The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Vol. 29, No. 4 ECONOMIC REVIEW 1993 Quarter 4 Required Clearing Balances 2 by E.J. Stevens The CPI as a Measure of Inflation 15 by Michael F. Bryan and Stephen G. Cecchetti The Inaccuracy of Newspaper Reports of U.S. Foreign Exchange Intervention by W illiam P. Osterberg and Rebecca Wetmore Humes FEDERAL RESERVE BANK OF CLEVELAND 25 G E E M IC R E V I E W 1993 Quarter 4 Vol. 29, No. 4 Required Clearing Balances 2 by E.J. Stevens More than 20 percent of the funds that banks have on deposit with the Federal Reserve Banks are required clearing balances, not required re serve balances. Since 1981, when they first earned a market return, clearing balances have become widespread among banks of all sizes. Here, the author takes a look at the reasons for the popularity of this rela tively new phenomenon as well as its impact on the setting and measur ing of monetary policy. The Consumer Price Index as a Measure of Inflation 15 Economic Review is published quarterly by the Research Depart ment of the Federal Reserve Bank of Cleveland. Copies of the Review are available through our Public Affairs and Bank Relations Depart ment. Call 1-800-543-3489, then immediately key in 1-5-3 on your touch-tone phone to reach the pub lication request option. If you pre fer to fax your order, the number is 216-579-2477. Coordinating Economist: William T. Gavin by Michael F. Bryan and Stephen G. Cecchetti One problem associated with using the Consumer Price Index as a focal point in monetary policy deliberations is the likelihood that it is a biased measure of inflation. The authors use a simple statistical frame work in this paper to estimate a price index that is immune to some of these weighting biases. By computing the common inflation element in a broad cross-section of consumer price changes, they find evi dence of a positive weighting bias between 1967 and 1981, and an insignificant bias in the years since then. The Inaccuracy of Newspaper Reports of U.S. Foreign Exchange Intervention 25 by W illiam P. Osterberg and Rebecca Wetmore Humes This paper presents a comparison of official data on U.S. foreign exchange intervention with newspaper reports. The authors find that the series are systematically different, which calls into ques tion the ability of intervention to signal monetary policy accurately. Alternatively, this divergence may reflect the fact that not all mar ket participants have equally accurate information about exchange market intervention. Advisory Board: Ian Gale Jagadeesh Gokhale Joseph G. Haubrich Editors: Tess Ferg Robin Ratliff Design: Michael Galka Typography: Liz Hanna Opinions stated in Economic Re view are those of the authors and not necessarily those of the Fed eral Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material may be reprinted pro vided that the source is credited. Please send copies of reprinted material to the editors. ISSN 0013-0281 Required Clearing Balances by E.J. Stevens E.J. Stevens is an assistant vice president and economist at the Federal Reserve Bank of Cleve land. The author thanks Cheryl L. Edwards for helpful comments on an earlier draft of this article and Ann Dombrosky for invaluable re search assistance. Introduction Few people realize that in addition to complying with a Federal Reserve System regulation for holding a required reserve balance, many banks simultaneously meet an additional requirement to hold a clearing balance in their account at a Federal Reserve Bank. This clearing balance re quirement differs from the familiar reserve re quirement in three significant ways. First, a bank’s agreement to meet the requirement is typically a business decision, not a legal necessity. Second, the amount of the requirement is mostly discre tionary, not a fixed percentage of the bank’s de posit liabilities. And third, the rate of return on a clearing balance is about equal to the federal funds rate, not zero, although a bank can use the earnings only to pay for services it buys from a Federal Reserve Bank. About 5,000 banks now maintain required clearing balances, ranging from small retail de positories with a $25,000 minimum requirement to giant money center banks with clearing bal ance requirements of several hundred million dollars.1 Forty-five percent of all required re serve balances and 85 percent of all required clearing balances are held by banks that use a http://fraser.stlouisfed.org/ mixed deposit account at a Federal Reserve Federal Reserve Bank of St. Louis Bank to comply with the combined require ments. For each of these institutions, the conse quences of modest account deficiencies or surpluses are reckoned on the basis of re quired clearing balance rules. Only if a bank fails to meet any of its clearing balance require ment does it face the familiar “discount rate plus 2 percent” penalty for a required reserve deficiency. Likewise, such a bank wastes sur plus balances by earning no rate of return only if its actual balance exceeds its required bal ance by more than a preestablished margin. Almost all analyses of bank reserve manage ment behavior focus entirely on reserve require ments, ignoring clearing balances because they are a relatively recent innovation.2 Clearly, how ever, an important set of banks now maintains balances at the Fed following a somewhat differ ent set of rules than they would if they held only required reserve balances. Knowledge of these rules is of more than accounting interest. For one thing — contrary to most models of the banking ■ 1 For simplicity, I use the term banks to mean all depository institu tions. ■ 2 A good survey might start with Poole (1968) and include Coats (1976), Spindt and Tarhan (1978), Friedman and Roberts (1983), Evanoff (1989), and Feinman (1993). FIGURE 1 Required Balances at Federal Reserve Banks Billions o f dollars SOURCE: Board o f Governors o f the Federal Reserve System. system and of monetary policy implementation — standard indicators of Federal Reserve policy, including total and excess reserves and the monetary base, probably contain a growing (albeit very small) component that effectively has a positive rate of return. For another thing, banks seem to be substitut ing required clearing balances for required re serve balances. In the aggregate, banks now hold about $33 billion of required balances at the Fed eral Reserve, including $27 billion of required re serve balances and $6 billion of required clearing balances. Required reserve balances have de clined about $6 billion over the past three years, while required clearing balances have almost tripled (see figure l).3 Although banks surely welcome lower reserve requirement taxes, the Federal Reserve must deal with the payment sys tem risk and monetary policy implementation re percussions of a banking system operating on a smaller cash deposit base. These repercussions may be muted, however, to the extent that banks replace required reserve balances with required clearing balances. The question, then, is why have required clearing balances grown so rapidly, and, more particularly, would further cuts in re serve requirements be offset by further growth of required clearing balances? ■ 3 The dollar volume of required clearing balances is reported as foot note 3 In Factors Affecting Reserve Balances of Depository Institutions and Condition Statement o f F.R. Banks (Federal Reserve Release H.4.1), and as part of the larger “service-related balances and adjustments" item in the data table "Reserve and Depository Institutions and Reserve Bank Credit" (Federal Reserve Bulletin, table 1.11). Clearing balance data are also reported in the pro forma balance sheet for Federal Reserve priced services in the Board of http://fraser.stlouisfed.org/ Annual Report. Federal Reserve BankGovernors’ of St. Louis This article is intended primarily to describe the little-known rules governing required clear ing balances and to introduce some related issues. The first two sections include back ground information about clearing balances and a look at how a bank might manage a combined required reserve and required clear ing balance. More precise institutional details are spelled out in the appendix. The third sec tion outlines three areas in which issues war rant further investigation. One is the way in which measures of bank reserves and the m on etary base have come to include an interestbearing component as a result of required clearing balances. The next points to ambigu ity in explanations of the rapid growth of re quired clearing balances. A third sketches some related central banking concerns about monetary policy implementation and payment service delivery. I. Some Background The Depository Institutions Deregulation and Monetary Control Act of 1980 extended coverage of Federal Reserve System reserve requirements from member banks to all depository institutions. At the same time, all depositories gained access to the Federal Reserve discount window and to Reserve Bank payment services. Services had to be priced at levels intended to recover their full cost of provision, including an allowance for the interest costs and profit required by competing private suppliers, such as conespondent banks. Fair pricing requires a careful cost-accounting distinction between Federal Reserve non priced, central bank activities such as reserve requirements and the discount window, and its priced semice activities such as check clearing. Both activities may lead a bank to maintain a deposit balance in an account at a Federal Re serve Bank. Banks whose vault cash is not sufficient to satisfy their reserve requirement can meet the remainder of the requirement in either of two ways. They can maintain funds in a deposit ac count at a correspondent bank on a pass through basis, making it easier for them to use the services of correspondents without also having to maintain an account at a Federal Re serve Bank to handle their payment needs. Al ternatively, banks can maintain the required funds on deposit at a Federal Reserve Bank. The Federal Reserve’s priced service activi ties are those for which it is not the sole sup plier. These include collection of commercial checks, processing of commercial automated clearinghouse items, wire transfer of funds and government securities, safekeeping of definitive securities, collection of noncash items, and transportation of cash.4 A bank that uses these priced services needs an account to which its payments and receipts can be posted. Three possibilities exist. First, a bank can contract to have its activity posted to a correspondent bank’s Fed account. As just noted, this might be especially appealing to those institutions already maintaining required reserve deposits at a correspondent on a pass through basis. Second, a bank that maintains a required reserve deposit at a Federal Reserve Bank might have its activity posted to that ac count. Flistorically, this has been the typical choice of banks using the Reserve Banks’ pay ment services. Third, if required reserve depos its are unnecessary or inadequate for transaction purposes, a bank might maintain additional bal ances in its Fed account under the required clearing balance anangement.’ Since 1981, banks have been able to pay for priced services with earnings credits on required clearing balances. ■ 4 The Reserve Banks also provide fiscal agent services for the U.S. Treasury. A bank may use its Fed account to make and receive payments associated with these services, whose costs are covered by fees paid by either the banks or the Treasury. ■ 5 Federal Reserve Banks have always been able to provide an ac count for customers who do not need to keep a required reserve balance but who wish to make and receive payments there. Flistorically, this nor mal banking practice apparently involved only an incidental aggregate http://fraser.stlouisfed.org/ amount of overnight balances in clearing accounts. Federal Reserve Bank of St. Louis Whatever its choice, a bank will want to en sure that unexpected charges to its account do not result in either penalties for daylight or over night overdrafts or reserve deficiencies, and that unexpected receipts do not lead to “wasted” excess reserves. Moreover, the Federal Reserve Bank will need some assurance, as a prudent banker, that charges to a bank’s account are made against a sufficient balance. Required clearing balances address these needs. The mechanism for maintaining required clearing balances and receiving earnings credits was introduced in 1981 and modified in 1982 to include a penalty-free band.6 (See the ap pendix for a more detailed description.) This anangement is comparable to the compensating balance method that some respondent banks and commercial firms use to pay for commercial bank services. A required clearing balance is an average amount that a bank contracts to hold in a deposit account during a reserve maintenance period. This balance is over and above any re quired reserve balance it must hold in that period. A bank’s required reserve balance is determined by deducting its holdings of vault cash from its total required reserve, which in turn is a percent age of the bank’s deposits specified by regula tion. A bank’s required clearing balance is selfdetermined, presumably so that it may avoid overdrafts and receive earnings credits commen surate with its monthly bill from the Federal Re serve Bank. Periodically, a bank’s actual maintained bal ance is compared with its required balance. Each business day, payments flow into and out of a bank’s Federal Reserve account. For each institution, the Reserve Bank records end-ofday account balances and then averages these maintained balances for a reserve maintenance period of one or two weeks, depending on the size of the bank. If the bank has no required clearing balance, its average maintained balance (after certain “carryover” adjustments discussed below) should equal or exceed the required re- ■ 6 Each of the 12 District Banks provides official circulars and other marketing materials informing customer banks about the terms on which priced services are available, including the required clearing balance op tion. Operating to some extent as 12 distinct businesses, Banks have fol lowed procedures for maintaining clearing balances that have differed somewhat in detail in the past. The most notable ditference was in allocat ing maintained balances between reserve and clearing requirements. Some Banks allocated balances first to the required and then to clearing requirements, while others did the reverse. This practice affected the pen alty structure on the initial amount of a deficiency, with some Banks as sessing required clearing balance penalties and others assessing required reserve penalties. However, banking consolidation across Fed eral Reserve District lines, as well as consolidation of some operations among the 12 Banks, has led to the uniform current set of procedures de scribed here (see Conference of First Vice Presidents [1993]). 5 serve balance (required reserves minus applied vault cash). If the bank also has a required clear ing balance, the average maintained balance, after carryover, should equal or exceed the total re quired balance plus or minus a penalty-free band.7 Maintained balances satisfy the reserve re quirement first, and the remainder is used to satisfy the clearing balance requirement. A bank is penalized if its balance falls short of the required amount by more than a penalty-free band, at the rate of 2 percent on amounts up to 20 percent of the required clearing balance, then 4 percent on amounts up to the whole re quired clearing balance, and at the discount rate plus 2 percent on any remaining deficien cies in required reserves. The bank receives earn ings credits, based on the daily effective federal funds rate, on balances in excess of its required reserve up to the amount of its required clearing balance plus the penalty-free band (adjusted according to the bank’s marginal reserve ratio; see appendix).8 Beyond that point, a bank penal izes itself for balances in excess of the required amount plus the penalty-free band, because the excess funds receive no earnings credits. The penalty-free band is the greater of 2 per cent of a bank’s required clearing balance or $25,000. Thus, a bank with a minimum required clearing balance of $25,000 could satisfy the re quirement and receive earnings credits on the amount by which its balance exceeds its required reserve by anywhere from zero to $50,000, with out penalty. A bank with a $200 million required clearing balance would receive earnings credits on any balance above its required reserve up to $204 million, and would satisfy its clearing bal ance requirement without penalty when this bal ance reached $196 million. Required clearing balances affect the Fed’s cost of providing priced services in two largely offsetting ways. Total cost includes the earn ings credits that Reserve Banks grant on clear ing balances ($177.8 million in 1992), reduced by an offset for unused credits. Total cost also is lowered by an offset for the income that Re serve Banks earn on assets financed with re quired clearing balances. This offset is imputed at the coupon-equivalent yield on three-month Treasury bills ($180.2 million in 1992).9 Frequent stops to fill up take time and may preclude unforeseen opportunities to buy gas at a lower price. However, buying gas at a low price only when the tank is nearly empty risks running out of gas. So, too, a bank that buys and sells funds frequently in order to keep its balance close to the required amount at all times may waste opportunities to buy or sell at bargain rates, while buying or selling only when the funds rate is a bargain raises the risk of overdrafts and failure to satisfy require ments, or of wasted balances. Banks work toward a target balance over seven or 14 calendar days (normally, five or 10 banking days), depending on the size of the insti tution. Thus, a bank’s cumulative required balance is seven or 14 times its average required balance. During the period, the account manager has a daily opportunity to target the day’s closing bal ance to add to the cumulative maintained balance. The cost of financing an extra dollar of bal ances is essentially the rate at which a bank might borrow or lend in the federal funds mar ket. This rate can vary noticeably over the course of a single banking day, over differing risk categories of borrower, and over days of a maintenance period. A bank does better, the more likely its manager is to “hit” the market when the rate is attractive, in effect filling up the fuel tank at places where gas is cheapest. Clearly, the attractiveness of the rate depends on the manager’s judgment about how expen sive funds are relative to what they might be over the remainder of the period (and, with carryover, over the following period). Ultimately, at the end of a maintenance pe riod, the value of an extra dollar of balances is determined by the structure of penalties and earnings credits within which the Reserve Banks administer requirements, including their permissiveness in waiving penalties. For a bank operating with only a required reserve bal ance, ignoring carryover, a deficiency would be penalized at a rate 2 percent above the discount rate, and frequent deficiencies would bring con sultations aimed at changing management’s be- ■ 7 Hereafter, “required balance" will be used to indicate the sum of a required reserve balance and a required clearing balance. II. Managing a Bank’s Fed Balance In general, managing a bank’s balance at its Fed eral Reserve Bank is rather like managing one’s fuel supply on an extended automobile trip. ■ 8 Earnings credits are not added to the balance in the account, but accumulate for use in offsetting charges for priced services (on a first-in, first-out basis) within 52 weeks. ■ 9 The amounts of these two items for calendar years are reported as components of “Other income and expenses’’ in the pro forma balance sheet for Federal Reserve priced services, published in the Board of Gov ernors' Annual Report. 6 havior. An excess would be wasted, costing something to finance but earning no interest. With this general background, the rationale for a bank’s decision to hold a required clearing bal ance can be investigated in three different time dimensions — a day, one or two maintenance periods, and the long run of many maintenance periods. A Day Uncertainty can be a dominant factor in a sin gle banking day. For a full-service bank, par ticularly a very large one, the level of its final end-of-day balance results from its last-minute interbank loan market maneuvering. A bank is likely to be involved in daily payments and re ceipts whose aggregate value is thousands of times larger than the required reserve balance. Thus, even slight deviations of payments or re ceipts from projected levels might flood or drain the bank’s Fed account during a day rela tive to the typical desired end-of-day balance. The manager of this account nonetheless must be able to come close to a targeted daily bal ance by arranging overnight borrowing or lend ing in the waning hours of the banking day in amounts that can be far larger than the target balance itself, and at attractive rates. The protection that required clearing bal ances provide against daylight and overnight overdrafts has become increasingly important over the past decade. Reductions in reserve re quirements and increased use of vault cash to satisfy requirements have left banks with smaller required reserve balances, but with no necessary change in the volume, time pattern, or predictability of charges and receipts for transactions. All else equal, this would be ex pected to increase the size and incidence of overdrafts, both daylight and overnight. The Federal Reserve Banks have measured and monitored daylight overdrafts with increas ing precision over the last 10 years, with amounts in excess of a minimum slated to be come subject to a fee in April 1994. Overnight overdrafts already are subject to penalty (the greater of 2 percentage points above the dis count rate or 10 percent). A bank without a re quired clearing balance might employ a variety of strategies to reduce the probability of over drafts, including targeting a higher level of noninterest-bearing excess reserves. Contracting to hold a required clearing balance would accom plish the same thing at almost no net cost, as long as the bank could use its earnings credits. A Maintenance Period A bank needs a strategy for maintaining a set of overnight Fed balances whose average will most profitably satisfy its balance requirements for the period, taking into account both the previous period’s reserve surplus or deficiency and the possibility of carrying a reserve defi ciency or surplus into the next period. No single reserve management strategy ap pears to dominate banking practice. Some banks target the average daily balance neces sary to meet the required balance (perhaps in cluding a margin of safety), recalculated daily for the remaining days of the maintenance pe riod. Others try to accumulate balances toward requirements only when the funds rate seems low relative to the expected rate for the pe riod. Still others deliberately keep a lean posi tion early in a period, lest a negative surprise in required reserves or a positive surprise in re ceipts provide more excess reserves than they could work off over the remainder of the pe riod without overnight overdrafts. Also, some banks try to alternate surplus and deficient pe riods, while others aim for a stable positive av erage balance, using the carryover feature only to deal with big surprises. Carryover. Without a required clearing bal ance, a large bank’s average balance for a maintenance period could be above or below the required level by as much as 8 percent of its required reserves (not just its required re serve balance). Large banks are permitted to carry over to the next period a surplus or defi ciency of up to 4 percent of required reserves, but they cannot carry any resulting surplus or deficiency into the following period. Eight per cent would result from using the maximum al lowable carryover of a surplus/deficiency from the previous period and carryout of the maxi mum deficiency/surplus to the next period. Adding a required clearing balance widens the range within which a bank can allow its maintained balance to fluctuate from one period to the next while still satisfying requirements. With the addition of a required clearing bal ance, the bank could be above or below the (higher) required level by as much as 8 per cent of required reserves plus 8 percent of the required clearing balance. This is because the clearing balance requirement itself provides a penalty-free band of plus or minus 2 percent of the required clearing balance, and maximum allowable carryover is 4 percent of required re serves, plus 4 percent of the required clearing balance, minus the penalty-free band of 2 per cent of the required clearing balance. A smaller bank (required reserves less than $1.25 million) without a clearing balance require ment could be above or below the required level of balances by as much as $100,000, because banks may carry over the larger of 4 percent of required reserves or $50,000. However, adding a $25,000 minimum clearing balance requirement would not change the range within which that same small bank could allow its balance to vary. Allowable carryover would actually decrease to $25,000 ($50,000 net of the minimum $25,000 penalty-free band), offset by the ability to utilize that penalty-free band. The Long Run Opting to maintain a required clearing balance has obvious advantages for a bank. It can earn a market rate of return on relatively small bal ances that might not fetch such an attractive rate if sold as odd lots. Targeting a larger bal ance means, on average, holding a larger bal ance, thereby creating a greater buffer against daylight and overnight overdrafts. Moreover, the bank gains flexibility in managing its bal ance, with the penalty-free band providing a convenient, costless margin of error around the targeted balance that would be absent if it were targeting a zero balance or only a re quired reserve balance. With these benefits in mind, it might seem surprising that all banks do not obligate them selves for as large a clearing balance as their need for earnings credits would support. Pre sumably, this is because a bank’s capital is a scarce resource, if for no other reason than that banking regulations specify minimum lev els of capital per dollar of total assets. Banks may restrict the volume of their required clear ing balances to the level at which, at the mar gin, it is more profitable to allocate scarce capital coverage to other assets that promise a better return than the expected spread be tween the earnings credit rate on required clearing balances and their cost of financing. Adopting a required clearing balance thus has the following effects on a bank’s manage ment of its Fed balance: • It holds a larger balance, with the addition likely to be financed at little or no net out-ofpocket cost, but requiring a modest allocation of capital. • Its incidence of daylight and overnight overdrafts would likely be lower. • Allowable carryover, whether positive or negative, may be greater, providing a larger base either for interperiod rate arbitrage or for a bigger pool of funds from which to absorb unforeseen shocks to the closing balance on the last day, as well as for the period. • The penalty-free band absorbs small de viations of actual from target balances without either penalty or wasted earnings. III. Three Issues Related to Required Clearing Balances Measuring Bank Reserves and the Monetary Base Traditional measures of Federal Reserve mone tary policy activity, including total and excess re serves and the monetary base, are being affected by the growing influence of required clearing bal ances. The required clearing balance facility can create an interest-bearing component of meas ured bank reserves quite distinct from the tradi tional non-interest-bearing reserve assets. The potential influence of clearing balances can be seen by considering how the reserve and monetary base aggregates are constructed from six measured values, each of which is an average for a two-week reserve maintenance period. In addition to required clearing bal ances, the other five measured values include 1) Fed balances: The aggregation of over night balances of all depository institutions. 2) Applied vault cash: The amount of priorperiod vault cash holdings being used to sat isfy current-period reserve requirements. 3) Other vault cash: The difference between banks’ current and applied vault cash.10 4) Currency in M l: The portion of currency in circulation held by the nonbank public. 5) Required reserves: The total amount of reserves that banks are required to hold, as speci fied in Federal Reserve Regulation D. Data are derived from banks' reports of depos its to the Federal Reserve and are assembled with and without “adjustments to eliminate the effects of discontinuities,” or “breaks,” associ ated with changes in reserve requirements. The adjusted series estimates the amount of trans action deposit reserve requirements that would have prevailed in the past, had cunent reserve requirements been in effect; the unadjusted se- ■ 10 This is not the same as surplus vault cash (see Gartinkel and Thornton [1991]). F I G U R E 2 Reserve and Clearing Balance Requirements: The Rules NOTE: XR = excess reserves; RCB = required clearing balance; RRB = required reserve balance; AVC = applied vault cash; PFB = penalty-free band; Rf = fed eral funds rate; and Rd + 2 = discount rate. SOURCE: Author. ries reports the then-current actual requirements. Required clearing balances are excluded from all measures of reserves, by definition, and from the adjusted monetary base, but are included in the unadjusted monetary base. This treatment is consistent with the differing purposes of the two measures of the monetary base. The adjusted series emphasizes the role of base money as actual or potential reserve as sets. These are the high-powered “tickets” that banks must hold when issuing reservable de posits, with the amount issued per ticket con strained by a reserve requirement. The adjusted monetary base includes the reserve assets held both by banks (adjusted total reserves plus vault cash not being used to meet reserve requirements) and by the nonbank public (cunency in M l). A historically consistent measure of adjusted total reserves has been derived by adding the actual historical quantity of excess reserves to adjusted required reserves. Similarly, because banks’ bal ances held to meet a clearing balance require ment cannot be used to satisfy reserve require ments, the adjusted monetary base excludes http://fraser.stlouisfed.org/ required clearing balances. Federal Reserve Bank of St. Louis The unadjusted monetary base emphasizes the federal government’s role in providing monetary assets directly to users in the private sector, rather than distinguishing between the quantities of private and public issues of money. The monetary base consists of all feder ally issued currency held by banks and the public (applied vault cash plus other vault cash plus currency in M l), plus all deposit li abilities of Federal Reserve Banks to private banks (Fed balances, including required clearing balances). The associated measure of total reserves adds applied vault cash to Fed balances and then subtracts required clearing balances (because they are not reserves). Ex cess reserves is the difference between this measure of total reserves and required reserves. Measuring total or excess reserves thus in volves distributing aggregate account balances between reserve balances and clearing balances. The current method does so by measuring re serve balances as all cunent balances other than required clearing balances.11 Any excess of maintained balances above the required clearing balance level, even within the penalty-free band (for example, point A in figure 2), thus augments aggregate total and excess reserves. Similarly, any deficiency of maintained balances from the required clearing balance level, both within and below the penalty-free band (for example, point B in figure 2), reduces aggregate total and excess reserves, even though the bank may have satis fied its reserve requirement. A potential implication of this measurement convention can be illustrated by imagining an extreme case. Suppose that a ll banks were to move simultaneously from the upper to the lower edge of their respective penalty-free bands between adjacent maintenance periods. Within a 2 percent penalty-free band above and below the current $6 billion of required clearing balances, actual total and excess re serves would vary by about a quarter of a bil lion dollars, with banks largely indifferent to the change. That is, their earnings loss from holding a lower balance at the Fed would be approximately equal to their earnings gain from financing a lower balance. With many banks, some holding more and some holding less than their required clearing balances, positive and negative deviations from required clearing balances within penalty-free bands would likely ■ 11 Adjusted total reserves equals adjusted required reserves plus actual excess reserves, which in turn equals applied vault cash plus Fed balances net of required reserves and required clearing balances. Unad justed total reserves equals applied vault cash plus Fed balances net of required clearing balances. tend to be offsetting. More generally, however, the greater the participation in required clearing balance anangements, the more probable that modest variations either in the supply of bank balances at the Fed, in total and excess re serves, or in the adjusted monetary base would be a matter of little moment to banks, since their net earnings would be unaffected.12 The essential issue here is whether total and excess reserves, as now measured, match any useful economic concept. The measures have no necessary counterpart at the level of an individ ual bank managing its reserve position, because carryover and penalty-free bands are unrecog nized. Banks that perpetually maintain cunent balances in excess of cunent requirements truly have cunent-period “excess” reserves. Other banks, however, will be in different stages of us ing carryover, either satisfying some of their cur rent reserve requirements with surplus balances from adjacent periods, or using cunent surplus balances to satisfy some of their reserve require ments in adjacent periods. Carryover itself does not destroy the utility of the current measures: A positive shock to the supply of reserves in one period, for exam ple, tends to imply a comparable negative shock to demand for total and excess reserves in the next period, and the System can rely on that carryover relationship in managing nextperiod supply. The difficulty comes from the addition of a penalty-free band, which makes it impossible to know whether a shock to re serve supply will affect next-period demand through carryover, or simply be accommo dated as earning assets this period through the penalty-free band. Sources of Growth in Required Clearing Balances Managing a bank’s required reserve balance at successively lower levels of reserve require ments has been likened to landing an airplane on a shrinking aircraft carrier. As the target bal ance gets closer to zero, there is less room for error. Averaging within a maintenance period provides less opportunity to absorb surprises, as does the possibility of carrying forward ex cesses and deficiencies. Overall, the banking ■ 12 Paying interest on total or excess reserves would not preclude effective monetary policy. See Dotsey’s (1991) investigation of monetary policy operating procedures in New Zealand, where there are no reserve requirements and where banks settle using a below-market interest-bearing http://fraser.stlouisfed.org/ asset whose supply is controlled by the central bank. Federal Reserve Bank of St. Louis system becomes less effective in smoothing interest rates.13 With these impediments in mind, the rapid growth of required clearing balances in recent years might be linked to the cuts in reserve re quirements of December 1990 and April 1992.14 Banks increased their required clearing bal ances by more than a third in the month follow ing the December 1990 cut and doubled their requirements within a year (see figure 1). How ever, it would not be easy to distinguish the im pact of lower reserve requirements from that of either rising bills for priced services or declining interest rates. The utility of earnings credits lies in paying bills for priced services, so the size of these bills places an upper limit on the volume of clearing balances that banks could find useful. In the aggregate, the percentage of total sales of priced services paid with earnings credits, while growing, was still less than 20 percent in 1992. There is some indication that banks, including some with the largest required clear ing balances, do tend to adjust their require ments in concert with the magnitude of their bills. What would be difficult to discover, how ever, is the extent to which annual growth of billings has “caused” the growth of required clearing balances. More important for the fu ture would be to determine what portion of the remaining 80 percent of the priced services revenue billed to banks would be capitalized as additional required clearing balances if re serve requirements were cut further.1'’ Many banks could be expected to adjust balances to keep pace with bills because their required clearing balances are likely to be fi nanced at a slightly positive rate spread, mak ing priced services cheaper when paid from earnings credits. Earnings credits are calculated on the basis of the daily effective federal funds rate, which is the quantity-weighted average rate paid by all borrowers of unsecured over night balances each day. Large banks operat ing actively in the interbank funds markets ■ 13 Feinman (1993) provides an excellent analysis of these relation ships. ■ 14 The 1990 action reduced the 3 percent reserve requirement against nontransaction deposits to zero, lowering required reserves by an estimated $13.7 billion. The 1992 action reduced from 12 percent to 10 percent the highest marginal reserve requirement on net transaction de posits, cutting required reserves by an estimated $8.9 billion. ■ 15 Hilton, Cohen, and Koonmen (1993) have Investigated this question, as well as a variety of techniques that might expand the use of required clearing balances. thus might expect to acquire marginal financ ing at rates averaging less than the effective rate, because foreign buyers and some others typically pay risk premiums that large domestic banks, for example, do not pay. This would in sert a profit wedge between the effective rate used in calculating earnings credits and the cost of financing required clearing balances. With this in mind, some of the past growth in required clearing balances probably reflects the increase in total sales of priced services and the attraction of paying with earnings cred its. In fact, if this relationship were one for one, about 14 percent of the growth of re quired clearing balances since 1990 might re flect growth of total sales of priced services. Putting aside billing magnitudes, the level of the federal funds rate can also exert an inde pendent, powerful influence on the size of a required clearing balance needed to produce a dollar’s worth of earnings credits. For example, to hold earnings credits constant at their 1990 value, the substantially lower federal funds rate would have called for a 61 percent increase in required clearing balances by 1992. Even if, for purposes of argument, demand for required clearing balances had been direct ly proportional to billings and inversely propor tional to the level of the federal funds rate, banks added about $1 billion more to their holdings of required clearing balances after 1990 than the hy pothetical amounts these two forces would have produced. This suggests that banks have been induced to replace required reserve balances with required clearing balances. The relative in fluences of the three forces are not clear, how ever, because their movements have been conelated. Clarifying their relative importance will be crucial in dealing with some of the policy issues with which required clearing balances may be come associated. Monetary Policy Issues Reserve requirements are a tax whose cost has become a serious issue in the United States in recent decades, as the competitive niche of tra ditional banking has faded in financial markets. Lower requirements can increase Federal Re serve payment system risk exposures through daylight and overnight overdrafts, can contrib ute to volatile overnight interest rates that could hamper monetary policy implementation, and can degrade the value of central bank payment services (Stevens [1989, 1991b, 19931). Further cuts in reserve requirements might bring signifi cant institutional changes in banking and pay ment arrangements, with increased privatization of payment services to avoid daylight over drafts, or with new Federal Reserve arrange ments to ensure that deposit balances at the Fed remain an effective vehicle for monetary policy implementation (Meulendyke, ed. [1993], Stevens [1991a, 1992]). Additional cuts in required reserves could re duce the System’s effectiveness in interperiod smoothing of short-term interest rates. Reserve carryover plays a role in this smoothing proc ess, allowing the banking system to absorb un intended variations in the System’s supply of balances. The penalty-free band can serve the same purpose, but has different implications for policy implementation. Banks tend to “make u p ” reserve deficiencies and surpluses in the next period, providing the System with a vital clue to interperiod variations in demand for the balances it supplies.16 This is lacking in the op eration of the penalty-free band. Thus, the Sys tem could face multiperiod runs of demand for balances below or above a required level. An additional policy implementation prob lem may arise from the earnings credit feature of required clearing balances. Restrictive policies will carry within themselves the seeds of their own disorganization. That is, as the federal funds rate rises, the quantity of clearing balances needed to pay for a given quantity of Reserve Bank priced services will decline, increasing the possibility of the interest-rate variability associ ated with low balances. A high interest-rate pol icy might also discourage use of some Federal Reserve payment services, by reducing the nomi nal quantity of Fed balances available for imme diate transfer within overdraft limits. Relying on required clearing balances as the vehicle for implementing monetary policy thus raises a more general question. Is a bank’s re quired clearing balance a by-product of its choice of the Fed as the best among alternative suppliers of services, or is the choice of the Fed’s priced services a by-product of the bank’s need for a larger balance? In either case, the Monetary Control Act’s neat distinction between central bank activities and priced service activities is not as clear-cut as it once appeared. ■ 16 Feinman (1993) finds that for a sample of large banks from 1987 to 1991, excess reserves and carry in had opposite signs about 90 percent of the time. KB IV. Conclusion The emergence of required clearing balances is changing the institutional setting in which indi vidual banks manage their Fed balances. Banks are able to hold balances substantially larger than dictated by reserve requirements, provid ing greater flexibility in avoiding overdrafts and meeting reserve requirements— and at minimal cost. Familiar aggregate data series are being af fected by bank holdings of required clearing balances. In effect, a definable, probably small, but as yet unmeasured portion of the total and excess reserves of the banking system is now earning assets, rather than being held as noninterest-bearing vault cash or reserve deposits. More important, marginal variations in banks' Fed balances increasingly take place within the earnings and cost structure of required clearing balances, not required reserve balances. Growth of required clearing balances relative to required reserve balances raises questions that need further investigation. Can required clearing balances be expected to replace re quired reserve balances if reserve requirements are cut further? How would monetary policy implementation be influenced when a change in the money market stance of policy affects not only the marginal cost but also the mar ginal revenue of many banks’ Fed balances? To what extent does the demand for clearing balances reflect a desire to pay bills with earn ings credits, and to what extent does it reflect a demand for larger balances? If demand is mainly for convenient bill paying, could Fed eral Reserve priced services generate a pool of balances large enough to maintain a smoothly operating money market when interest rates are high? O n the other hand, if demand is mainly for a level of balances high enough to accommodate transaction needs, could banks use all of the Federal Reserve priced services their balances could buy when interest rates are high? Congress created the Federal Reserve Sys tem as a single response to the joint desire for a more uniform national payment system and for a regulator of the nation’s money supply. The mandate of the Monetary Control Act of 1980 was that these two functions should exist independently, in the sense that the Federal Reserve Banks could no longer provide free payment services to offset banks’ costs of main taining required reserves. Subsequent cuts in re serve requirements have allowed the banking system to reduce its holdings of non-interest bearing required reserve balances at the Federal Reserve Banks to historically low levels relative to bank deposits and the monetary base. All else equal, continuing along this trend would require some combination of changes in mone tary policy implementation and in the payment system to accommodate the absence of cash in ventories in the banking system. Alternatively, required clearing balances could provide a new basis for banks to hold deposits at the Federal Reserve Banks, but whether this is feasible re mains to be demonstrated. Appendix Required Balances: The Rules17 Current reserve and clearing balance require ments include two types of rules: those for computing and maintaining required balances and those for calculating earnings credits and penalties. Many banks are “unbound”— that is, either they have a zero reserve requirement or they meet the requirement entirely with vault cash. These banks nonetheless may maintain a re quired clearing balance. Other banks are “bound” by a positive reserve requirement that exceeds their vault cash. They must maintain a required reserve balance, but do not hold a re quired clearing balance. A large number of banks, however, are both bound to hold a re quired reserve balance and elect (or have been asked) to hold a required clearing balance. The rules laid out here, and summarized in figure 2, are for a bank that must meet a com bined reserve and clearing balance, maintained on the biweekly basis that is typical of a rela tively large institution. The other two cases may be derived by dropping all references to a required reserve balance or to a required clear ing balance, as the case may be. Note that in maintaining a balance, a bank that holds only a required clearing balance cannot use the car ryover feature, and a bank that holds only a re quired reserve balance cannot use the penalty-free-band feature. ■ 17 From Standard Operating Procedure 10.0, Conference of First Vice Presidents (1993). See also Board of Governors of the Federal Re serve System, Monetary Policy and Reserve Requirements Handbook, Washington, D.C.: Federal Regulatory Service. Computing and Maintaining a Required Balance A bank’s required balance, RB, is not a unique dollar amount, but a range around the com bined required balance. RB = (RRB + RCB) ± PFB. The combined required balance includes a required reserve balance, RRB, that is the bank’s total reserve requirement, RR, net of its applied vault cash, A VC. RRB = RR —A VC. The total reserve requirement is computed by applying appropriate marginal reserve re quirement ratios to the amount of a bank’s transaction deposit liabilities in each of three “tranches.” In 1993, requirements are zero on the first $3-8 million of deposits, 3 percent on additional deposits up to $46.8 million, and 10 percent on deposits in excess of $46.8 million. Requirements typical of large banks are computed on the basis of daily average transac tion deposit liabilities outstanding during suc cessive two-week reserve computation periods ending every other Monday. Applied vault cash is the bank’s daily average holdings dur ing the 14-day period that ends three days be fore the beginning of the maintenance period. The required clearing balance, RCB, is nor mally a dollar amount agreed to by the bank and its Federal Reserve Bank, with a $25,000 minimum. As stated by the Federal Reserve Bank of Cleveland (1992), The prescribed level of an institution’s clearing bal ance will be determined in consultation with the in stitution on the basis of the deposit size of the institution, the volume and type of services that are or will be used, and the need to avoid account overdrafts.... This Bank may make adjustments in the prescribed level of an institution’s clearing bal ance from time to time as may be appropriate. Such adjustments will normally be made no more than once a month and will be effective on the first Thursday of the month that coincides with the first day of a maintenance period. The penalty-free band, PFB, for required clear ing balances establishes a range of balances that will satisfy the combined required balance, be cause actual holdings are allocated first toward the required reserve balance, with the remainder allocated toward the required clearing balance. The band is 2 percent of the required clearing balance, or $25,000 if the required clearing bal ance is less than $1.25 million. A bank’s maintained balance, MB, is the av erage daily closing balance in its Fed account, averaged over a two-week maintenance period (for a typical large bank) that begins on a Thursday and ends on a Wednesday, two days after the end of the required reserve computa tion period. Carryover provisions allow a bank to carry forward to the next maintenance period an ex cess or deficiency in its maintained balance to the extent that it is offset by a deficiency or ex cess in the next period. The amount of carry over can be no more than (0.04 [RR + RCB] PFB) and cannot be carried forward more than one period. Note that the limit on eligible carry over is based on a bank’s reserve requirement, RR, not on its reserve balance requirement. Earnings Credits, Penalties, and Wasted Balances Earnings credits provide a return on required clearing balances that can be used only to pay for Federal Reserve Bank priced payment serv ices.18 Required reserve and surplus balances earn nothing. The return is based on the average federal funds rate during the maintenance period in which the required clearing balance was held. The funds rate is applied on an annualized basis to the actual average daily clearing balance (with in the upper limit of the penalty-free band), ad justed by the bank’s marginal reserve requirement ratio. A bank subject to the 10 percent marginal reserve requirement will earn the funds rate on its entire allowable clearing balance, a bank subject to a 3 percent marginal requirement will earn the funds rate on only 93 percent of that balance, and a bank subject to a zero mar ginal reserve requirement will earn the funds rate on only 90 percent of the balance. The marginal reserve requirement adjustment incorporates two factors that allow Federal Re serve Banks and correspondent banks to pro vide similar services to customer banks on the “level playing field” envisioned in the Mone- ■ 18 More specifically, earnings credits be used to pay pen alties for clearing balance deficiencies or to cover charges related to non priced service functions of the Federal Reserve Banks, such as penalties for deficient required reserve balances, Interest on discount window loans, and cost recoveries for providing accounting information services. tary Control Act. To illustrate, first suppose that correspondent banks have a 10 percent mar ginal reserve ratio and that the incidence of the cost of a correspondent’s reserve require ment is on its respondent customer banks. This suggests that the Reserve Banks might give earnings credits on only 90 percent of a required clearing balance to avoid placing themselves at an advantage relative to corre spondent banks in providing priced services. Second, recognize that a bank paying for correspondent bank services with earnings credits on balances held with the correspon dent is able to deduct the amount of those bal ances from its own deposit liabilities subject to reserve requirements. (Deducting amounts “due from other banks” avoids double-reserving of interbank deposits.) If the same bank were to buy services of equal value from a Federal Re serve Bank and pay for them with earnings credits on a required clearing balance, it would lose the deduction. This is irrelevant for a bank with a zero marginal reserve requirement, but not for those reserving 3 percent or 10 percent at the margin. Therefore, the Fed should give earnings credits on 93 percent or 100 percent of required clearing balances, depending on the customer’s marginal reserve ratio, to avoid placing itself at a disadvantage relative to corre spondents in providing priced services. Penalties are imposed on a bank whose maintained balance is deficient, to the extent that the deficiency is not offset by carryover from the previous period or to the next period. Maintained balances are allocated first toward the required reserve balance, with the remain der allocated toward the required clearing bal ance. A bank pays a penalty at an annual rate that rises with the size of the deficiency: no penalty on the first 2 percent (or $25,000) of the required clearing balance (the penalty-free band), 2 percent of the next 18 percent of the required clearing balance (or of the next 20 percent minus $25,000), and 4 percent of the remainder of the required clearing balance. Deficiencies that extend into the required re serve balance are penalized at a rate 2 percent age points above the discount rate. Balances can be said to be wasted to the ex tent that they exceed the required range and are not carried forward to the next period. Such balances do not contribute to satisfying a re serve or clearing balance requirement and do not receive earnings credits. References Coats, Warren L., Jr. “What Do Reserve Carry overs Mean for Bank Management and for Free Reserves?” Jo u rn a l o f B ank Research, Summer 1976, pp. 123-27. Conference of First Vice Presidents. Subcommit tee on Accounting Systems, Budgets, and Ex penditures, Standard Operating Procedure 10.0. Washington, D.C.: Board of Governors of the Federal Reserve System, 1993Dotsey, Michael. “Monetary Policy and Operat ing Procedures in New Zealand,” Federal Re serve Bank of Richmond, Economic Review, vol. 77, no. 5 (September/October 1991), pp. 13-19. Evanoff, Douglas D. “Reserve Account Manage ment Behavior: Impact of the Reserve Ac counting Scheme and Carry Forward Provision,” Federal Reserve Bank of Chi cago, Working Paper No. 89-12, June 1989. Federal Reserve Bank of Cleveland. “Mainte nance of Reserve and Clearing Accounts,” Operating Letter No. 4, August 27, 1992. Feinman, Joshua. “Bank Reserve Management, Overnight Overdraft Penalties, and Carryover: Theory and Evidence,” Board of Governors of the Federal Reserve System, unpublished manuscript, June 1993Friedman, Richard M., and William W. Roberts. “The Carry-Forward Provision and Manage ment of Bank Reserves "Jo u rn a l o f Finance, vol. 38, no. 3 (June 1983), pp. 845-55. Garfmkel, Michelle R., and Daniel L. Thornton. “Alternative Measures of the Monetary Base: What Are the Differences and Are They Im portant?” Federal Reserve Bank of St. Louis, Review, vol. 73, no. 6 (November/December 1991), pp. 19-35. Hilton, Spence, Ari Cohen, and Ellen Koonmen. “Expanding Clearing Balances,” in Ann-Marie Meulendyke, ed., Reduced Reserve Require ments: Alternativesfo r the Conduct o f Mone tary Policy an d Reserve Management. New York: Federal Reserve Bank of New York, April 1993, pp. 109-35. Poole, William. “Commercial Bank Reserve Man agement in a Stochastic Model: Implications for Monetary Policy,” Journal o f Finance, vol. 27 (December 1968), pp. 769-91Spindt, Paul, and VefaTarhan. “The Liquidity Structure Adjustment Decision of Large Money Center Banks,” Board of Governors of the Federal Reserve System, Special Stud ies Paper No. 121, October 24, 1978. Stevens, E. J. “Removing the Hazard of Fedwire Daylight Overdrafts,” Federal Reserve Bank of Cleveland, Economic Review, vol. 25, no. 2 (1989 Quarter 2), pp. 2-10. ___________ . “Federal Funds Rate Volatility,” Federal Reserve Bank of Cleveland, Eco nom ic Commentary, August 15, 1991a. ___________ . “Is There Any Rationale for Re serve Requirements?” Federal Reserve Bank of Cleveland, Economic Review, vol. 27, no. 3 (1991b Quarter 3), pp. 2-17. ___________ . “Comparing Central Banks’ Rulebooks,” Federal Reserve Bank of Cleveland, Economic Review, vol. 28, no. 3 (1992 Quar ter 3), pp. 2-15. ___________ . “Price Isn’t Everything,” Federal Reserve Bank of Cleveland, Economic Com mentary, April 1, 1993- I B The Consumer Price Index as a Measure of Inflation by Michael F. Bryan and Stephen G. Cecchetti Introduction As the economy approaches the Federal Reserve’s stated objective of price stability, it has become necessary to examine carefully the price indices on which policy is based. The most popularly used aggregate price statistic in the United States is the Consumer Price Index (CPI). This fact alone probably accounts for the prominence it has achieved as a measure of inflation and as a focal point in the Federal Reserve’s inflation fight. As an expenditure-weighted index of cost-of-living changes, though, the CPI was never intended to be used as an indicator of inflation. Broadly speaking, there are two problems as sociated with using the CPI to measure inflation. The first concerns the transitory noise created by nonmonetary events, such as sector-specific shocks and sampling enors. The second involves a potential bias in the index that results both from the expenditure-based weighting scheme the CPI employs (weighting bias) and from persistent er rors in measuring certain prices (measurement bias). In an earlier paper, Bryan and Cecchetti (1993), we study the first of these issues.1 Here, we examine the second. The existence of bias, or deviations between http://fraser.stlouisfed.org/ the trend in the price indices and inflation, im Federal Reserve Bank of St. Louis Michael f. Bryan is an economist at the Federal Reserve Bank of Cleve land, and Stephen G. Cecchetti is a professor of economics at The Ohio State University and a re search associate at the National Bureau of Economic Research. The authors wish to thank Mark Watson for helpful discussions and for providing his software. This paper benefited from the sug gestions of Martin Feldsteln, Dennis Fixler, Spencer Krane, Pok-sang Lam, Nelson Mark, Angelo Melino, Alan Viard, and Mark Wynne. In addition, the authors thank Edward Bryden, Christopher Pike, and Matthew Mercurio for research assistance. plies that any fixed-weight price index will be an imperfect long-run target for a policy aimed at aggregate price stability. The magnitude of the bias in the CPI is an empirical matter. Previous researchers have addressed the issue of bias in price statistics by performing calculations based on highly disaggregated information.2 This ap proach provides at best only a broad approxima tion. Moreover, the bias in price statistics depends on the severity and origin of supply shocks, on changes in technology and tastes, and on other time-varying phenomena, so the time-invariant estimates derived from these saidies are of only limited value to policymakers. Our strategy is different. Using a simple sta tistical framework, we compute a price index that is immune to the weighting bias inherent ■ 1 That paper shows how the use of limited-influence estimators, such as the median of the cross-sectional distribution of individual con sumer goods prices, removes transitory elements that create difficulties with interpreting month-to-month movements in the aggregate CPI. We find that the median CPI performs well as a high-frequency measure of the persistent component of inflation. ■ 2 See Wynne and Sigalla (1993) for a thorough review of the literature. in the CPI as a measure of inflation. The recent work of Stock and Watson (1991) provides a method for combining information in many time series to generate an index of coincident economic conditions. This paper attempts to do for prices what Stock and Watson have done for output. We use a dynamic factor model analogous to theirs to compute the common inflation element in a broad cross-section of consumer price changes. Unlike expenditure or output-weighted price indices, the dynamic factor index is an unbiased estimate of the component common to each of the individual price changes in the crosssection of data we examine. By comparing the trend in the dynamic factor index with the trend in the CPI, we are able to gauge the extent of the weighting bias in the CPI as a measure of inflation. Our results suggest that over the 25year period from 1967 to 1992, the weighting bias in the CPI averaged roughly 0.6 percent age point per year. But, since we can construct a time series for the dynamic factor index, we are able to estimate the bias over two economi cally distinct periods. We find that there was a large positive weighting bias during the 15 years beginning in 1967, but that the weighting bias has been insignificant since 1981.' The following section discusses the sources of bias in fixed-weight price indices. We con tinue with a brief description of the dynamic factor model employed to construct an unbiased measure of consumer price inflation together with its standard error, and then present a sum mary of our results. I. Bias and Expenditure-Weighted Price Indices In order to understand the bias in fixed-weight price indices as measures of inflation, we be gin by defining measured inflation, n t , as a constant expenditure-weighted index of price changes from period t-1 to t, or (1 ) nt =Y.wpPr mon inflation component and an idiosyncratic relative price movement, represented as (2) where mt is inflation and xjt is a relative, or real, price disturbance. Substituting equation (2) into (1), and not ing that = 1, we can write measured infla tion as (3) which states that the growth rate of a standard fixed-weight price statistic sums inflation and a weighted average of relative price disturbances. For purposes of policy formulation, we need to obtain a measure of the common element mt or, alternatively, a measure of n t constructed so that the expectation of the sum on the right side of equation (3) is zero.4 Unfortunately, the expectation of Kt does not equal mt\E( ¿JVp xjt) * 0. There are two rea sons for this “bias.” First, the individual prices may, on average, be measured inconectly. We broadly refer to this as a “measurement bias.” In addition, actual expenditure shares, wjt , and xj{ are conelated, producing a “weighting bias.” In either case, the expectation of the observed xJt’s will be nonzero. Our approach is designed to minimize enors caused by weighting bias. And although the dynamic factor approach we have chosen will have little directly to say about meas urement biases— inasmuch as they are unrelated to the choice of weighting schemes employed— we can make inferences about certain types of these biases by examining subsets of the data. ■ 3 Strictly speaking, the weights used by the Bureau of Labor Statistics (BLS) in the construction of the CPI vary slightly with relative price changes from year to year. This Is necessary in order to hold constant the Implicit real quantity of any item used in the calculation of the index. This fixed-weight price index also differs slightly from the CPI because we are summing the weighted logs of the individual prices rather than the weighted levels. ■ n t = mt + ^ w Joxj t , j j where wJO is a set of base-period expenditure weights and pjt is the percentage change in the price of good j from period t-1 to t? The ex penditure weights are defined to sum to one. The next step is to note that changes in the individual goods prices, the pJt’s, share a com pj{ = mt + xjt, 4 If the xjt's are mean zero and the weights are constant, then E (n { )= mr However, realizations of n , are unlikely to equal m t , and we can also think of nt as a noisy measure of inflation. There are several reasons why realizations of £ wj0 xjt will not equal zero period by pe riod. First, there is simple sampling error in the Individual price data. But in its absence, wj0 xjt may not equal zero period by period because of the way the economy adjusts to real shocks. In our earlier paper, we use a simple model derived from Ball and Mankiw (1992) to describe how supply shocks may cause price indices such as n t in equation (1) to con tain transitory movements away from mt . It should be clear at this point that the bias in a price statistic as a measure of inflation, which is a statistical concept, is distinct from the bias as a measure of the cost of living, al though the two may share similar origins, as we explain shortly. In a strict sense, the choice of the term “bias” may be somewhat unfortu nate here, as it does not reflect an error in the calculation of the CPI per se, but rather an er ror caused by applying the CPI to a problem it was never intended to address. Bias in the CPI as a measure of inflation is simply the devia tion in the trend of n t from mt , whereas bias in the CPI as a measure of the cost o f living is defined as the deviation in the CPI trend from a constant utility price index. Consider the case of substitution bias, in which the price of a single good rises. Label this as good k, so that xkt > 0. In the absence of monetary accommodation, the household budget constraint requires the sum of the rela tive price disturbances weighted by actual ex penditure shares to be zero, or where (3 measures the covariation of actual expenditure weights and relative price distur bances. Substituting equation (5) into (4) yields (6) I> y 0 ^ i+ X P / * / i = ° j j or (6') 2 > y o Xj X = “ X M v i j j This is the weighting bias — only if = 0 will the sum of the base-period weights and the relative price disturbances be zero. Otherwise, a weighting bias will arise that has the oppo site sign of the covariation of the expenditure weights and the relative price disturbance. Nevertheless, there exists a set of weights, wjt, such that (7) E ( ^ w jt xjt) = 0. j (4) 5 > A = °- j For each relative price increase xkt, the relative price of the remaining goods must fall propor tionately such that wktxkt + ^ wjtxjt = 0. Yet, j*k consumer theory implies that expenditure shares will change depending on the price elasticity of demand for the product: Goods having an elastic ity greater than one will experience declines in their relative expenditure, and vice versa. The implication here is that if an actual expenditure weight tends to fall for a product whose relative price rises, it reduces the exactly offsetting relative price influence of the remaining set of commodi ties when applied to their original expenditure weights and creates a positive bias in the infla tion statistic: wkoxkt + Xjt > 0 . j* k Substitution bias is simply a specific form of a general weighting bias. To see this more clearly, consider a simple two-period example. We can represent actual expenditure weights in period 1 as a function of the base period weight and the relative price disturbance in period 1, (5) wjX = wjQ + pyxyl , The Wjt ’s can be thought of as the inflation weights — those that yield a price index with out a weighting bias. II. Origins of Bias in the CPI In general, we think of all of the biases in the CPI as a measure of inflation as arising from some combination of weighting and measurement bias. As we have already described, weighting bias is the consequence of covariation between relative price changes and a set of properly con structed weights. The classic example of such a weighting bias is substitution bias, where the py’s are negative and the weighting bias is positive. Studies of the size of the commodity substi tution bias conducted in recent years have con cluded that the amount of substitution bias in the CPI is relatively small. For example, Manser and McDonald (1988) estimate that the com modity substitution bias averaged between 0.14 and 0.22 percentage point per year over the period 1959 to 1985. This is largely a confirma tion of Braithwait’s (1980) earlier estimate of 0.1 percentage point per year over the 1958 to 1973 period. Moreover, Manser and McDonald find the level of the bias to be one-third greater for the high-inflation period (1972 to 1985) than for the more moderate inflation period of 1959 to 1972. It is entirely conceivable that there are cases in which the correlation between expenditure weights and measured relative price changes is positive, imparting a downward weighting bias in fixed-weight inflation measures. One such case would be a demand-induced relative price increase resulting from a change in tastes, where the relative price of a commodity rises because the relative expenditure on it has risen. Consider also the case in which new goods are introduced. The market basket purchased by households will expand to include items not given any weight in the cunent index or, alternatively, actual expenditure weights on the included goods will fall. As a consequence, price changes for the goods included in the price index are given too much weight relative to a conectly measured price index. If the relative price change for the new good is negative, the new good produces a positive bias in the price index that is analogous to substitution bias. But it is possible to imagine a case in which the relative price change of the new good is positive, resulting in a negative bias in the price statistic. This would hold true if new goods cause a substitution away from, as well as a decrease in the relative price of, the goods in cluded in the index. Similarly, changes in relative product quality produce a weighting bias by introducing a cor relation between actual expenditure weights and relative prices. Quality changes imply that the same effective quantity is available for a generally lower price and, depending on the elasticity of demand for the product, the share of expenditure on such a good could either rise or fall as its effective price drops.5 In many instances, weighting bias is not the sole source of the error from using the CPI as a measure of inflation. A number of potential biases arise when the prices of individual com modities are mismeasured. To see how this af fects the indices we are studying, consider the case in which measured price changes have three components: the common element, mt, the correctly measured relative price change, xJt, and a common, nonzero measurement error, et . We can write this as ■ 5 As an empirical matter, measuring new goods bias is much more difficult than measuring commodity substitution bias, since new goods prices are unobservable prior to their introduction. As noted in Diewert (1987), Hicks (1940) suggests that the price of the new good prior to its in troduction should be the shadow price at which demand is equal to zero. While this is an excellent theoretical criterion, implementation is simply not possible. As a result, little work has been done on estimating the importance of new goods bias. There are, however, several rough estimates of the size of this problem. Diewert (1987) suggests that the bias caused by new goods could be as high as 0.5 to 1.0 percentage point annually, while Lebow, Roberts, and Stockton (1992) gauge the amount as no more than 0.5 per http://fraser.stlouisfed.org/ centage point per year. Federal Reserve Bank of St. Louis (8) pjt = mt + xjt + et. It is readily apparent that the measured price index will be (9) n = m t + et + ^ w j0 xjr j That is, measurement enor will be embedded in the inflation statistic independent of the weight ing scheme. New goods (and other excluded goods more generally) introduce the potential for measurement bias to the extent that the set of prices is no longer complete. Moreover, insofar as average quality changes are reflected in the price data, they also create a measurement bias by producing a common trend in the price data that is unrelated to inflation.6 So-called “outlet substitution bias,” arising from the tendency of consumers to escape some part of price increases by shifting purchases toward lower-priced (discount) stores, is another recently identified source of measurement bias. We can think of this bias as some combination of newgoods bias and quality bias, as the goods sold by the discount retailers might be considered sepa rate commodities from those sold by full-service, higher-priced stores.7 ■ 6 The quality adjustment problem has been the subject of the bulk of academic work on price measurement bias. Beginning with Griliches’ (1961) study of automobile prices, this literature has concentrated on esti mating the quality bias in the prices of specific durable goods, presum ably because the quality of durable goods Is more easily quantifiable and data are usually readily available. Estimates of quality bias in the aggre gate price index are then extrapolated from the measurements derived for specific commodity groups. For example, Gordon (1992) estimates that quality changes account for slightly more than 1.5 percentage points of the average rise in the prices of consumer durable goods over the 1947 to 1983 period. By applying this estimate to goods that they presuppose to be subject to quality improvements, Lebow, Roberts, and Stockton (1992) estimate aggregate quality bias in the CPI to be 0.3 percentage point annually. ■ 7 The recent growth in the discount retail business has led econo mists to increase their concern over outlet substitution bias. When con sumers substitute between retail outlets on the basis of price, and this shift in the buying pattern is not captured in the point-of-purchase survey conducted by the BLS, the CPI overstates inflation. While the Labor De partment adjusts its sample over time, no more than 20 percent of the change in outlet patterns is incorporated into a particular year’s survey. Consequently, this measurement problem can affect the aggregate price statistic for a period of several years. A recent study by Reinsdorf (1993) examines the effect of outlet substitution during the 1980s on food and fuel commodities. Assuming that none of the price differences among outlets reflect quality differentials, he concludes that outlet bias accounts for between 0.25 and 2.0 percentage points annually for food, and between 0.25 and 1.0 percentage point annually for energy. III. A Dynamic Factor Index Approach Our objective is to compute a reduced-bias esti mate of inflation from consumer price data. Re call from equation (3) that we can write a fixed expenditure-weight price index as the sum of common inflation, rht, and a term representing the weighted sum of relative price changes, ^ w J0xJt. This makes clear that the measure ment of inflation requires a set of weights that allow us to construct an estimate of the com m on element in all price changes. Price indices such as the CPI, the Producer Price Index (PPI), or the implicit price deflator for personal con sumption expenditures (PCE) share a common core, but as a result of their weighting method ologies, each has a unique weighting bias as a measure of inflation. As an alternative to the expenditure weighting schemes generally used, we propose weighting commodity prices based on the strength of the inflation signal, mt, relative to the noise, xJt, in each time series. To do this, we assume that the log of each individual product price is the sum of two components: a nonstationary, common core, and a nonstationary, idiosyncratic component measuring movements in relative prices. Taking first differences, the model can be written as Maximum likelihood estimation of mt is ac complished by applying a Kalman filter to a set of either aggregate or individual price data. The result is AanA estimate of both the 1parameter A A vector, a = {'P jB .r}, where T is the diagonal covariance matrix of r\and the common factor, mt . We can write mt as a weighted sum of cur rent and past individual pJt’s. Expressly, (13) which is an unbiased estimate of mt . Put slightly differently, the dynamic factor index is an esti mate of the common trend in the individual inflar \ tion series such that E Wj(L) xjt = 0 . I v J Our main interest is in measuring the average weighting bias in the CPI over various sample pe riods. This is the difference between the average inflation in the CPI and the average mt, which we label rht . We would also like to construct an estimate of the standard enor of this bias. Rewriting (13) in matrix form, we have (14) (11) y (Z )m ,= 5 + ^f, (12) Q(L)xt = P + ri,, where p t and xt are vectors; and 0 are, respectively, a vector and matrix of lag polyno mials with stationary roots; i, and r\are i.i.d. random variables; and P and 8 are vector and scalar constants.8 We identify mt by assuming that relative price disturbances are unconelated with common inflation at all leads and lags. This is what is meant by a common component. If mt were conelated with any of the x ’s, then they would contain a part of the common core. In addition, it is necessary to restrict the (3’s to sum to zero. For computational convenience, we further assume that 0(Z) is a diagonal matrix of lag polynomials, that T), is serially unconelated, and that the covariance matrix of r\t is diagonal.9 http://fraser.stlouisfed.org/ 8 See Stock and Watson (1991 ) for details. Federal Reserve Bank of St. Louis mt = W ( L ) p t . It follows that (1 5 ) (10) p t= m t + x t , mt = ^ W j ( L ) pJt, m = \ r ( l) p /;, A where |i/; is the vector of estimated means of in flation in the individual component price series and 11/(1) is a function of the elements of a .10 It is useful to rewrite the CPI in a way analo gous to (15). From equation (1), we have ^6) n= which is the estimate of average inflation in the CPI constructed as a constant weighted log-linear index. An estimate of the bias follows as (17) BUts=K- rh = W0ÿip - W(l)\ip = [W0 - W ( l ) ] Î p . The construction of standard error estimates is slightly more complicated, but still straightfor ward. To do this, we require an estimate of all ■ 9 Throughout, we assume that both mt and the x jf 'scan be modeled as AR(2)’s. ■ 10 The notation H/(1) represents the evaluation of the lag polyno mials at L = 1, and so is the sum of the polynomial coefficients. F I G U R E 1 Comparison of the CPI, PCE Deflator, and DF2 SOURCES: U.S. Department o f Labor, Bureau o f Labor Statistics; U.S. Depart ment o f Commerce, Bureau o f Economic Analysis; and authors’ calculation« TABLE 1 Comparisons of the CPI and the CPI/PCE Dynamic Factor Index (annualized percent changes) of the parameters used to calculate m and n. This includes the estimated covariance matrix of a as well as an estimate of the covariance matrix of the vector of estimated means The first of these is a by-product of the maximum likelihood estimation of a , while the second can be con structed from the raw inflation data. Calculation of the covariance matrix of |x is complicated by the fact that the p f s have sub stantial serial conelation. In fact, the model (10)(12) implies that when 'F(Z) and the 0 (Z)’s are all second-order polynomials, the individual infla tion series will follow an ARMA(4,2).n This leads us to use the Newey and West (1987) heteroskedasticity and autoconelation consistent co variance estimator, with 24 lags. We can now construct an estimate of the co variance matrix of the entire parameter vector y = {a , jiip} , called jr. Assuming that a and |i^ are independent, then X is block diagonal. Be cause m and n are both functions of y, we can construct standard errors by computing the vector of first partial derivatives of each with re spect to y. The variance estimates follow by pre- and post-multiplying X by this vector of derivatives. A It is worth noting that the uncertainty in Bias comes from variation in W {\), which is a func tion of a , and variation in . But the uncer tainty in the mean vector creates variation in the estimation of mean CPI inflation as well, and so the variance in the estimated bias is likely to be lower than the variance in either m or 71.12 Feb. 1967Dec. 1981 Jan. 1982Dec. 1992 Full Sample CPI all items 7.05 (0.94) 3.75 (0.33) 5.65 (0.71) IV. The Results PCE deflator 6.36 (0.71) 4.05 (0.26) 5.38 (0.53) DF2 6.65 (0.81) 3.75 (0.33) 5.48 (0.59) Weighting bias 0.39 (0.23) 0.00 (0.00) 0.17 (0.15) We constructed two alternative dynamic factor indices of inflation based on consumer price data from 1967 to 1992. The first, labeled DF2, is the common element derived from the CPI and the PCE deflator, two aggregate consumer NOTE: Numbers in parentheses are standard errors. The covariance matrix of the means o f the two components was com puted using a Newey and West (1987) robust covariance estimator with 24 lags. Subperiod calculations were made independently from the full sample. All values are the average an nual difference in the natural log of the index. SOURCE: Authors. ■ 11 It is simple to show that the model implies that, ignoring con stants, each individual inflation series can be written as 0¡(L) 4* (L) pjt = 0y(/.)£f+¥ (L) r\ ]t, which is a restricted ARMA (4,2). ■ 12 As implied by the discussion at the end of the previous section, the block diagonality of the covariance matrix allows us to measure the relative contribution of variation in the model parameters, the elements of â , and the mean vector, j i to the estimated variance of the bias. In virtu ally all of the cases we examine, the uncertainty from estimation of the means accounts for more than 95 percent of the uncertainty in Bias. F I G U R E 2 12-Month Growth Rates of the CPI, DF2, and DF36 SOURCES: U.S. Department o f Labor, Bureau o f Labor Statistics; U.S. Depart ment o f Commerce, Bureau o f Economic Analysis; and authors’ calculations. Comparisons of the CPI and the 36-Component Dynamic Factor Index (annualized percent changes) Feb.1967Dec. 1981 Jan. 1982Dec. 1992 Full Sample CPIa 6.93 (0.85) 4.04 (0.26) 5.71 (0.63) DF36 6.05 (0.68) 4.11 (0.25) 5.11 (0.52) Weighting bias 0.88 (0.26) -0.07 (0.13) 0.60 (0.17) a. The CPI used here was constructed as the weighted sum o f the difference o f the natural logs o f the individual components (1985 weights). NOTE: Numbers in parentheses are standard errors. The covariance matrix o f the means o f the 36 components was com puted using a Newey and West (1987) robust covariance estimator with 24 lags. Subperiod calculations were made independently from the full sample. All values are the average annual difference in the natural log o f the index. SOURCE: Authors. price statistics that are constructed from essen tially the same price data, but that employ dif ferent weighting schemes (figure 1). Over the full sample, this dynamic factor index averaged 5.48 percent per year with a standard error of 0.59 percentage point. This yields a weighting bias over the period of 0.17 percentage point with a standard error of 0.15 percentage point (table 1). Subperiod estimates, which are com puted separately using data for only the sub samples, reveal more bias in the 1967 to 1981 interval, about 0.4 percentage point annually. Over the latter period, there appears to have been no bias in the CPI. The aggregate CPI and the PCE deflator may not provide a rich enough set of price data to measure the common element accu rately. As an alternative, we calculated the dy namic factor index from disaggregated price data for 36 components of the CPI (DF36), spanning the complete set of the consumer market basket over the same January 1967 to December 1992 period.13 The 12-month growth rates of the CPI, DF2, and DF36 are re produced in figure 2. The average rate of increase of this more com prehensive dynamic factor index over the sample period is 5.11 percent, compared with 5.71 per cent for the CPI, implying an average annual bias in the CPI of 0.60 percentage point over the 1967 to 1992 period with a standard enor of 0.17 per centage point (table 2). Using 36 rather than two indices increases the estimated weighting bias with virtually no change in precision. But again, we find substantial differences in the magnitude of the CPI weighting bias between the two sub periods. Between 1967 and 1981, we estimate the weighting bias at 0.88 percentage point annually (with a standard enor of 0.26). But since 1981, we fix the bias in the CPI to be nearly zero (-0.07 percentage point with a standard enor of 0.13 percentage point). The dynamic factor indices have limitations, of course. First, the degree of disaggregation and the extent of the sample covered by the price data used are incomplete. More generally, our calculations do not account for the poten tially important measurement biases that arise when goods are systematically excluded or when there is a common measurement error, such as unmeasured aggregate quality changes. While we cannot address such measurement biases directly, we can gauge their severity by ■ 13 A catalog of the 36 components can be found in Bryan and Cecchetti (1993). Comparisons of the Dynamic Factor Indices of Goods and Services Prices (annualized percent changes) Feb. 1967Dec. 1981 Jan. 1982Dec. 1992 Full Sample CPIa 6.93 (0.85) 4.04 (0.26) 5.71 (0.63) DF36 6.05 (0.68) 4.11 (0.25) 5.11 (0.52) DFGOODS 5.43 (0.69) 3.55 (0.30) 4.47 (0.54) DFSERVICES 7.06 (0.70) 4.90 (0.27) (0.53) 6.02 Estimated Bias CPI-DF36 0.88 (0.26) CPI-DFGOODS 1.50 (0.30) -0.07 (0.13) 0.60 (0.17) 0.49 (0.15) 1.23 (0.20) a. The CPI used here was constructed as the weighted sum o f the difference o f the natural logs o f the individual components (1985 weights). NOTE: Numbers in parentheses are standard errors. The covariance matrix of the means o f the 36 components was com puted using a Newey and West (1987) robust covariance estimator w ith 24 lags. Subperiod calculations were m ade independently from the full sample. All values are the average annual difference in the natural log o f the index. SOURCE: Authors. comparing dynamic factor indices computed from commodity subsets of the data.14 In our statistical model, equations (10) to (12), relative price changes are taken to be stationary. With the additional assumption that relative price changes are zero on average (that is, that the P’s in equation [12] are all zero), we can estimate the common factor from any subset of the data. Some economists have suggested that the most serious problem may be in measuring service output. This means that services prices are un reliable, and we use that insight to examine the size of this potential measurement bias.15 14 Measurement bias might manifest itself as low-frequency com ponents in the X jt ’s of certain series. The implication is that the single factor model we employ may not be sufficiently general to capture the time-series behavior of some prices. If this were a serious problem, then we should find that some of the roots of the estimated AR(2) coefficients in 0 (L) imply nearly nonstationary behavior. Our estimates suggest that this may be a problem for medical commodities, motor fuel, and transpor tation services, but is unlikely to affect the commodities generally thought to suffer from significant measurement difficulties. To test the hypothesis that there is a systematic bias in the measurement of services prices, and to evaluate the recommendation that these prices be excluded from the calculation of inflation, we have split the CPI into goods and services compo nents and have computed a dynamic factor in dex for each. The results are reported in table 3. Assuming that the difference between infla tion in goods prices and inflation in services prices is entirely a result of measurement bias in the latter category, we can gauge the weighting bias in the CPI from the difference between the dynamic factor index estimated using goods only (DFGOODS) and the aggregate CPI. Again, while we note rather substantial differences be tween the two prior to 1982, for the recent pe riod, we estimate the weighting bias in the CPI at less than 0.5 percentage point per year. These results also allow us to estimate the size of the measurement bias in services prices di rectly by comparing the dynamic factor indices for goods only (DFGOODS) and services only (DFSERVICES). Curiously, the deviation between the dynamic factor indices calculated from the component data, while relatively large for the 1967 to 1981 period (1.63 percentage points an nually), is slightly smaller in the post-1981 period (1.35 percentage points annually). While there appears to have been a systematic bias in serv ices prices before 1982, which may be attribut able to their mismeasurement, that difference was reduced after 1981.16 V. Conclusion Gauging the accuracy of price indices, which has a long tradition in economics, has taken on new enthusiasm in the recent era of rela tively moderate inflation. At issue is whether a goal of zero inflation literally means zero or whether, because of various biases in the calcu lation of inflation, some low but nonzero rate of measured inflation is sufficient. We have computed dynamic factor indices of consumer prices, which are constructed by essentially weighting commodities on the strength ■ http://fraser.stlouisfed.org/ ■ 15 A recent example is in Poole (1992). Federal Reserve Bank of St. Louis ■ 16 In the early 1980s, the methodology used to construct the shel ter component of the CPI, which accounts for roughly half of all services in the index, was changed from a relatively volatile purchase-price basis to a rental equivalence basis. To account for this change, we reconstruct ed the shelter component to conform to a rental equivalence basis for the entire sample. This change, not surprisingly, had little impact on the dy namic factor index calculations. Nevertheless, the results reported here are on the adjusted basis. of a common inflation signal, in an attempt to assess a potentially important source of bias in the CPI as a measure of inflation— weighting bias. Our estimate of weighting bias in the CPI is roughly 0.6 percent annually in the 1967 to 1992 period, but the size of that bias varies sub stantially within subperiods. In fact, on the ba sis of the estimates provided here, we conclude that since 1981, weighting bias in the CPI as a measure of inflation has been negligible. If there is measurement bias common to the consumer prices in our data set, such as may occur from the systematic mismeasurement of quality changes, it would still be embedded in the estimates presented here. We found signifi cant differences between the dynamic factor es timates derived from all items and the dynamic factor indices derived from goods prices only. In this paper, we have considered only the case of consumer prices, given their impor tance in the monetary policy setting and also allowing for comparisons with other studies of bias. Conceivably, a measurement bias com mon to all consumer prices caused by, say, a reallocation of the economy’s resources be tween investment and consumption goods may be embedded in the dynamic factor indi ces presented here.17 This could presumably be conected by allowing the dynamic factor index to include a broader range of prices, particularly asset prices. An area of future research, then, would involve the integration of investment goods into these dynamic factor calculations. References Alchian, Armen A., and Benjamin Klein. “O n a Correct Measure of Inflation,” Jo u rn a l o f Money, Credit, a n d Banking, vol. 5, no. 1 (February 1973), pp. 173-91. Ball, Laurence M., and N. Gregory Mankiw. “Relative Price Changes as Aggregate Sup ply Shocks,” National Bureau of Economic Research Working Paper No. 4168, Septem ber 1992. Braithwait, Steven D. “The Substitution Bias of the Laspeyres Price Index: An Analysis Us ing Estimated Cost-of-Living Indexes,” Am erican Economic Review, vol. 70, no. 1 (March 1980), pp. 64-77. Bryan, Michael F., and Stephen G. Cecchetti. “Measuring Core Inflation,” National Bureau of Economic Research Working Paper No. 4303, March 1993. Diewert, W.E. “Index Numbers,” in John Eatwell, Murray Milgate, and Peter Newman, eds., The New Palgrave D ictionary o f Eco nomics. London: Macmillan Press, 1987, pp. 767-80. Gordon, Robert J. “Measuring the Aggregate Price Level: Implications for Economic Per formance and Policy,” National Bureau of Economic Research Working Paper No. 3969, January 1992. Griliches, Zvi. “Hedonic Price Indexes for Auto mobiles: An Econometric Analysis of Quality Change,” in Price Statistics o f the Federal Gov ernment. New York: National Bureau of Eco nomic Research, 1961, pp. 173-96. Hicks, John R. “The Valuation of Social Income,” Economica, vol. 7 (1940), pp. 105-24. Lebow, David E., John M. Roberts, and David J. Stockton. “Economic Performance under Price Stability,” Board of Governors of the Federal Reserve System, Working Paper No. 125, April 1992. ■ 17 The potential for a systematic measurement bias, caused by the exclusion of investment goods in the CPI, has been suggested by Alchian http://fraser.stlouisfed.org/ and Klein (1973). Federal Reserve Bank of St. Louis Manser, Marilyn E., and Richard J. McDonald. “An Analysis of Substitution Bias in Measur ing Inflation, 1959-1985,” Econometrica, vol. 56, no. 4 (July 1988), pp. 909-30. Newey, Whitney K., and Kenneth D. West. “A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Co variance Matrix,” Econometrica, vol. 55, no. 3 (May 1987), pp. 703-08. Poole, William. “Where Do We Stand in the Battle against Inflation?” Report to the Shadow Open Market Committee, March 8-9, 1992. Reinsdorf, Marshall. "The Effect of Outlet Price Differentials on the U.S. Consumer Price In dex,” in Murray F. Foss, Marilyn E. Manser, and Allan H. Young, eds., Price Measure ments a n d Their Uses. Chicago: University of Chicago Press for the National Bureau of Economic Research, 1993, pp. 227-54. Stock, James H., and Mark W. Watson. “A Prob ability Model of the Coincident Economic Indicators,” in Kajal Lahiri and Geoffrey H. Moore, eds., Leading Economic Indicators: New Approaches a n d Forecasting Records. Cambridge: Cambridge University Press, 1991, pp. 63-89. Wynne, Mark A., and Fiona Sigalla. “A Survey of Measurement Biases in Price Indexes,” Fed eral Reserve Bank of Dallas, Research Paper No. 9340, October 1993- The Inaccuracy of Newspaper Reports of U.S. Foreign Exchange Intervention by William P. Osterberg and Rebecca Wetmore Humes Introduction Central bank intervention in foreign exchange markets most recently came into prominence dur ing the period of exchange-rate volatility in the autumn of 1992. Speculators doubted that Euro pean central banks would be able to defend the exchange rates agreed upon as part of the Euro pean Rate Mechanism. After massive intervention, central banks eventually capitulated, and several key exchange rates were allowed to fall radically against the German mark (DM). While this se quence of events would seem to have cast con siderable doubt on the usefulness of sterilized intervention, disagreement continues both within policy circles and among researchers as to whether sterilized central bank intervention is a useful tool for exchange-rate management.1 Until recently, studies of intervention have been hampered by a lack of official data, as direct measures of central bank intervention ■ 1 Researchers would point out that this most recent period was not a good test of intervention's efficacy because exchange-rate management was not the sole objective of the central banks. In addition, some of the intervention may not have been sterilized, making it difficult to isolate its impact. “Sterilization” occurs when the effect of intervention on the money supply is offset by open market operations. Nonsterilized intervention is http://fraser.stlouisfed.org/ thus, in some sense, equivalent to monetary policy. Federal Reserve Bank of St. Louis William P. Osterberg is an econo mist and Rebecca Wetmore Humes is an economic analyst at the Fed eral Reserve Bank of Cleveland. The authors thank Owen Humpage and Mark Sniderman for helpful comments. have usually not been made available to the public. Now, however, the Board of Governors of the Federal Reserve System provides a time series of U.S. dollar intervention vis-à-vis the DM and the Japanese yen from 1985 to 1992. One consequence of the former lack of imme diately available and accurate intervention in formation has been the use of daily newspaper reports as proxies for actual intervention mag nitudes in related studies. The possibility that intervention is not re ported accurately may have important implica tions for understanding the signaling mechanism of intervention. For example, such inaccuracy may call into question the ability of intervention to signal future monetary policy with precision. In addition, it may reflect differences in the infor mation available to foreign exchange traders, sug gesting that some traders may be able to profit from inside information. In this paper, we begin with a discussion of issues regarding information about intervention. We then describe the data on actual intervention and newspaper reports. In the third section, we outline the procedure that we use to test for sys tematic differences between reported and actual in tervention series. In the final section, we briefly discuss the implications of our results. I. Information about Intervention: Reported versus Actual Data There is by now a substantial literature de voted to understanding the impact of central bank intervention on foreign exchange mar kets. Recent useful summaries of this literature have been provided by Dominguez and Frankel (1993), Edison (1993), Humpage (1991), and Obstfeld (1990).2 While most recent studies, such as Baillie and Humpage (1992), Baillie and Osterberg (1993), and Hung (1992), use official daily intervention data, others, such as Klein and Rosengren (1991) and Kaminsky and Lewis (1993), use daily news paper reports of intervention.3 If the focus of a given study is on the signaling role of interven tion. then it makes sense to utilize newspaper re ports that reflect the information available to the average trader. One concern is that the choice of interven tion data, reported or actual, may influence re searchers’ conclusions about the efficacy of intervention. However, we would like to raise two other possible concerns, namely, that if there is a systematic difference between actual and reported intervention, 1) the signals as rep resented by the newspaper reports may be mis leading, and 2) some market participants may have more accurate information about interven tion than do others. That the latter is possible can be seen simply by considering the mecha nisms of intervention. U.S. intervention counter parties are either brokers or commercial banks. If brokers are utilized, they will not reveal that the transaction is official intervention. If com mercial banks are utilized, the wire services should accurately reveal that the Federal Re serve has entered the market.4 In either case, the only market participants with definitive knowledge are the counterparties chosen by the Federal Reserve Bank of New York. If we are willing to assume that the newspaper reports indicate what is known about interven tion by the uninformed trader, then a systematic difference between actual and reported interven tion implies a systematic difference in knowledge among market participants. However, it is not clear how much time passes before all market participants learn of the intervention, or even if they ever obtain accurate information short of the official release one year later by U.S. authori ties. In addition, it is unclear if the newspaper re ports are written during the course of the day and are thus affected by changing and uncertain views about intervention activity, or whether they represent a presumably more accurate, end-of-day assessment. We know of only three previous comparisons of actual and reported U.S. intervention data. Klein (1993) uses multinomial logit analysis to calculate the probability that intervention is re ported, conditional on the size of the interven tion. He estimates that, without conditioning on size, the probability that actual intervention is reported is 72 percent, and the probability that reported intervention actually occurred is 88 percent. He also shows that newspaper reports are more likely if the intervention is relatively large. Dominguez (1992) examines the impacts of reported and “secret” intervention on the volatility of the DM/U.S. dollar exchange rate. She assumes that actual intervention not re ported in the newspapers is “secret.” No signifi cant difference is seen between the impacts of the two categories of intervention on volatility. Dominguez and Frankel (1993) tabulate actual and reported interventions by the United States and Germany from November 1982 through October 1989. The accuracy of newspaper re ports varied across different time periods. For example, while all 22 U.S. interventions in the period September through November 1985 were reported, only 73 percent of interven tions from March 1989 through October 1989 appeared in the print media.5 We make two contributions to the literature on central bank intervention. First, we construct a comprehensive data set from newspaper re ports of central bank intervention for the period January 2, 1985 to October 11, 1991. This data set improves on those constructed by other research ers by quantifying qualitative reports (such as “small” intervention) rather than disregarding them. Second, we test for the existence of sys tematic components in the differences between ■ 2 A consensus of the literature is that if sterilized intervention mat ters at all, it does so because it signals a change in information about monetary policy. ■ 3 Still others have 1) constructed monthly numbers intended to capture the shift in international portfolios due to intervention (for exam ple, Ghosh [1992]), 2) attempted to define intervention in terms of the monetary authorities’ balance sheets (see Danker et al. [1987]), or 3) used measures of central banks’ foreign reserves (for example, Glick and Hutchison [1992] and Watanabe [1992]). ■ 4 However, the market sometimes seems to make guesses that con fuse intervention operations with correspondent transactions. ■ 5 Dominguez (1992) and Dominguez and Frankel (1993) utilize re ports of intervention from The London Financial Times, The New York Times, and The Wall Street Journal. Klein (1993) uses the first two sources. 27 actual and reported intervention, calculating these differences using either dummy variables or numerical magnitudes. We also either in clude “rumors” in the reported series or dis card them. With few exceptions, we find that there are systematic components; that is, the differences are serially conelated. II. Data Actual The Board of Governors of the Federal Reserve System provided us with time series of U.S. net daily dollar transactions from January 1985 to Oc tober 1991. All data are in dollars, representing the actual net dollar purchases (sales) rather than dollar equivalents that have been translated into dollars via application of the exchange rate.6 These data are now publicly available, with a one-year lag, from the Board of Governors. We report the results of our analyses with three categories of intervention: U.S. intervention visà-vis unspecified cunencies carried out in terms of U.S. dollars, U.S. intervention vis-a-vis the DM, and U.S. intervention vis-a-vis the yen. Using these data, we created dummy vari ables, each of which equals +1 for positive net dollar purchases, -1 for negative net dollar pur chases (positive sales), and 0 if the country did not intervene (its net dollar transaction was 0). Newspaper Reports After having searched The Wall StreetJournal, The Neu> York Times, and The Financial Times, we ultimately decided to record the daily press reports of intervention from the foreign exchange column of The Wall StreetJournal.7 We recorded all mentions of intervention that were indicated as pertaining to the previous day or previous business day. Thus, if there was first mention of intervention a week after its occunence, we do not record it, on the presumption that it would not have been known by the market at the time. As in the case with the actual data, for each category of intervention, a buy/sell variable was created to indicate whether a country was a net buyer or seller of dollars. It equals +1 if the country bought dollars, -1 if it sold dollars, and 0 if it did not intervene. To correspond to the way in which the actual intervention data were constructed, we documented U.S. interven tion in the DM/dollar and yen/dollar markets. A buy/sell variable was constructed for each mar ket, indicating whether the United States bought (+1) or sold (-1) dollars. Thus, reported U.S. in tervention in each of these two markets is re corded in two places. For example, if the United States was reported to be buying yen, we would record this under the United States selling dol lars vis-a-vis the yen, and also in the (overall) U.S. selling category described previously (and denoted as U.S. vs. $U.S. in the tables). For all groups, we recorded the size of the interven tion if given. This includes qualitative terms such as small, moderate, and large, as well as dollar magnitudes when given. After all data were recorded, we calculated the minimum, median, and maximum of the re ported dollar magnitudes for each U.S. inter vention variable when such magnitudes were reported in the newspaper. We substituted for qualitative terms. For terms indicating “small,” “light,” or “token,” we used the minimum for the particular category of intervention. For “modest” or “moderate,” we substituted the me dian. For “large” or “heavy,” we substituted the maximum. If no indication of size was given, we used the median. For example, if the United States was reported to be intervening heavily against the yen, we would substitute the maximum of all numeric reports of the United States buying or selling dollars against the yen. We then created a net transaction vari able for each category by multiplying the buy/sell dummy variable by that amount. This variable is comparable to the actual net inter vention variable. The minimums, maximums, and medians for all of the reported intervention variables are provided in table 1. We also recorded specific mention of rumors.8 For a given country A, two types of rumors are recorded: 1) whether country A intervened on its own behalf, and 2) whether country A inter vened on behalf of country B (or whether coun try B intervened on behalf of country A). In the white noise tests that we describe below, we either disregard the rumors (treat them as being nonreports) or count them (treat them the same as other reports). The details of our treatment of rumors and “on behalf o f” transactions are de scribed in the appendix. ■ 6 Such a procedure would embed simultaneity into any sub sequent analysis of the relation between intervention and exchange rates. ■ 7 This source is the most consistent of the three. While the use of only one source may seem to make our series less comprehensive than it would otherwise be, the amount of information that we obtain from this news report is greater. In addition, we avoid having to determine how to code reports when disparities arise among different sources. ■ 8 Thus, an erroneous report is not the same as an erroneous rumor. y T A B L E 1 III. White Noise Tests M in im u m , M e d ia n , and M a xim u m for the S ize of Reported Intervention (m illio ns of U .S . dollars) Minimum U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Median Maximum 35.00 150.00 600.00 60.00 1 1 8 .3 3 1 4 3 .7 5 250.00 50.00 200.00 SOURCE: Authors’ calculations based on newspaper reports from January 2, 1985 to December 31, 1991. B 0 X 1 Calculations of the W hite N oise Te st Statistics The two test statistics utilized in this article are those calcu lated by the SAS/ETS routine SPECTRA. They are the Kappa (K ) statistic suggested by Fisher (1929) and the KolmogorovSmirnov (K-S) statistics suggested by Bartlett (1966). Fuller (1976) presents their formulas as follows: k= 1 K - S = maximum absolute difference of Ck , the cumulative distribution function of a uniform random variable, where m k C*= [ X j= 1 X 4<<V' ./= 1 In(L) is the largest periodogram of a sample of m periodogram ordinates with two degrees of freedom. Here, co indicates fre quency, with m - ( n —l)/2 and n being the number of observations. In both K and K -S, the periodogram is being used to search for periodicities of unspecified form.3 Fuller (1976), p. 282, states that “for many nonnormal processes we may treat the periodogram ordinates as multiples of chi-squared random variables.” He further discusses how this assumption helps to motivate the formulas given above. However, as we note in the text, the peculiar nature of the data here requires us to qualify our application of these test statistics to our data and to con sider alternate sample periods and alternate calculations of the series. Fuller (1976), p. 284, gives the distribution of K, and Bimbaum (1952) gives the distribution of K - S. a. The concept of a periodogram is detailed in Fuller (1976), p. 275. The white noise tests focus on the U.S. inter vention categories. For each reported intervention variable, we vary the series along two dimen sions: First, we either count all rumors (about whether there was intervention or rumored “on behalf o f” intervention) or discount all rumors.9 Second, we use either numerical values or dum my variables. The use of dummy variables may help to ameliorate some problems discussed below regarding the appropriate use of our statis tical technique. Although we could see if errors in reports of intervention were of economic significance by comparing the impacts of actual and re ported intervention on exchange rates, that pro cedure would require us to specify a model of the interaction between intervention and ex change rates. Given the multiplicity of frame works used to study intervention, we elected to utilize a technique that is not model-specific: testing for whether the differences between re ported and actual intervention are white noise. A time series is white noise if it has a mean value equal to zero and if observations are seri ally uncorrelated. The two statistics we report below are those of the Kolmogorov-Smirnov and Kappa tests, provided by the SAS/ETS (1990) version 6 rou tine SPECTRA. A detailed discussion of these tests is found in Fuller (1976), pp. 282-85. The exact calculations are described in box 1. In our application of the tests, a finding that a se ries is not white noise implies that the series contains serial correlation rather than that it lacks a nonzero average.10 However, there are some limitations as to how one can interpret these test results. First, the interpretation of the enor equaling zero is ambiguous because it does so whenever 1) there was no intervention and no intervention was reported, and 2) there was intervention that was reported accurately. That this ambiguity is not a desirable characteristic of our procedure can be seen by comparing three scenarios. In one ■ 9 As an example, consider a report that “the Federal Reserve pur chased 100 million yen, rumored to be on behalf of the Bank of Japan." In a series that counts rumors, this would be entered as a purchase of yen (sale of dollars) by the Japanese, while in the “no rumors'' series, it would count as a U.S. purchase of yen. ■ 10 Utilizing the ADJMEAN option in the SPECTRA routine sets the average of the series to equal zero. 29 TABLE 2 A Descriptive Statistics for Actual, Reported, and Rumored Intervention: Full Sample Period N u m b e r of Occurrences Average Size Total Buying Selling Buying Selling 294 203 185 98 61 66 196 142 119 160.34 111.83 134.73 177.74 141.21 124.25 184 38 37 52 6 12 132 32 25 148.08 140.28 131.25 148.56 108.64 137.25 38 4 3 16 22 3 2 142.81 118.33 143.75 140.68 98.89 143.75 Actual intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Errors in Reported Intervention Total 1 1 Actual but Reported but Not Reported Not Actual Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen 160 171 158 135 168 153 25 3 5 R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen 24 4 2 Categories of intervention: U.S. vs. $U.S.: U.S. intervention vis-à-vis unspecified currencies, carried out in terms o f U.S. dollars. U.S. vs. DM: U.S. purchases or sales o f DM in terms o f U.S. dollars. U.S. vs. Yen: U.S. purchases or sales of yen in terms of U.S. dollars. NOTE: “Buying” and “Selling” columns are in terms o f purchases and sales of millions o f U.S. dollars. SOURCE: Authors’ calculations. case, imagine a typical day in the midst of a long period in which there was no interven tion and no reason to expect intervention. In the second case, imagine that the newspapers cor rectly report the cessation of intervention at the end of a period of turbulent markets and frequent intervention. In the third case, assume that a non zero amount of intervention is conectly reported. In all three cases, the error is zero, although differ ent information is provided in each case.11 We hope to ameliorate the impact of this factor on our result by varying the data in two ways. First, we split the sample in half to con trol in part for changes in the frequency and patterns of intervention. Second, we calculate the enors using both dummy variables and numeric variables. Using dummy variables will reduce the number of enors if the newspapers seldom cor rectly report the amount of intervention. Another limitation to our procedure is that our data may violate the maintained hypothesis that they are generated by a continuous random variable. Intervention either takes the value of zero (the vast majority of days) or jumps to a number of the magnitude of 100 (100 million U.S. dollars). Here again, we hope that by using dummy variables, which exhibit smaller jumps, we reduce the impact of such discontinuities. IV. Intervention Data and Errors Tables 2A-2C describe the actual intervention data, the reports of intervention, and rumored interventions.12 The first line, “U.S. vs. $U.S.,” de notes U.S. purchases or sales of unspecified cur rencies. This includes the number of days that the United States intervened in all cunencies, in cluding the DM and yen, as indicated on the next two lines.13 We use this measure in our as sessment of the overall accuracy of reports about U.S. intervention, since newspaper reports often do not specify the foreign cunency in which the United States is intervening.14 In table 2A, we see that there were 294 actual U.S. interventions for the full sample period, 184 reports of intervention, and 38 rumors of inter vention. Thus, at most, 76 percent of interven tions were mentioned in the newspaper ([184 + 38] 7294 = 0.76). At the bottom of the table, we ■ 11 This problem would be ameliorated if we were able to model the joint process governing the intervention/exchange-rate interaction. This process presumably will yield an expected intervention variable and in turn will specify the significance of errors in reported intervention on the exchange rate. ■ 12 We compiled many more categories of reports than are ana lyzed in the tables. Our comparisons were restricted to those series for which we had actual intervention data. ■ 13 Note that the United States sometimes intervened with more than one currency within one day. ■ 14 The official data are In dollars, so in our comparison of re ported and actual intervention, we have restricted ourselves to reports of dollar intervention. Fortunately, when reports specify amounts, they indi cate the dollar magnitudes, eliminating the need to convert via applica tion of the exchange rate. TABLE 2 B Descriptive Statistics for Actual, Reported, and Rumored Intervention: January 2 , 1985-May 20,1988 N u m b e r of Occurrences Average Size Total Buying Selling Buying Selling 100 60 78 61 33 53 39 27 25 176.68 116.83 130.60 119.99 115.41 62.53 55 4 8 33 2 8 22 2 0 153.18 118.33 132.03 138.18 109.17 0 20 2 0 12 0 0 8 2 0 140.42 0 0 124.38 89.17 0 Actual intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Errors in Reported Intervention Total Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen 65 56 72 Actual but Reported but Not Reported Not Actual 55 56 71 10 0 1 R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs Yen 13 2 0 Tables 2B and 2C present similar information for the two sample halves.15 Almost twice as much actual intervention in the U.S. vs. $U.S. category occurred in the second half of the sample as in the first. In the U.S. vs. DM cate gory, intervention was much heavier in the sec-, ond half of the sample, as the United States shifted to buying DM (selling dollars). Reports of intervention appear to capture these pat terns. However, in the U.S. vs. DM and U.S. vs. yen categories, occurrences of reports fall far short of the number of actual interventions. This finding stands in sharp contrast to the find ings in the previous paragraph regarding the U.S. vs. $U.S. category. Table 3 presents the results of the white noise tests, separated by whether the reported series omits or includes rumors and by whether we use numeric or dummy variables.16 All of the white noise tests were performed on both the full sam ple and on each half of the sample. Splitting the sample is an attempt to see if the results are sen sitive to choosing sample periods that vary re garding either the intensity of intervention or its pattern. In this case, intervention activity was heavier during the second subsample. Generally, with both tests, the full sample and split samples reject the hypothesis that the time series of enors are white noise. Thus, there are systematic components to the differences be tween actual and reported intervention. For dum my variables, we reject the hypothesis of white noise in all cases. Categories of intervention: U.S. vs. $U.S.: U.S. intervention vis-à-vis unspecified currencies, carried out in terms o f U.S. dollars. U.S. vs. DM: U.S. purchases or sales o f DM in terms o f U.S. dollars. U.S. vs. Yen: U.S. purchases or sales o f yen in terms o f U.S. dollars. NOTE: “Buying” and “Selling” columns are in terms o f purchases and sales o f millions o f U.S. dollars. SOURCE: Authors’ calculations. report errors, either from comparing actual and reported or from comparing actual and ru mored intervention. O n the one hand, there were 135 days on which intervention occuned but was not reported, implying that it was re ported only 54 percent of the time. On the other hand, only 25 of the 184 reports were er roneous (86 percent accuracy). In the case of rumors, however, most were in enor: For 24 of 38 rumors, there was no actual intervention. V. Summary Newspaper reports of central bank intervention are often used as if they are interchangeable with actual intervention data. Except in rare cases, ac tual data have become available only recently for the United States, with a one-year lag. Here we describe detailed time series culled from The Wall StreetJournal and compare them to actual inter vention data. We quantify qualitative reports of intervention for all of the series. To the best of ■ 15 We have also compiled analogous tables for the subperiods January 2—December 31,1985; January 1,1986—February 20,1987; February 21,1987—February 19,1990; and February 20,1990—October 11,1991. These tables are available from the authors and facilitate com parison with previous research on the effectiveness of intervention over various subsamples. ■ 16 Rumored intervention includes rumors about both “own" and “on behalf of” intervention. 31 TABLE 2 C Descriptive Statistics for Actual, Reported, and Rumored Intervention: May 23,1988—October 11,1991 N u m b e r of Occurrences Average Size Total Buying Selling Buying 194 143 107 37 28 13 157 115 94 133-41 105.93 151.54 192.09 147.27 140.66 129 34 29 19 4 4 110 30 25 139.21 151.25 129.69 150.64 108.61 137.25 18 2 3 4 1 1 14 1 2 150.00 118.33 143.75 150.00 118.33 143.75 Selling Actual intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen Errors in Reported Intervention Total Actual but Reported but Not Reported Not Actual Reported intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen 95 115 86 80 112 82 15 3 4 R u m o r e d intervention U.S. vs. $U.S. U.S. vs. DM U.S. vs. Yen 11 2 2 Categories of intervention: U.S. vs. $U.S.: U.S. intervention vis-à-vis unspecified currencies, carried out in terms o f U.S. dollars. U.S. vs. DM: U.S. purchases or sales of DM in terms o f U.S. dollars. U.S. vs. Yen: U.S. purchases or sales o f yen in terms of U.S. dollars. NOTE: “Buying” and “Selling” columns are in terms o f purchases and sales of millions of U.S. dollars. SOURCE: Authors’ calculations. our knowledge, this is the first such treatment of qualitative reports. Whether we examine numeric values or dum my variables, count or discount rumors, or split the sample, we find that there usually are system atic components in the differences between the actual and reported intervention series. While the economic significance of any such differences is unclear, we believe that these findings may have important implications for understanding the sig naling mechanism of intervention. If the newspa per reports reflect the markets’ final assessment of intervention activity, then reporting enors im ply that the market (with the exception of the intervention counterparties) is misinformed and that intervention is unlikely to signal monetary policy accurately. 32 T A B L E 3 White Noise Tests for Errors in Reported Intervention First Half: January 2, 1985M a y 20, 1988 Full Sample Variable Second Half: M a y 23, 1988October 11, 1991 K K-S N K K-S N K K-S N 35.3842a 39.7927a 0.26563 0.4l42a 884 884 10.2977b 20.31093 0.19363 0.36733 442 442 27.51593 30.16263 0.29163 0.43673 442 442 33.68173 43.4345a 0.25883 0.4l20a 884 884 8.6115 20.20563 0.18343 0.36773 442 442 27.25513 32.7564a 0.28973 0.43193 442 442 55.60833 58.98623 0.3180a 0.3472a 884 884 22.20643 39.45403 0.27903 0.38143 442 442 35.7343a 27.4339a 0.33403 0.30693 442 442 53.28133 62.6117a 0.3035a 0.34793 884 884 20.7967s 39.14643 0.25503 0.38663 442 442 34.99083 32.41263 0.32753 0.30133 442 442 N o R umors U.S. vs. DM U.S. vs. Yen With Rumors U.S. vs. DM U.S. vs. Yen D u m m y Variables N o Rumors U.S. vs. DM U.S. vs. Yen With Rumors U.S. vs. DM U.S. vs. Yen Categories of intervention: U.S. vs. DM: U.S. purchases or sales o f DM in terms o f U.S. dollars. U.S. vs. Yen: U.S. purchases or sales o f yen in terms o f U.S. dollars. NOTE: N = num ber of observations. For K and K-S, see box 1. a. Significant at the 5 percent level. b. Significant at the 10 percent level. SOURCE: Authors’ calculations. Appendix Treatment of Rumors and “On Behalf of” Intervention We created two sets of variables from the re ported intervention data: The first treats all ru mors as true, and the second treats all rumors as false. The first step in the creation of both data sets was the formulation of the net dollar transaction variables for each category of inter vention. For the U.S. intervention categories, this variable is equal to the amount variable, which is always non-negative, multiplied by the buy/sell dummy variable. To compare reported and actual interven tion data, we must transfer intervention that was reported as being on behalf of another country to that particular country. For exam ple, if the United States actually purchased yen on behalf of Japan, the data that we receive from the Federal Reserve’s Board of Governors will attribute such intervention to Japan rather than to the United States. To accomplish this adjustment, we created two variables for each country, FORI and FOR2. FORI equals 1 if the country intervened on behalf of another coun try. FOR2 equals the number of countries re ported to be intervening on its behalf. There is also is a third dummy variable, FORRUMOR, which equals 1 if intervention by the country was rumored to be on behalf of an other country. To create the data set in which all rumors are considered true (false), we trans ferred (did not transfer) all of the intervention that was rumored to be on behalf of another country. Additional details regarding these pro cedures are available from the authors. References Baillie, Richard T., and Owen F. Humpage. “Post-Louvre Intervention: Did Target Zones Stabilize the Dollar?” Federal Reserve Bank of Cleveland, Working Paper No. 9203, Feb ruary 1992. Baillie, Richard T., and William P. Osterberg. “Central Bank Intervention and Risk in the Forward Premium,” Michigan State Univer sity, Econometrics and Economic Theory Paper No. 9019, May 1993. Bartlett, M.S. An Introduction to Stochastic Proc esses, 2nd ed. Cambridge: Cambridge Uni versity Press, 1966. Bimbaum, Z.W. “Numerical Tabulation of the Dis tribution of Kolmogorov’s Statistic for Finite Sample Size,” American Statistical Association Jo u rn a l September 1952, pp. 425-41. Danker, Deborah J., et al. “Small Empirical Mod els of Exchange Market Intervention: Appli cations to Germany, Japan, and Canada,” Jo u rn a l o f Policy Modeling, vol. 9 (Spring 1987), pp. 143-73. Dominguez, Kathryn. “Does Central Bank Inter vention Increase the Volatility of Foreign Exchange Rates?” Harvard University, manu script, November 1992. ________ , and Jeffrey Frankel. Does Foreign Ex change Intervention Work? Consequences fo r the Dollar. Washington, D.C.: Institute of International Economics, 1993. Edison, HaliJ. “The Effectiveness of Central-Bank Intervention: A Survey of the Literature after 1982,” Princeton University, International Fi nance Section, Special Papers in International Economics No. 18, July 1993Fisher, R.A. “Tests of Significance in Harmonic Analysis,” Proceedings of the Royal Society of London, series A, vol. 125 (1929), pp. 54-59. Fuller, Wayne. Introduction to Statistical Time Series. New York: John Wiley & Sons, 1976. Ghosh, Atish R. “Is It Signaling? Exchange Inter vention and the Dollar-Deutschemark Rate,” Jo u rn a l o f International Economics, vol. 32, nos. 3/4 (May 1992), pp. 201-20. Glick, Reuven, and Michael Hutchison. “Mone tary Policy, Intervention, and Exchange Rates in Japan,” paper presented at Federal Reserve Bank of San Francisco Conference on Exchange Rate Policies in Pacific Basin Countries, September 1992. Humpage, Owen F. “Central Bank Intervention: Recent Literature, Continuing Controversy,” Federal Reserve Bank of Cleveland, Eco nomic Review, 1991 Quarter 2, pp. 12-26. Hung, Juann H. “Assessing the Effect of Steril ized U.S. Foreign Exchange Intervention: A Noise Trading Perspective,” Federal Reserve Bank of New York, manuscript, January 1992. Kaminsky, Graciela L., and Karen K. Lewis. “Does Foreign Exchange Intervention Signal Future Monetary Policy?” Board of Governors of the Federal Reserve System, Finance and Econom ics Discussion Series No. 93-1, February 1993. Klein, Michael W. “The Accuracy of Reports of Foreign Exchange Intervention,” Jo u rn a l o f International Money a n d Finance, 1993 (forthcoming). ________, and Eric Rosengren. “What Do We Learn from Foreign Exchange Intervention?” Tufts University, manuscript, September 1991. Obstfeld, Maurice. “The Effectiveness of ForeignExchange Intervention: Recent Experience, 1985-1988,” in William H. Branson, Jacob A. Frenkel, and Morris Goldstein, eds., Interna tional Policy Coordination an d Exchange Rate Fluctuations. Chicago: University of Chicago Press, 1990. SAS/ETS User’s Guide. Version 6, 1st ed. Cary, N.C.: SAS Institute Inc., April 1990. Watanabe, Tsutomu. “The Signaling Effect of Foreign Exchange Intervention: The Case of Japan,” paper presented at Federal Reserve Bank of San Francisco Conference on Ex change Rate Policies in Pacific Basin Coun tries, September 1992. Economic Review H 1992 Quarter 4 ■ White- and Blue-Collar Jobs in the Recent Recession and Recover}7: W ho’s Singing the Blues? by Erica L. Groshen and Donald R. Williams Using Bracket Creep to Raise Revenue: A Bad Idea Whose Time Has Passed by David Altig and Charles T. Carlstrom Cyclical Movements of the Labor Input and Its Implicit Real Wage by Finn E. Kydland and Edward C. Prescott Assessing the Impact of Income Tax, Social Security Tax, and Health Care Spending Cuts on U.S. Saving Rates by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff History of and Rationales for the Reconstruction Finance Corporation by Walker F. Todd M 1993 Quarter 1 Generational Accounts and Lifetime Tax Rates, 1900-1991 by Alan J. Auerbach, Jagadeesh Gokhale, and Laurence J. Kotlikoff Has the Long-Run Velocity of M2 Shifted? Evidence from the P* Model by Jeffrey J. Hallman and Richard G. Anderson Examining the Microfoundations of Market Incentives for Asset-Backed Lending by Charles T. Carlstrom and Katherine A. Samolyk 1993 Quarter 2 Money and Interest Rates under a Reserves Operating Target by Robert B. Avery and Myron L. Kwast H 1993 Quarter 3 Capital Requirements and Shifts in Commercial Bank Portfolios by Joseph G. Haubrich and Paul Wachtel FDICIA’s Emergency Liquidity Provisions by Walker F. Todd Efficiency and Technical Progress in Check Processing by Paul W. Bauer Fourth Quarter Working Papers Current Working Papers of the Cleveland Federal Reserve Bank are listed in each quarterly issue of the Economic Review. Copies of specific papers may be re quested by completing and mailing the attached form below. Single copies of individual pa pers will be sent free of charge to those who request them. A mailing list service for personal subscribers, however, is not available. Institutional subscribers, such as libraries and other organiza tions, will be placed on a mail ing list upon request and will automatically receive Working Papers as they are published. ■ 9307 Loan Sales, Implicit Contracts, and Bank Structure ■ 9309 Lender Consistency in Housing Credit Markets ■ 9311 The Equity of Social Services Provided to Children and Senior Citizens by Joseph 6. Haubrich and James B. Thomson ■ 9308 Dynamic Optimal Fiscal and Monetary Policy in a Business Cycle Model with Income Redistribution by Kevin J. Lansing by Robert B. Avery, Patricia E. Beeson, and MarkS. Sniderman by Laurence J. Kotlikoff and Jagadeesh Gokhale ■ 9310 Accounting for Racial Differences in Housing Credit Markets by Robert B. Avery, Patricia E. Beeson, and Mark S. Sniderman Please complete and detach the form below and mail to: Research Department Federal Reserve Bank of Cleveland P.O. Box 6387 Cleveland, Ohio 44101 Check item(s) requested Please send the following Working Paper(s): □ 9307 □ 9309 □ 9308 □ 9310 □ 9311 Send to: Please print Name Address City State Zip Economic Commentary N A F T A and the Midwest by Randall W. Eberts and Lydia K. Leovic October 15, 1992 T h e Importance of Structure in Decisionmaking by Jerry L. Jordan November 1, 1992 Federal Credit and Insurance Programs: Beyond the Deficit Diversion by David Altig November 15, 1992 H o m e Mortgage Lending by the N um b e r s by Robert B. Avery, Patricia E. Beeson, and Mark S. Sniderman February 15, 1993 A n Overview of the Clinton Budget Plan by David Altig and Jagadeesh Gokhale March 1, 1993 A Market-Based Approach to Regulatory Reform by Jerry L. Jordan March 15, 1993 Skepticism about the Direction of Inflation: Causes, Costs, and Cures • by Jerry L. Jordan Price Isn’ t Everything December 1, 1992 by EJ. Stevens April 1, 1993 FDICIA’ s Discount W i n d o w Provisions by Walker F. Todd December 15, 1992 Th e Cost of Buying Time: Lessons from the Thrift Debacle by James B. Thomson January 1, 1993 Bank Exposure to Highly Leveraged Transactions by William P. Osterberg January 15, 1993 S o m e Fiscal Advice for the N e w Government: D o n ’ t Let the Sun G o D o w n on B E A by David Altig February 1, 1993 O n the Political E c o n o m y of Trade Restraints by Jerry L. Jordan April 15, 1993 T he Energy Tax: W h o Pays? by Mark E. Schweitzer and Adam D. Werner May 1, 1993 Does Small Business N e e d a Financial Fix? by Katherine A. Samolyk and Rebecca Wetmore Humes May 15, 1993 Are the Great Lakes Cities Becoming Service Centers? by Erica L. Groshen and Laura Robertson June 1, 1993 D o Deficits Matter? by Owen F. Humpage June 15, 1993 Enterprise Liability: A Prescription for Health Care Reform? by Charles T. Carlstrom July 1, 1993 The Evolving Loan Sales Market by Joseph G. Haubrich and James B. Thomson July 15, 1993 Free Markets and Price Stability: Opportunities for Mexico by Jerry L. Jordan August 1, 1993 Assessing Real Interest Rates by John B. Carlson August 15, 1993 Airline Deregulation: Is It Time to Finish the Job? by Paul W. Bauer and Ian Gale September 1, 1993 Th e Decline in U.S. Saving Rates: A Cause for Concern? by Jagadeesh Gokhale September 15, 1993 Credibility Begins with a Clear C o m m i t m e n t to Price Stability by Jerry L. Jordan October 1, 1993 Th e Budget Reconciliation Act of 1993: A S u m m a r y Report by David Altig and Jagadeesh Gokhale October 15, 1993