory Economic Hyphotheses for Monetary Management (48 pages) Nov. 1968 A Model of the Markets for Consumer Instalment Credit and New Automobiles (60 pages) Jan. 1969 A Summary of the Brunner-Meltzer Non-Linear Money Supply Hypothesis (65 pages) Revised May 1969 The Market For Deposit-Type Financial Assets (205 pages) March 1969 Impact of Changing Conditions on Life Insurance Companies (23 pages) March 1969 Adjustments of Selected Markets in Tight Money Periods (206 pages) June 1969 11 A Study of Money Stock Control (41 pages) July 1969 12 Empirical Test on the Effect of Changes in Money Supply in Developing Economies May 1970 13 Historical Analysis of the “Crowding Out” of Pri vate Expenditures by Fiscal Policy Actions January 1971 14 Money Stock Control and Its Implications for Monetary Policy: Technical Appendices October 1971 Number 1 2 3 4 5 6 7 8 9 10 Page 23 S u b s c r i p t i o n s to this bank’s R e v ie w are available to the public without charge, including bulk mailings to banks, business organizations, educational institutions, and others. For information write: Research Departinent, Federal Reserve Bank of St. Louis, P. O. Box 442, St. Louis, Missouri 63166.