The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
The Federal Reserve Bank of San Francisco’s Economic Review is published quarterly by the Bank’s Research and Public Information Department under the supervision of Michael W. Reran, Senior Vice President. The publication is edited by William Burke, with the assistance of Karen Rusk (editorial) and William Rosenthal (graphics). Opinions expressed in the Economic Review do not necessarily reflect the views of the management of the Federal Reserve Bank of San Francisco, or of the Board of Governors of the Federal Reserve System. For free copies of this and other Federal Reserve publications, write or phone the Public Infor mation Section, Federal Reserve Bank of San Francisco, P.O. Box 7702, San Francisco, Califor nia 94120. Phone (415) 544-2184. 2 Windfall Proffits? Interest !at© s and Index Numbers I. Introduction and Summary 4 II. Crude Oil Price Controls and the Windfall Profit Tax: Deterrents to Production? Yvonne Levy 6 . .. As a measure to reduce dependence on foreign-oil imports and improve the allocation of resources, decontrol is a step in the right direction — but the windfall profit tax is a step in the wrong direction. III. Interest Rate Forecasts and Market Efficiency 29 Adrian W. Throop . .. By making use of the information contained in market professionals’ forecasts, an investor in Treasury bills could have improved his return. IV. Indexes, Inflation, and Public Policy 44 Herbert Runyon ...T h e personal consumption “ deflator” has provided a more equitable choice recently than the consumer-price index for determining cost-ofliving adjustments. Editorial committee for this issue: John Scadding, Brian Motley, and Robert Jacobson 3 Production, investing, and household-buying decisions have all come to be influenced by the past decade's upsurge in inflation and increased government participation in market decisions. This issue of the Economic Review presents three studies of this type of influence. One article examines the production losses attributable to price controls - and more recently, windfall-profit taxes - on domestic petroleum producers. Another article evaluates the ability of financial markets, in an inflationary era, to incorporate the best possible interest-rate forecast into current market prices. And a third article analyzes the composition of the price indexes which play such a prominent role, through indexation formulas, in determining household incomes and government expenditures. Yvonne Levy describes the nation's recent experience with price controls on domestically-produced crude oil, which held the average price of domestic crude below the world market level for almost a decade. Controls thus tended to reduce production substantially below the level that would otherwise have prevailed, aggravating a decade-long downtrend in U.S. production. "As a reSUlt, controls worked against the nation's goal of energy independence and led to an inefficient allocation of resources." Crude-oil price controls were finally eliminated this past January, but producers still are not realizing the world price - although they are receiving considerably more than they would have done with continued controls. The reason is the windfall-profit tax, which has been returning to the Federal Government much of the added revenue that otherwise would have accrued to producers through decontrol. "With the tax, future domestic production will be lower than it otherwise would have been, although higher than with continued controls. Similarly, imports will be higher than otherwise, and the misallocation of resources will continue." Levy provides a number of scenarios of production possibilities with different long-run supply elasticities and with different rates of increase of (uncontrolled) crude prices. All of the scenarios show that the decline in domestic production expected over the 1979-85 period with continued controls might have been reversed without controls. But the windfallprofit tax would substantially reduce the positive impact. Levy thus concludes, "As a measure to reduce dependence on foreign-oil imports and improve the allocation of resources, decontrol is a step in the right direction. But by the same token, the windfall-profit tax is a step in the wrong direction." She adds that policymakers should consider altering the tax to make it a true tax on profits, rather than an excise tax on a portion of the selling price. In that way, it would not affect production decisions at the margin. Adrian Throop points out that volatile financial markets in this inflationary period have put a premium on accurate forecasts of interest rates. Thus, he examines the degree to which the Treasury-bill market efficiently utilizes all available information so as to incorporate the best possible interest-rate forecast into current market prices. This involves an evaluation of the applicability of the "efficient market" hypothesis to the Treasury-bill market. Hence he examines two types of independent forecasts - an autoregressive forecasting 4 equation based on the past history of the bill rate, and the forecast of a selected panel of market professionals. If the market is not efficient, Throop argues, a group of investors could improve their returns by altering the maturity of their investments in light of superior interest-rate forecasts. "Whether or not investors can profit by 'speculating' on interest rates, through holding other than their normally preferred maturities, depends critically on whether available interest-rate forecasts are more accurate than the market's." If the market is efficient, the information in these forecasts would already be incorporated into the prices of securities, and therefore nothing would be gained. In that case, investors would be better off by simply "hedging" their positions with maturities equal to their planned investment periods, thereby avoiding possible risk. Throop shows that professional analysts' predictions of the Treasury- bill rate two quarters ahead are significantly more accurate than market predictions. "This indicates that the market does not efficiently utilize all available information in making bill-rate forecasts." The analysts' ability to "beat the market" stems from a better utilization of information about past movements in the bill rate, and also from a more efficient utilization of other sorts of information. Herbert Runyon then turns to the question of measuring inflation, which, as he emphasizes, means an increase in the general price level rather than separate price increases for individual commodities. Since numerous goods and services are bought and sold daily in the markets, an index number represents the only feasible way of describing the general movement of prices through time. But this index number should give an accurate representation of changes in living costs - at least indirectly through market observations as consumers reveal their individual preferences by purchasing certain goods and services. For the household sector, the standard measures include the Consumer Price Index (Cpn and the Personal Consumption Expenditure (PCE) deflator. The CPI - a "Laspeyres" index - uses quantities purchased in a certain base year as a reference point from which to measure changes in prices of a basic (presumably unchanging) consumer-expenditure pattern. The PCE - a "Paasche" index - uses the current-period expenditure pattern as a reference base for comparison with expenditures in earlier periods. For structural reasons, the cpr tends to overstate - and the PCE to understate - what consumers actually pay for at the checkout counter. As a practical matter, the CPI and PCE were quite close in their measurement of living costs from 1960 through 1977. But in the 1978-80 period, the CPI rose at a much faster rate - not so much because of the indexes' different statistical composition as because of their different treatment of sharply rising homeownership costs. Among other things, the CPI's weightiDg of homeownership overstates its importance in the cost of living. Runyon argues that the use of the CPl in recent years has overcompensated many income recipients - especially beneficiaries of Federal programs, since the Federal government depends on the CPI as a standard index in its efforts to offset the impact of inflation on such individuals. This overcompensation has several public-policy consequences. "Overcompensation leads to unwarranted increases in Federal expenditures and in the Treasury deficit. Beyond that, it introduces inequities relative to those individuals not receiving indexed benefits, and thus amounts to an unintended redistribution of income." Thus, he concludes, recent experience suggests that the PCE index is the more equitable choice for determining cost-of-living adjustments. 5 Yvonne levy'" Throughout the decade of the 1970's, a complex system of Federal controls governed prices for domestically-produced crude oil. Those controls held the average price of domestically-produced crude below the world market level. Consequently, controls tended to reduce production substantially below the level that would otherwise have prevailed, aggravating a decade-long downtrend in U.S. crude-oil production. For any given level of refiners' crude-oil demand, the reduction in domestic production raised imported-oil requirements by an equivalent amount. Moreover, that added import volume involved greater resource costs than would have been required through domestic production. As a result, controls worked against the nation's goal of energy independence and led to an inefficient allocation of resources. Despite President Reagan's January 1981 lifting of controls on domestically-produced crude, producers still are not realizing the world price. The Windfall Profit Tax - which went into effect March 1, 1980 and could extend through 1990 - has been returning to the Federal government much of the added revenue that otherwise would have accrued to producers through decontrol. With the tax, producers are realizing less than the world market price,. although of course more than they would have realized with continued con- trois. Thus, with the tax, future domestic production will be lower than it otherwise would have been with decontrol and no tax, although higher than with continued controls. Similarly, imports will be higher than otherwise, and the misallocation of resources will continue. Section I presents a simple model of the supply of domestically-produced crude oil which shows how supply responds positively to prices received by producers. It also shows that, in the absence of controls, domestic crude oil would sell at approximately the world price because domestic producers operate in a world market. Section II describes the major features of the crude-oil price-control program contained in the recently terminated Energy Policy and Conservation Act of 1975. That section shows how that program held the average domestic price below the world market price, and thus kept domestic output below the amount that would have been produced in a free market. Section III outlines the major provisions of the Crude Oil Windfall Profit Tax of 1980. It shows how the tax leaves the price realized by producers still below the world , reducing the positive impact of decontrol on domestic crude-oil production. Finally, Section IV presents a range of estimates of the domestic production losses that resulted from the Energy Policy and Conservation Act, and that potentially could result over the next decade from the Windfall Profit Tax. The recently-passed Reagan tax program contains minor changes in the windfall-profit tax, but these do not materially affect the conclusions of this paper. 'Senior economist, Federal Reserve Bank of San Francisco. Research assistance provided by Alane Sullivan and Lloyd Dixon. 6 I. Domestic Oil Production in a Free Market Long-run Supply Model Building upon this foundation, we can further clarify the concept of the long-run supply schedule for representative individual properties, the domestic crude-oil producing industry and the combined domestic and import sectors through Charts lA-C. The charts show, at a particular period in time, the quantity of crude oil that will be available to the U .5. market from those various sources at various selling prices. We assume, throughout the analysis, that the crude-oil producing industry is workably competitive, in line with the bulk of the evidence presented in the recent academic literature. With regard to structure, the producing and refining sectors of the industry are clearly different. Thousands of U.5. firms are engaged in the exploration and extraction of crude oil, with no firm dominant - in contrast to the oligopolistic refining sector of the industry.4 Chart 1A shows the long-run supply schedule for the representative individual property. This is the long-run marginal cost curve (MC) - the addition to total cost resulting from the last unit of output. The marginal cost of producing additional barrels from a given reservoir increases because firms must invest in higher-cost recovery techniques, such as enhanced oil-recovery methods and deeper development wells, as more oil is extracted from a finite reservoir. In a competitive market, each firm maximizes profit by expanding output to the point where marginal cost equals price. The schedule is upward sloping; as the price rises, it pays firms to the higher-cost barrels that would have been uneconomic to produce at a lower price. Chart lB shows the long-run supply schedule, 5 0 5 0 , for the entire domestic crudeoil producing industry. That schedule is derived by summing the amounts produced by all properties at each price - that is, summing horizontally the marginal cost-output curves for all producing properties. Again, the schedule is upward sloping because, with Domestic crude-oil production can be increased over the long-run in a number of ways, all of which are encouraged by a rise in selling price. "Long-run" means a planning period long enough to permit producers to invest in new productive capacity to achieve higher production rates. In crude-oil production, new productive capacity can be installed either at existing (i.e., already-producing) properties or at entirely new sites. I At existing properties, producers can increase the rate of extraction by drilling more development wells. Alternatively, they can invest in enhanced oil-recovery technologies. This involves drilling service wells through which steam, chemicals, or gases may be injected to increase well pressure, thus raising the recoverable proportion of the total reservoir. In addition, producers can expand production through the discovery of new properties - reservoirs in unproven regions as well as near already-producing areas. But first, producers must do some exploratory drilling and identify resources that are recoverable under current economic and technological conditions. The addition of development wells at known reservoirs permits a higher rate of extraction from a given deposit, but it does not increase the ultimate, or total cumulative, production potential. 2. This potential can be expanded only through an increase in proved reserves, resulting from investments in enhanced oilrecovery technologies and the discovery of new economic resources. "Proved reserves" refer to the portion of the resource base that has been identified and explored, and from which crude oil can be recovered profitably at current prices and with current technology. J While the occurrence of oil is finite, being governed by geology, a host of other factors economic, technological, environmental and political - determine the rate at which oil resources are discovered, developed and transferred to the category of reserves. 7 Chart 1 United States Crude Oil Supply Without Controls A. INDIVIDUAL DOMESTIC PROPERTY Price (dollars per barrel) Pw MC 1---------1 qw Quantity (barrels/unit of time) Price (dollars per barrel) Pw B. TOTAL DOMESTIC INDUSTRY t-------------.y Qw Quantity (barrels/unit of time) C. TOTAL U. S. CRUDE OIL MARKET Price (dollars per barrel) ___ Pw t------------,-~~------ S Domestic + Imports Qw QD Quantity Domestic Domestic (barrels/unit of time) production consumption 8 increased production, firms must locate, develop and extract oil from less accessible and poorer quality resources. 5 Chart lC shows the long-run supply schedule for domestic and foreign oil available to U.S. refiners at various selling prices. At prices below the world price, P w , the supply comes entirely from domestic sources. The price cannot exceed P w , since imports are available essentially without limit at that price. Hence, the total supply schedule is represented by the kinked curve, So as. Schedule DD meanwhile represents U.S. refiners' crude-oil demand schedule. In the absence of price controls, domestic crude oil would sell at approximately the world price, P w - the landed price for imported oil. 6 This is because U.S. producers operate in a world market. At the world price, domestic production cannot meet the total quantity demanded by U.S. refiners. The price of imported crude thus represents the marginal cost of an additional barrel - and thus determines the marginal domestic producer price. At the world price the domestic price that would prevail without controls - domestic producers would be willing to supply Qw barrels. Domestic demand would be QD, and imports in the amount of QD-Qwwould be required. Efficiency in the allocation of resources requires that the total cost of satisfying any given quantity demanded be as low as possible. 1 When the alternative to domestic oil is imported oil purchased at the world price, efficiency requires that production from all domestic properties be expanded to the point where the marginal cost of the last unit of output is equal to the price of imported oil. Beyond that point, resources could be saved by reducing domestic output and replacing that output with imported oil. But below that point, where the cost of the last barrel produced is less than the price of imported oil, resources could be saved by reducing imports and expanding domestic production. Since the supply schedule So So reflects the marginal cost of producing domestic oil, the uncontrolled market solution for domestic and foreign supply, So as, represents an efficient allocation of resources. In this allocation, the marginal cost of production for all domestic producers is equal to the world price. There is no opportunity to reduce total cost by shifting supply between domestic and foreign sources. II. Domestic Oil Production Under Price Controls The Federal price-control programs of the 1970's held the average selling price of domestically-produced crude below the world market price, and thereby disturbed the efficient free-market solution. But government attempts to influence domestic prices first developed in the 1930's - although their purpose was to hold the producer price above (rather than below) the competitive level. During the 1930's, oil producing states instituted "conservation" programs - ostensibly to prevent "wasteful" production practices, but in reality, to keep prices high by limiting production. 8 Those programs were effective until the mid-1950's, when increas- ing quantities of foreign oil became available at prices well below the average domestic producer price. After trying (unsuccessfully) to restrict imports voluntarily, the Federal government in 1959 introduced a program of mandatory import quotas, using national security as justification. The early 1970's witnessed a fundamental change in the nation's demand for imports. Despite the quota system, domestic crude-oil production peaked by 1970, as import competition prevented the domestic price from rising as fast as production costs. By 1973, U.S. petroleum consumption had outgrown domestic production, and imported oil had 9 become the required source of marginal supplies. The marginal cost of imported oil, i.e., the world price, thus became the determinant of the domestic producer price. tives.Since decisions to expand production are determined on the margin, they permitted production in excess of the base level to receive a higher ceiling price. In that way, they hoped to provide sufficient incentive to stimulate production. But they also wanted to prevent owners of wells brought into production before the OPEC price run-up from receiving profits far higher than originally anticipated and so held the lower-tier price not only below the world level but the upper-tier price. In this view, the removal of such unanticipated profits - "windfalls" - would have little impact on production decisions. (A similar philosophy apparently underlies the new Windfall Profit Tax.) Nevertheless, because controls generally held the price at the margin below the world price, production was less than it otherwise would have been. Charts 2A-C illustrate the price and production effects of the Energy Policy and Conservation Act for representative properties and the entire domestic industry. 12 Properties in existence before 1975 faced the marginal revenue (MR) schedule shown in Chart 2A. Output of q, or less qualified as stripper oil and received the uncontrolled (world) price, P w. Output from qs to the base production control level (BPCL) received the lower-tier price, PL' Output in excess of BPCL received the upper-tier price, P u. "New" properties - those brought into production after 1975 - faced the marginal revenue schedule shown in Chart 2B, and except for small stripper properties, received the upper-tier price. Existing stripper properties were unaffected by this system of price controls. Those properties received the world price, with or without controls, and hence production remained the same at q,. But for other existing properties, production was lower than it would have been without controls, since production was not permitted to receive the world price. The magnitude of the impact depended on whether costs had risen since the 1975 base year. Producers not incurring higher costs would have found it profitable to expand production past the base-period level, to earn the uppertier price on the new production. Production Energy Policy and Conservation Act The Federal government first placed direct controls on prices of domestic crude oil on August 15, 1971, when Presiden t Nixon froze wages and prices throughout the economy. 9 A multi-tier pricing system evolved in Phase IV of the controls program, announced in August 1973, and in the Emergency Petroleum Allocation Act, passed in November 1973 during the Arab oil embargo. These then led to the Energy Policy and Conservation Act of 1975 (EPCA). That legislation controlled domestic producer selling prices on a propertyby-property basis, with production above and below the "base production control level" (BPCL) -the 1975 average monthly production - subject to different price ceilings. Under the act, "lower-tier" oil referred to output at or below the BPCL, while "upper-tier" oil referred to output in excess of this base level or output from new properties brought into production after 1975. The law stipulated a lower price ceiling for lower-tier oil than for upper-tier oil, and stipulated that ceilings would be set to achieve a target average price for domestic oil. That price could rise to reflect inflation, but by no more than 10 percent annually. 10 Initially, the law classified oil from "stripper" properties those producing ten barrels or less daily - as upper-tier oil. But in September 1976, the energy agency decontrolled stripper oil and thus permitted it to receive the world market price. Policy makers designed the control program to hold the average price of domestically-produced crude below the world market level, and thereby protect consumers from the full impact of sharply rising world prices - in effect, transfering to consumers much of the added income that would otherwise have accrued to producers. II At the same time, policy makers sought, through the multi-tier system, to accomplish that objective with the least possible reduction in production incen10 Chart 2 United States Crude Oil Supply Under Energy Policy and Conservation Act A. INDIVIDUAL DOMESTIC PROPERTY B. INDIVIDUAL DOMESTIC PROPERTY (High-cost Existing) (New) Price Prir.e (dollars (dollars per barrel) per barrel) Pw Pw MR Pu Pu qs ql quqw BPCL Quantity (barrels/unit of time) Price (dollars per barrel) Quantity (barrels/unit of time) C. TOTAL U. S. CRUDE OIL MARKET ------------r--ilIItf--------- Pw ...... S Domestic + Imports Pc 1-"""""""""""'''''''''''''''''''''''''''''''''''''''' So 0c QW QD Quantity Domestic Domestic (barrels/unit of time) production consumption 11 still would have been lower than it would have been without controls, however, since the upper-tier price was below P w. Producers incurring higher costs than faced in 1975 would have fared less well. (See the marginal cost schedule, MC I> in Chart 2A.) For those properties, producers would not have found it profitable to expand production beyond the base-period level. The new production up to the base level would earn only P b the lowertier price, which would be below the cost of producing it. The profit earned on upper-tier oil (shaded area) would not cover the loss incurred on lower-tier oil (striped area) and the firm would choose to remain producing at ql' Producers incurring an extremely sharp rise in costs since 1975 would have an incentive to actually lower production to 10 barrels or less per day, to qualify for the uncontrolled price afforded strtpper properties. Producers with new properties would have found less incentive to produce than in a situation with no controls (Chart 2B). When the world price was P w , output from new properties was only qu because they were permitted to receive only the price P u. Had price controls not existed, output on new properties would have been qw, because that output would have received the world price. At the industry level, this system of controls, like its predecessors, tended to hold the average domestic producer price below the world market level, thereby reducing production below the level that would otherwise have prevailed (Chart 2C). When the world price is P w , controls hold the average domestic producer price at Pc. As a result, the domestic industry produces only Qc instead of the quantity Qw produced in the absence of controls. For a given level of refinery crude oil demand, the consequent reduction in domestic output is offset entirely by imports. Thus the control program tended to increase the nation's dependence on foreign oil. The control program also led to inefficiency in the allocation of resources. Between output levels Qc and Qw, each additional barrel of crude could be produced domestically at a cost below the world price, and thus at a smaller expenditure of resources than for imported oil. Area dca, which equals the difference between the world price and the domestic supply schedule SoSo at each increment between Qc and Qw, represents resources wasted on expenditures for imported oil because of controis. 1] Intertemporal Production Decisions Thus far, we have considered the effects of controls only in a static framework. We have assumed that firms consider only the current price, without regard to price expectations, when making production decisions. Also, we have ignored the potential effects of the current level of output on future production. We have assumed that firms could obtain optimal production and profit paths over time by producing at the point where marginal cost equals price in each planning period. But there is an important difference between the marginal-cost (supply) schedule of a typical manufacturing firm and the schedule of a firm extracting an exhaustible resource such as petroleum. As petroleum is removed from a reservoir, the pressure of the reservoir declines, and so too does the total amount of petroleum available - a tendency known as the "natural decline function." Because of the exhaustible nature of the resource, a barrel of oil produced today will not be available in some future period. Petroleum producers thus face an additional cost of production not incurred by manufacturers. That additional cost the user cost - is the opportunity cost or profit foregone of being unable to sell that unit of output in the future. In order for a producer to decide to produce a barrel of oil in the present, the price of each additional barrel of oil produced today must be sufficient to cover this opportunity cost as well as normal production costs. Moreover, in view of the effect of current output on future output - the "natural decline" problem - the firm cannot simply select the output level in each period where marginal revenue equals marginal cost (defined to include user cost). Instead, the firm must maximize a stream of profits over time, which involves a discounting 12 troIs on current prices may have created just such an expectation of higher prices in the future when controls eventually might be lifted. The expected price path in moving from control to decontrol would have been steeper than had prices never been subject to controls. As a result, controls may have raised the user cost of controlled oil, thereby causing producers to restrict current production even more than they would have done because of receiving less than the world-market price. This was especially true of the 1978-80 period, when market participants widely expected eventual decontrol. 15 procedure. This present-value analysis, in effect, states that for production of an additional barrel to take place, the present profit invested at the market rate of interest must at least be equal to the profit from selling that barrel any time in the future. 14 By affecting price expectations, the controls program may have exerted still another restraining effect on current production. If the expected path of future prices rises, the user cost increases and producers can expand profits by deferring current production to the future. In this case, the present value of the future profit would exceed today's profit. Con- III. Domestic Oil Production Under "Decontrol" On June 1, 1979, the Energy Department began to implement a program, mandated by President Carter, for decontrolling domestic crude-oil prices by October 1, 1981. Under that program, production that previously would have been subject to the lower-tier price was permitted to move gradually to the uppertier category. Then, beginning on January 1, 1980, production previously classified as upper-tier oil, plus the lower-tier oil moving into the upper-tier category, was permitted to move to a free-market classification at a rate of 4.6 percent per month. The Energy Department decontrolled oil discovered after January 1, 1979 on June 1 of that year, and it lifted controls on "heavy" crude on August 17,1979. Finally, the Reagan Administration - in its first major economicpolicy move - lifted all remaining price controls on domestically produced crude on January 28, 1981. In moving to decontrol domestic prices, both the Carter and Reagan Administrations hoped to encourage domestic production and to slow down the growth of U.S. petroleum consumption. To the extent that the higher refiner costs for domestic crude were reflected in higher refined-product prices, consumption should be curtailed. 16 Windfall Profit Tax At the same time, Congress was unwilling to permit producers to realize all the added revenue that would accrue through decontrol, especially since that step could boost producer revenues by about $1 trillion over the 1980-90 period, according to estimates of the Joint Committee on Taxation. 17 As a result, Congress enacted the Crude Oil Windfall Profit Tax of 1980 to divert to the U.S. Treasury some of the incremental revenues that would otherwise be received by producers through decontrol. The tax became effective March 1, 1980. The tax is perhaps the largest ever imposed on a single industry. Over the 1980-90 period, the tax could yield about $236 billion, in addition to a $332-billion increase in corporate income taxes resulting from decontrol. Thus, the U.S. Treasury could receive $568 billion of the projected $l-trillion additional industry revenue received from decontrol. 18 Although called a tax on profit, the tax really is a Federal excise tax on a portion of the selling price received from crude oil. The tax paid per barrel is determined by applying various tax rates to the "windfall profit" - the difference between the decontrolled producer price and the price that would have prevailed 13 under continued controls (less state severance tax) . The tax rates and base prices applicable to various properties vary according to type of production and size of producer (Table 1). The Internal Revenue Service established these new oil categories for tax purposes. Producers are classified either as "majors" or as "independents" (producers with gross annual sales of $5 million or less and with refining capacities of no more than 50,000 barrels a day). Identical tax rates are applied, except for the first 1,000 barrels a day of Tier 1 and Tier 2 production by independents. To provide greater incentive for certain investments, the lowest tax rates apply to newly discovered and incremental tertiary oil, the latter being oil obtai ned t hroug h a qual ified tertiary (enhanced) recovery method, i.e., production in excess of the projected decline rate for the property without the tertiary technique. 19 In computing the tax, producers first subtract a base price, adjusted for inflation, from the decontrolied producer price (Appendix A). Next they subtract a state severance tax - estimated to average about 5.4 percent of the selling price - to determine the "windfall profit." Then they apply the appropriate tax rate to determine the amount of tax to be paid. Table 1 Provisions of the Windfall Profit Tax 1 Tax Rate Integrated Independent Producer Producer 3 (percent) Base Price ($ per bbl.) Annual Adjustment to Base Price (percent) Tier 1 Controlled oil discovered before 1979 2 70 50 1281 Inflation 6 Tier 2 Stripper well oil 60 30 15.20 Inflation National Petroleum reserve oil 60 30 15.20 Inflation Tier 3 Newly discovered oil 30 30 16.55 Inflation Heavy oil 30 30 16.55 Inflation Incremental tertiary oil 4 30 30 16.55 + 2% + 2% Inflation + 2% ExemptS 1 The Windfall Profit Tax of 1980 was enacted into law on April 2, 1980. But the tax was retroactive, i.e., applicable to crude-oil production removed from properties after February 29, 1980. 2 For purposes of the windfall-profit tax, the pricing categories conform to those in effect in May 1979, before the process of gradual decontrol began. 3 The special reduced-tax rates afforded independent producers for Tiers I and 2 are applicable only to their first 1,000 barrels per day of production. Production in excess of 1,000 bid is taxed at the regular windfall-profit tax rates, i.e., the rates applicable to integrated producers. 4 Incremental tertrary oil is the amount of production from a property on which a producer uses a qualified tertiary (enhanced) recovery method in excess of the projected decline rate if a tertiary technique had not been used on that property. 5 Categories of crude exempt from the tax include: qualified governmental-interest oil, qualified charitable-interest oil, certain Indian oil, Alaskan oil (other than from the Sadlerochit reservoir) north of the Alaska-Aleutian mountain range and over 75 miles from the pipeline, and tertiary oil from properties owned by independent producers. 6 Inflation is measured by the change in the Gross National Product deflator. 14 Chart 3 shows how the windfall-profit tax will affect U.S. crude oil production as long as the uncontrolled domestic price remains above the adjusted base price. 20 So So represents the long-run supply schedule for the domestic crude-oil producing industry. Decontrol with no tax would provide producers with the maximum incentive to increase production, and would lead to an efficient ailocation of resources in meeting with the nation's petroleum requirements. In this situation, producers would realize the uncontrolled price, P w , and produce at output Qw. With continued controls, producers would realize price, Pc, and supply Qc. The tax lowers the price realized by producers below the free-market price to an intermediate price, PT' Since the uncontrolled domestic selling price is determined by the world price, producers must absorb the tax as a reduction in realized price, and cannot pass it on to consumers. The tax thus reduces the incentive to increase production provided by decontrol. At price, P T, quantity QT is produced - more than would be produced with continued controls but less than would be produced with decontrol and no tax. Production is greater than with continued controls because only part of the so-called "windfall" is diverted to the U.S. Treasury. Chart 3 Effects of Windfall Profit Tax on Domestic Crude Oil Production Price per barrel Pw t--------~" Pr Q C Q r Q W Quantity of Crude Oil IV. Estimating Production losses Crude-oil production in the United States dropped 10 percent over the 1970-79 period, and would have dropped 24 percent except for the addition of 1.4 million barrels per day of Alaskan North Slope production. The decadelong decline in production resulted inevitably from a decline in the amount of oil added to reserves annually, through new discoveries and enhanced oil recovery, during the 1970-79 period compared with the decade of the 1960' s. Unless gross annual addi tions to reserves exceed the rate of extraction, the total inventory of proved reserves declines. Producers were forced to lower production so as not to experience an even greater run-down in their proved reserve inventory. Even with a cutback in production, proved reserves - the industry's working "in the ground" inventory - declined steadily over the 1970-79 period from 39.0 to 27.1 billion barrels. Even with controls, the average wellhead price for domestically-produced crude rose three-fold over the 1973-79 period (TapIe 2). This price upsurge led to a reversal of the prolonged decline in exploration activity that had occurred over the preceding decade and a half. Between 1973 and 1979, the total number of oil wells drilled nearly doubled, rising at an average annual rate of 16 percent. Nonetheless, additions to reserves still dropped from an average annual rate of 2.7 billion barrels during the 1965-69 period to 2.0 15 servation Act production? billion barrels during the 1971-74 period, and then to only 1.3 billion barrels during the 1975-78 period. The rate of reserve additions only began to pick up, to 2.2 billion barrels, with a new price upsurge during the period of gradual decontrol in 1979. Without price controls, drilling activity would have increased somewhat faster, raising annual additions to reserves as weB as production at existing properties through development drilling. How much did the controls more specifically the Energy Policy and Con- contribute to the decline in Energy Policy and Conservation Act To answer that question, we could develop a petroleum-supply model to project production under an uncontrolled price assumption. Then, that outcome could be compared with actual production to estimate production losses. Given the fact that the oil-supply process involves several phases - exploration, reservoir development, and production - the development of such a model would be a vast Table 2 Relationship Between Controlled Domestic Crude Oil Prices and World Price, 1972·80 Average Domestic Price: Controlled as Percent of Uncontrolled Average Refiner Acquisition Price Average Domestic Price at Wellhead Year Imported Crude' Without Controls (est.) With Controls 1972 3.22 3.39" 3.39 100.0 1973 4.08 3.89" 3.89 100.0 1974 12.52 11.66' 6.87 1975 13.93 12.95' 7.67 1976 13.48 12.56' 8.19 5.13 1977 14.53 13.59" 8.57 1978 14.57 13.95" 9.00 1979 21.67 22.93" 33.82 35.54'; 1980 Old 011' New 011" 5.03 10.13 58.9 5.03 12.03 Lower-Tier Upper-Tier Stripper 59.2 11.71 12.16 5.19 11.22 13.59 6.35 12.34 63.1 5.46 12.15 13.95 5.22 12.85 64.5 12.64 5.95 13.20 22.93 10.57 19.40 55.1 21.20 6.49 14.34 3554 14.14 3306 59.6 Naval Alaska Petroleum North Slope Reserves 65.2 I Before entitlement benefit. 2 In 1972 and most of 1973, petroleum prices were subject to the Phase I-IV price provisions of the Economic Stabilization Act. The average domestic wellhead price during that period did not differ substantially from the refiner acquisition cost of crude, so it can be assumed that controls had little effect on domestic crude-oil prices. As a reSUlt, prices are assumed to be the same with or without controls. ;i Estimated by excluding domestic producers' transportation costs from the refiner acquisition price for imported oil, the adjustment being based on the post-1976 price of stripper oil. This adjustment reflects the fact that the wellhead price for stripper oil-deconlrolled in September 1976-represents an uncontrolled price for domestic oil exclusive of transportation costs. 4 Under the Emergency Petroleum Allocation Act in effect throughout 1974 and 1975, "old" oil in any given month was defined as the quantity produced from a given property in the corresponding month of 1972. " Under the Emergency Petroleum Allocation Act, "new" oil was defined as any production in excess of output in the same month of 1972. " The post-1976 (decontrolled) stripper price was used as a proxy for the uncontrolled domestic producer price. Actually, in 1979 and 1980, stripper oil sold at a substantial premium above comparable-quality imported oil, due to tight worldwide supply conditions and the willingness of refiners to pay a premium for security of supply. Source: Uncontrolled domestic price at the wellhead estimated by author as described in footnote 3 above. Actual prices as published by U.S. Department of Energy, Monthly Energy Review. 16 production, as we also have done here. 23 These studies have yielded a wide range of elasticity estimates - ranging from .3 to .8 - with the variation perhaps due in part to the different time periods involved. 24 Utilizing these published elasticity estimates, we estimated production losses for the 1976-79 price-control period under several different elasticity assumptions (Table 3). The elasticity of production from existing properties was assumed to be .2 in each scenario. However, the elasticities of production from new properties ranged from .3 (low), to .5 (medium) to .8 (high assumption). All these elasticity estimates pertain to the long-run. They show how production eventually might respond to a given change in price after time had elapsed for exploration and development. In our estimates, however, we have recorded the response as if it shows up fully within a year. For example, the estimated production losses for 1976 represent the additional production that would be forthcoming in the long-run as a result of closing the differentials between controlled and uncontrolled prices prevailing during that year. Although recorded for that single year, in reality the production response would take considerably longer. The results indicate a substantial lowering of production by controls in the 1976-79 period, under all scenarios (Table 3). The likeliest outcome would probably arise from the lowresponse assumption, because elasticity of supply would surely fall in the wake of a substantial widening of price differentials. Under this low-response scenario, production with uncontrolled prices eventually could have been 10 percent higher in 1976, and 29 percent higher in 1979, than actual production with controlled prices. The comparable production increases under the high-response scenario would have been 15 percent in 1976 and 73 percent in 1979 - although the underlying elasticity estimates in that scenario appear to be unrealistically high. The limited nature of the resource base, and the further depletion in recent years, would preclude the likelihood of very high supply elasticities - .8 for new dis- undertaking. Instead, we have relied upon already existing models to obtain long-run price elasticities of supply upon which to estimate the production losses resulting froJ1.l that particular control program. Elasticity of supply is a measure of the responsiveness of the quantity supplied of a given product to an increase in its price. In the absence of controls, the average wellhead prices of lower and upper-tier crude oil would have risen from their respective controlled levels to the world market level. 21 At existing properties - those in operation before 1975 - an increase in prices would have: (1) raised investment in development wells, thereby raising the rate of extraction from proved reserves and (2) encouraged greater investment in enhanced oil-recovery methods, thereby increasing additions to reserves through improved technology. An increase in the upper-tier price also would have encouraged more exploratory drilling, leading to the discovery of more new reserves at new properties. This study attempts to estimate how much extra production would have been forthcoming had prices been allowed to rise to the world price. It does so by drawing on outside estimates of the elasticity of supply, the percentage change in quantity supplied divided by the percentage change in price. In estimating production losses as a result of controls, we calculated the percentage difference between the uncontrolled and controlled prices in any given year, multiplied that by the supply elasticity to get the percentage change in output, and then converted that percentage change to an arithmetic change by multiplying it by the existing quantity. Recent studies by the Department of Energy suggest a long-run price elasticity of about .2 for categories of production equivalent to "existing properties."22 Numerous econometric studies are available for deriving elasticity estimates for new properties. These studies relate price to additions to reserves from new discoveries. The authors then assume that a given increase in new reserves leads to an equivalent percentage increase in 17 Table 3 Domestic Crude Oil Production Losses Under the Energy Policy and Conservation Act, 1976-79 (Production in millions of barrels) Domestic Crude 011 Prices' (Dollars per barrel) Prices with Controls Domestic Crude Oil Production (Millions of barrels) Year lower-Tier Upper-Tier Estimated Price Without Controls" Total Production With Controls 1976 5.13 11.71 12.56 2968 1977 5.19 11.22 13.59 3009 1978 5.46 12.15 13.95 3178 1979 5.95 13.20 22.93 3115 Domestic Crude Oil Production (Millions of barrels) Estimated Production Losses Due to Controls" Low Response Year Existing Properties' 1976 267 1n","J ~..,.., New Properties' 18 Estimated Total Production Without Controls Additional Domestic Production Without Controls as a Percent of Production with Controls 3253 9.6 1 A 1 p" J'11"'''7 .... L.I "IJ.'7" 3495 10.0 1'7'( £.// 1978 228 89 1979 350 561 4026' 29.2 1976 267 115 3350 12.9 1977 277 330 3616 20.2 1978 228 244 3649 14.8 1979 350 1102 4566 46.6 1976 267 187 3421 15.3 1977 277 544 3830 27.3 1978 228 398 3804 19.7 1979 350 1924 5389 73.0 Medium Response Year High Response Year , Annual averages; producer prices at the wellhead. , For an explanation of the derivation of this price series, see Table 2. Derived by the author on the basis of three separate assumptions regarding price elasticity of supply for new properties. Each response is assumed to become fully effective within a single year, although in reality, the production response to closing any given differential between controlled and uncontrolled prices would take more than a year. :j 1 Elasticity figures for existing properties are .2 in all cases; for new properties .3 for low response, .5 for medium response, and .8 for high response. Source: Price and actual production data: U.S. Department of Energy, Monthly Energy Review. Estimated production losses computed by author using methodology described in text. 18 coveries - included in that scenario. 25 In all scenarios, the potential losses with controls would have been very large on the basis of 1979 prices because of the huge increase estimated for wellhead prices in the absence of controls. The average refiner acquisition price for imported crude (before entitlements) rose by only 41f2 percent over the entire 1975-78 period, but then jumped 49 percent within 1979 alone (Table 2), as OPEC members sharply boosted prices in the wake of the cutoff of Iranian exports and consequent world-wide tightening of supplies. The average domestic wellhead price without controls the average refiner acquisition price for imported crude less transportation costs from domestic wells to refineries - thus would have risen about 58 percent in 1979. Indeed, that wellhead price probably would have risen even more, because of refiners' willingness to pay a premium for domestic oil for security of supply. In this situation, controlled prices of lower-tier and upper-tier oil would have been only 17 and 39 percent, respectively, of the estimated uncontrolled price for domestic oil. Windfall Profit Tax The windfall profit tax will dilute the stimulus to increased production provided by decontrol. The amount of dilution will depend principally upon the future behavior of the uncontrolled domestic-producer selling price - and thus the after-tax realized price - and upon the production response to any given increase in realized price. Based upon an assumed 2 percent "real" average annual increase in the uncontrolled domestic producer selling price over the 198090 period, we have developed different sets of estimates of the offsetting effect of the windfall-profit tax on the positive production response from decontrol for the years 1985 and 1990. 26 This involves the development of three alternative policy assumptions: (l) Scenario I, continued price controls; (2) Scenario II, decontrol with no windfall-profit tax, and (3) Scenario III, decontrol with a windfall-profit tax. As before, we have assumed a lower elasticity of supply for oil from existing properties - Tiers 1 and 2, as well as heavy oil and incremental tertiary oil - Table 4 Estimated Domestic Producer Oil Prices, 1985 and 1990, Under Three Alternative Policy Assumptions! Scenario I I 2 :J 4 Scenario II Prices with Decontrol but No Windfall Profit Tax" Scenario III Prices with Decontrol and Windfall Profit Tax' Tier 1 Tier 3 Tier 2 Year Prices with Continued Controls' Tier 1 Tier 2 Tier 3 1985 19.92 23.64 28.23 53.93 29.57 35.10 45.25 1990 26.78 31.77 41.92 79.84 41.84 49.96 67.03 Annual averages in dollars per barrel; producer sales prices. For 1985 and 1990, prices with controls were estimated by making inflation adjustments to the base prices for each tier (as defined in the Crude Oil Windfall Profit Tax of 1980, Table O. For 1980, the adjustment factors were based on the GNP deflator. Thereafter, the inflation rate was assumed to decline gradually, reaching a 6.0-percent annual rate by mid-1986 and remaining at that rate through 1990. For Tier 3, the base price was adjusted upward by an additional 2 percent each year, as allowed in the law. The author estimated that domestic oil, without controls, would have sold for $35/barrel in the fourth quarter of 1980. For 1985 and 1990, the decontrolled price was estimated by adjusting the 1980 price to reflect inflation plus an assumed 2percent annual increase. The anticipated price of heavy oil in 1980 was assumed to be $7.50 less than the average domestic price, reflecting the traditional price differential. The author assumed that the windfall profit tax effectively reduces the selling price actually realized by domestic producers. For a description of the derivation of these prices, see Appendix A. The tiers represent the categories of domestic oil as defined by the Crude Oil Windfall Profit Tax of 1980, as shown in Table 1. 19 than for oil from newly discovered properties. And again, we have assumed a range of elasticities for Tier 3 newly-discovered oil, ranging from .3 (low), to .5 (medium) to .8 (high assumption). To develop production estimates, we first estimate producer prices under each of the three scenarios (Table 4), making the estimates in nominal terms to conform with the actual computation of the windfall-profit tax. The tax affects production in any given period through its impact on the realized relative price of oil compared with what it would be without the tax. Under continued controls (Scenario I), we calculate prices for Tier 1 and Tier 2 oil for the years 1985 and 1990 as equal to certain base prices (defined by the Windfall Profit Tax) Table 5: Estimated Domestic Crude Oil Production, 1985 Low Response' Tier 1 I :J 4 Scenario II Production with Decontrol but No Windfall Profit Tax' Total" Tier 1 Tier 2 Tier 3 Total" 2,046 500 97 3,113 U) U) (.3) 1985 730 536 907 2,748 891 632 1,530 3,628 1990 318 637 1,035 2,464 396 766 1,611 3,248 Medium Response' 1979 (Actual) 2,046 500 97 3,113 U) U) (.5) 1985 730 536 907 2,748 891 632 2,094 4,192 1990 318 637 1,035 2,464 396 766 2,112 3,749 High Response' 1979 (Actual) 2,046 500 97 3,113 U) U) (.8) 1985 730 536 907 2,748 891 632 2,993 5,091 1990 318 637 1,035 2,464 396 766 2,912 4,549 1979 (Actual) 2 Scenario I Production with Continued Price Tier 2 Tier 3 In millions of barrels. Low, medium and high responses refer to the assumed elasticities of supply for various categories of crude. Tier 1 and Tier 2 oil, as well as heavy oil and incremental tertiary (enhanced) oil in Tier 3, were assumed to be oil from existing properties. Oil from new properties appears in Tier 3. In each response case, oil from existing properties (Tiers 1 and 2 and heavy oil and incremental tertiary oil in Tier 3) was assumed to have a price elasticity of .2. The elasticity for new properties (Tier 3) varied from .3 in the low case to .5 in the medium case and .8 in the high case. Assumes domestic oil prices behave as described in Table 4, Scenario I, with a continuation of the controls embodied in the Energy Policy and Conservation Act. Production estimates for the years 1985 and 1990 under this assumption are from the Congressional Budget Office study cited below, page 76. These production estimates were used to derive our Scenarios II and III. Assumes domestic oil prices behave as described in Table 4, Scenario II. Production was estimated on the basis of the same supply elasticities utilized in estimating producing losses under the Energy Policy and Conservation Act. 20 ing the inflation-plus-2-percent adjustment to the uncontrolled stripper price ($35 a barreD in the fourth quarter of 1980. (We use the same methodology, although a higher base price, as the Joint Committee on Taxation uses in estimating windfall-profit tax revenue.) 28 Under decontrol and the tax (Scenario III) we calculate the realized producer price by adding the extra after-tax adjusted by inflation (measured by the GNP deflator). The Tier 3 inflation adjustment equals the inflation rate plus 2 percent annually, as specified in the law. 27 Thus, in this scenario, prices for Tier 1 and Tier 2 production remain constant in real terms, while the price for Tier 3 production rises at a 2-percent real annual rate. Under decontrol (Scenario II) , we estimate the free market price by apply- and 1990, Under Three Alternative Policy Assumptions] Scenario III Percent Increase in Total Production Due to DecontroF r, Production with Decontrol and Windfall Profit Tax' Tier 1 Tier 2 Tier 3 Total" Percent Increase In Total Production After Tax' (.2) (.2) (.3) 32.0 790 580 1,314 3,259 18.6 31.8 348 697 1,416 2,936 19.2 (.2) (.2) (.5) 52.5 790 580 1,690 3,635 32.3 52.2 348 697 1,752 3,272 32.8 (.2) (.2) (.8) 85.3 790 580 2,243 4,188 52.4 84.6 348 697 2,247 3,767 53.7 Assumes domestic oil prices behave as described in Table 4, Scenario Ill. ,; Alaskan oil from proved reserves has been included in the production totals, but not in any tier. The author estimated this Alaskan production at 575 and 475 million barrels in 1985 and 1990, respectively, compared with 471 million barrels in 1979. 7 This refers to the amount (percent) by which total production without controls (Scenario ll) would exceed total production with continued controls (Scenario during the years 1985 and 1990, respectively. n , This refers to the amount (percent) by which total production without price controls but with the windfall profits tax (Scenario III) would exceed total production with continued controls (Scenario I) during the years 1985 and 1990, respectively. Source: Production estimates in Scenario I, Congressional Budget Office, The Windfall Profits Tax: A Comparative Ana(ysis of Two Bills. 1979. All other estimates by the author. 21 revenue per barrel to the estimated price under continued controls (Appendix A). With these assumptions, the nominal free market price rises from $35/barrel during 1980-IV to about $54/barrel by 1985 and $80/ barrel by 1990 (Table 4). After-tax producer prices thus range (depending on tier) from 55 to 84 percent of the uncontrolled price by 1985, and from 52 to 84 percent of the uncontrolled price by 1990. With these price estimates, and with Congressional Budget Office estimates of production under continued controls, we derive (Scenario II and III) production estimates by applying appropriate supply elasticities, just as we did in estimating the effects of the Energy Policy and Conservation Act. 29 The use of nominal prices to estimate production losses is justified by the use of the same inflation adjustment for both controlled and uncontrolled prices, so that inflation has a neutral effect. As seen in Table 5, domestic oil production undoubtedly would have continued to trend downward over the 1979-90 period under continued price controls. But all three sets of assumed elasticities suggest that production is likely to rise substantially between 1979 and 1985 under decontrol. The Congressional Budget Office's forecast for production under continued controls shows total production dropping from 3,113 million barrels in 1979 to 2,748 million barrels by 1985, and then to 2,464 million barrels by 1990. Under the "low response" set of elasticities, total production with decontrol and no tax would reach 3,628 million barrels in 1985 and 3,248 million barrels in 1990. With the tax, production could still reach 3,255 million barrels in 1985 and 2,936 million barrels in 1990. Production figures in the "high response" case would be considerably higher, but as already indicated, the elasticity figures involved appear to be unrealistically high. Here again, the production differentials between the scenarios would not necessarily occur specifically in 1985 and 1990. Rather, the differentials represent the ultimate production responses to the estimated price differentials existing in those years as a result of the tax. Domestic production probably would rise Table 6: Estimated Domestic Under Three Alternative Domestic Oil PriCE Scenario I Total Production with Continued Controls" 2748 Total Production With Decontrol but No Windfall Profit Tax 3790 Percent Increase in Production With Decontrol but No Windfall Profit Tax 37.9 2748 3962 44.2 +2 2748 4192 52.5 +5 2748 4555 65.8 +10 2748 5209 89.6 Annual Percent Increase in the Decontrolled Domestic Price in Real Terms ~2 o l Annual production in millions of barrels. 2 Medium response assumes an elasticity of.2 for tiers 1 and 2, respectively, and also for heavy oil and incremental tertiary oil in tier 3. It assumes an elasticity of .5 for newly discovered oil in tier 3. 22 Unfortunately, increased drilling activity has not been translated into an increase in proved crude-oil reserves. In 1979, total reserves continued a decline that began in 1968. But gross additions to reserves amounted to 2.2 billion barrels, the highest figure since 1971. This meant a narrowing of the deficit between the gross volume added to reserves and the amount extracted. The inventory of proved reserves dropped by only 0.8 billion barrels the smallest amount since 1968. In any event, our analysis indicates that the windfall-profit tax would reduce the production response expected from decontrol. As a measure to reduce dependence on foreign-oil imports and improve the allocation of resources, decontrol is a step in the right direction. But by the same token, the windfall-profit tax is a step in the wrong direction. Perhaps policymakers should alter the tax to make it a true tax on profits rather than an excise tax on a portion of the selling price. In that way, it would not affect production decisions at the margin. even faster than projected, under all three sets of elasticity assumptions, if the uncontrolled price rose at more than 2 percent above the inflation rate (Table 6). For example, in the medium-response situation, an increase from 2 percent to 5 percent in the real rate of oilprice rise would mean an increase, from 53 percent to 66 percent, in the production differential achieved by 1985 because of the shift from controlled to uncontrolled prices. With the windfall profit tax in effect, the production differential would be 32 percent with a 2-percent real increase in prices, and 41 percent with a 5-percent real rate of increase. The higher prices being realized by producers as a result of decontrol apparently are now exerting a dramatic impact on exploration and development activity. Drilling activity set a new record in 1980, surpassing the previous highs reached in the mid-1950's. During the year, the industry drilled nearly 65,000 wells of all types - about 26 percent more than during 1979. For the year, total footage drilled rose 23 percent, while the number of drilling rigs in operation rose 32 percent. New records appear in prospect for 1981. Crude Oil Production, 1985, Assumptions l and with Medium Response 2 -'S=cc'-'e=nac.'-'rio~II"_I . Total Production With Decontrol and Windfall Profit Tax 3379 _ Percent Increase in Production With Decontrol and Windfall Profit Tax 23.0 Amount of Production Lost Due to the Tax as a Percent of Production with Decontrol and No Windfall Profit Tax' 10.8 3487 26.9 12.0 3635 32.3 13.3 3871 40.9 15.0 4301 56.5 17.4 , Production under this scenario depends upon the controlled price, and therefore is not influenced by alternative assumptions with regard to the decontrolled price. 4 Actual production losses under each price assumption are as follows: -2%, 411; 0%, 475; + 10%,908. 23 + 2%, 557; + 5%, 684; 24 V. Summary and Conclusions Federal price-control programs in effect throughout most of the 1970's held the average domestic producer price of crude oil below the world market level. By permitting refiners to pay less than the world price for domestic crude, Congress attempted to protect consumers from bearing the full impact of rising world prices. There is considerable debate as to whether that objective was achieved. Moreover, controls affected the supply side of the u.s. market, both by creating greater dependence on foreign oil and by leading to an inefficient allocation of resources. In this respect, the Energy Policy and Conservation Act - as well as earlier control programs reduced domestic production substantially below the level that would have prevailed without controls. Efficiency in satisfying any given level of national consumption requires that domestic producers expand crude-oil production to the point where the cost of the last barrel equals the cost of acquiring an additional barrel of foreign crude. Federal controls violated that condition, by holding the domestic selling price below the landed price of foreign oil, and thereby causing the industry to produce at less than the optimum production level. For every barrel not produced, the nation's dependence on foreign oil rose by an equivalent amount at a greater cost of resources. The removal of price controls (January 28, 1981) has forced refiners to pay free-market prices for domestic crude oil. That means higher refined-product prices also, to the extent that refiners pass on those higher costs to consumers. But decontrol also should raise domestic production above the level that would prevail under continued controls. In fact, decontrol may bring about at least a temporary reversal in the production decline of the past decade. But the windfall-profit tax will reduce that positive impact. For example, with supply elasticities of .2 and .3 for existing and new properties, respectively, and with a 2-percent Appendix A Windfall Profit Tax (WPT) Calculation, 19851 Tier III Estimated Decontrolled Price' Less Adjusted 1979 Base Price' Windfall Profit Before Severance Less Severance Tax' Windfall Profit Windfall Tax Heavy Oil 42.38 -28.23 All Other TIer III Oil 53.93 -28.23 Tier I 53.93 -19.92 Tier II 53.93 -23.64 34.01 - 1.84 30.29 1.64 14.15 .76 25.70 - 1.39 32.17 - 22.52 28.65 -1719 13.39 - 402 24.31 - 7.29 9.65 11.46 9.37 17.02 19.92 23.64 + 965 + 11.46 28.23 + 9.37 28.23 +17.02 29.57 35.10 37.60 45.25 Amount of Windfall Retained by Producers Producer Realized Prices with Decontrol after WPT Adjusted Base Price Amount of Windfall Retained by Producers Producer Price with Decontrol after WPT I All data in dollars per barrel. , Price assumed to rise at the inflation rate plus 2 percent annually between 1980 and 1985. 1 Base price adjusted upward over the 1980-85 period according to the method described in the law. , Tax imposed by the states, assumed 10 average 5.4 percent of the windfall profit before severance. Source: All computations by aUlhor. 25 real rate of price increase, decontrolled production would be 31 percent higher in 1985, and 30 percent higher in 1990, than the amounts that might be produced with continued controls. But the windfall-profit tax could reduce that production differential to about 18 percent in 1985 and 1990 respectively. And the more the decontrolled price of oil rises relative to prices of other goods and services, the greater will be the production losses attributable to the tax. FOOTNOTES 1, The term Ilproperty" is used as defined in the rather than the price actually paid. That approach made it possible to increase the price of stripper-well oil without lowering the price of some other category of domestic oil. See Montgomery (1977, pp. 9 and 12) for a discussion of this point as well as the inflation adjustment. Energy Policy and Conservation Act to mean a separate and distinct producing reservoir. (U.S. Department of Energy, Monthly Energy Review, January 1981, p. 99). A "reservoir" is a porous and permeable underground formation containing an individual and separate natural accumulation of producible oil or of oil and natural gas. In most situations, reservoirs are classified as oil or gas reservoirs by a regulatory agency. See American Petroleum Institute (1976, p. 7). 11. Economists generally agree that controls held the average domestic producer selling price below the world market level. Economists also agree that the so-called entitlement program equalized the average cost of crude oil to each refi ner regardless of the relative amounts of imported, lower-tier, or upper-tier imputs used - and that refiners paid a common average price for all oil that was below the world market level. See, for example, Cox and Wright (1978, p. 4) and Montgomery (1977, p. 37). There is widespread disagreement, however, about whether crude-oil price controls reduced refined-product prices below the level that would have prevailed without controls. Montgomery has argued that the entitlement program utilized competitive market forces to pass through the increased refiners' profits from price-controlled crude oil, from crude-oil producers to refined product consumers (1977, pp. 37 -40). Phelps and Smith (1977) maintain that refined-product prices were not held down by the controls and entitlements, and profits were transferred from producers to refiners. They argue that world refined products are made from world crude. The U.S. imports refined products and therefore, U.S. refined product prices must reflect world crude prices. This is basically an empirical question, but with much conflicing evidence. See, for example, Deacon (1978). 2. In this regard, crude-oil production differs from most manufacturing processes. With the latter, we assume that increasing capacity, (i.e., rate of attainable production), causes something like a proportionate increase in the amount of product that ultimately will be forthcoming. Crude-oil investments, however, must be described with reference to two dimensions: the rate of output to be achieved, and the total volume of crude available for ultimate production. For a discussion of this point, see Bradley (1967, p. 16). 3. American Petroieum Institute, American Gas Association, Canadian Petroleum Association (1980, p. 14). 4. See, for example, Duchesneau (1975) and Eppen (1975) 5. Herfindahl and Kneese (1974, p. 123). 6. This would be the average delivered price for imported oil at the refinery gate, including transportation costs. In actuality, average domestic producer prices probably would not exactly equal the average landed cost of imported oil because of quality differences. Crude oil is not a homogeneous commodity; viscosity, sulfur content and other characteristics vary and affect its value. Nevertheless, the price of imported oil would determine domestic prices in the manner described in the text. 12. The following analysis synthesizes an even more detailed analysis of the production effects of the Energy Policy and Conservation Act, made by Kall (1980, pp. 107-111). For an earlier discussion, see Roush (1976, pp. 16-20). 13. There are also resources wasted on the demand side as a a result of controls. The entitlement program reduced the price of imported oil to a common average price for domestic and imported oil (P r) which was below the world market price (P w)' In doing so, it raised the quantity of imported oil demanded above the quantity that would have been demanded had refiners been required to pay the world price for an incremental barrel of crude. The additional oil demanded had an incremental value to the economy of Pro But to realize that value, the nation paid the world price, Pw' to foreign sellers of crude. The resources consequently wasted on the demand-side equalled the difference between the world price and the average refiner acquisition price for both foreign and domestic oil, times the additional quantity 7. The requirements for efficiency on the supply and demand sides of the U.S. crude-oil market, as well as the inefficiencies created by price controls, are discussed in detail by Arrow and Kalt (1979, pp. 9-27). Their work draws upon an earlier study by Roush (1976). 8. For a detailed discussion of these state programs, see McDonald (1971, pp. 29-55) and Bohi and Russell (1978, pp. 250-253). 9. For a description of the various Federal crude-oil price-control programs of the late 1970's, see MacAvoy (1977) and Montgomery (1977 and 1978). 10. In computing the average target price, the energy agency assigned stripper-well oil an upper-tier price 26 demanded as a result of the entitlement program. We should note that the average refiner acquisition price for domestic and imported oil, Pr , under the entitlement program was between the world price, Pw ' and the average controlled domestic price, Pc as shown in Chart 2. That price was not shown on the chart because it did not affect the domestic production of crude oil. 22. In its 1979 Annual Report to Congress, the U.S. Department of Energy forecast production for 1985 and 1990 under several categories that would correspond to existing properties, the most responsive being production from enhanced oil-recovery techniques. The imputed price elasticity for existing properties derivable from these forecasts is approximately.2. 14. If marginal profit is increasing like compound interest, an owner of a reservoir will be indifferent at the margin between extracting and holding at every instant of time. Hotelling (1931) established the profit-maximizing condition for a firm managing a depletable resource. For a further discussion of the point see Solow (1974, pp. 1-6). 23. That assumption was employed in most models of reserve additions. See Bohi and Russell (1978, p. 237) for a discussion of this point. 24. For example, Fisher (1964) estimated an eillsticity for new-oil discoveries of .3 using data for 1946-55, but Erickson and Spann (1971) obtained an estimate of .8 using 1946-59 data. The U.S. Department of Energy, in its 1979 Annual Report to Congress, developed forecasts for 1985 and 1990 which imply an elasticity for new fields of around .3. For a survey of these and other models, see Kimmel (1977). 15. Kalt found that, on balance, controls tended "to encourage later rather than earlier extraction." See Kalt (1980, p. 132). 16. As indicated in footnote 11, some economists argued that refined-product prices already reflected world oil prices. They maintained that decontrolling domestic crude-oil prices would have no effect on refined-product prices. 25. Estimates vary widely concerning the total amount of oil remaining to be discovered in the United States, both on and offshore. For example, one official source places the total undiscovered recoverable resource base at somewhere between 50 and 127 billion barrels; see U.S. Geological Survey (1975, p. 4). Another recent assessment places the estimate at between 14 and 32 billion barrels; see Nehring (1981, p.175). 17. This estimate was based on the assumption that the uncontrolled domestic-producer price rose at the inflation rate plus 2 percent - i.e., at a 2 percent real annual rate - over the 1980-90 period. Reported in "U.S. Windfall Tax Bonanza Based on $75 Oil Price in 1990" (1980, p. 3) 26. This is the same price assumption employed by the Joint Committee on Taxation in developing its 1979 estimates of the Federal revenues to be derived from the windfall-profit tax. The price assumption is used to analyze the production effects of decontrol, with and without the windfall-profit tax. Note that there would be no tax unless the uncontrolled domestic price remains above the adjusted base price. 18. See assumption described in footnote 17. 19. The projected decline rate ("base level") is the average daily amount of oil removed from the property during the six-month period ended March 31, 1979, reduced by one percent per month (after 1978) for each month before the project-beginning date. See Price Waterhouse and Company (1980, p. 21). 20. There would be no "windfall" upon which to base the tax unless this condition prevailed. 27. Tier 3 encompases newly discovered oil, heavy oil and incremental tertiary oil. It receives preferential treatment in the law through a lower tax rate and an extra 2-percent annual increase in its base-adjusted price. 21. Since controls were imposed on the selling price, average selling prices of. lower- and upper-tier crude in the absence of controls would have risen to approximately the landed price for imported oil (before entitlements). The uncontrolled wellhead price would have roughly equalled the world price minus the average transportation costs incurred by domestic producers in supplying refiners. For reasons of data availability we used actual and estimated uncontrolled wellhead prices, rather than seIling prices, in calCUlating possible production losses resulting from the Energy Policy and Conservation Act. 28. Their estimate, made in early 1979, underestimated the actual increase in uncontrolled prices that actually occurred by the latter part of that year. 29. For estimates of production under continued controls, see U.S. Congress, Congressional Budget Office (1979, p. 76). REFERENCES American Petroleum I nstitute. Standard Definitions for Petroleum Statistics. Technical Report No.1. Washington, D.C.: American Petroleum Institute, 1976. Natural Gas in the United States and Canada as of December 31,1979. Washington, D.C.: American Petroleum Institute, 1980. Arrow, Kenneth J. and Kalt, Joseph P. Petroleum Price Regulation: Should We Decontrol? Washington, D.C.: American Enterprise Institute for Public Policy Research, 1979. American Petroleum Institute, American Gas Association, Canadian Petroleum Association. Reserves of Crude Oil, Natural Gas liquids and 27 Bohi, Douglas R. and Russell, Milton. limiting Oil Imports. Baltimore: Johns Hopkins University Press, 1978. Force. Washington, D.C.: American Enterprise Institute for Public Policy Research, 1977. Montgomery, W. David. "A Case Study of Regulatory Programs of the Federal Energy Administration," Study on Federal Regulations. U.S. Senate, Committee on Government Affairs. Washington, D.C.: U.S. Government Printing Office, 1978, pages 733-833. Bradley, Paul G. The Economics of Crude Petroleum Production. Amsterdam: North-Holland Publishing Company, 1967. Cox, James C. and Wright, Arthur W. "The Effects of Crude Oil Price Controls, Entitlements and Taxes on Refined Product Prices and Energy Independence," land Economics. February 1978, pages 1-15. __, The Transition to Uncontrolled Crude Oil Prices. Prepared for the National Bureau of Economic Research Conference on Public Regulation, Washington, D.C., December 15-17, 1977. Pasadena: California Institute of Technology, 1977. Deacon, Robert T. "An Economic Analysis of Gasoline Price Controls," Natural Resources Journal. October, 1978, pages 801-814. Nehring, Richard and Van Driest, E. Reginald. The Discovery of Significant Oil and Gas Fields in the United States. Santa Monica: Rand Corporation, 1981. Duchesneau, Thomas D. Competition in the U.S. Energy Industry. Cambridge: Ford Foundation, 1975. Eppen, Gary. Energy: The Policy Issues. Chicago: University of Chicago Press, 1975. Phelps, Charles E. and Smith, Rodney T. Petroleum Regulation: The False Dilemma of Decontrol. Santa Monica: Rand Corporation, 1977. Erickson, Edward W. and Spann, Robert M. "Supply Response in a Regulated Industry: The Case of Natural Gas," Bell Journal of Economics and Management Science. Spring, 1971, pages 94121. Price Waterhouse and Company. Windfall Profix Tax. New York: Price Waterhouse and Company, 1980. Roush, Calvin T., Jr. Effects of Federal Price and Allocations Regulations on the Petroleum Industry. U.S. Federal Trade Commission Staff Report, 1976. Fisher, Franklin M. Supply and Costs in the U.S. Petroleum Industry. Baltimore: John Hopkins Press, 1964. Haiimark, Fred O. Heavy Oil in California. Sacramento: California Division of Oil and Gas, 1980. (Unpublished memo) Solow, Robert M. "'The Economics of Resources or the Resources of Economics," The American Economic Review, Papers and Proceedings. May 1974, pages 1-14. Herfindahl, Orris C. and Kneese, Allan V. Economic Theory of Natural Resources. Columbus: Charles E. Merrill Publishing Company, 1974. U.S. Congress, Congressional Budget Office. The Windfall Profits Tax: A Comparative Analysis of Two Bills. Washington, D.C.: U.S. Government Printing Office, November 1979. Hotelling, Harold. "The Economics of Exhaustible Resources," Journal of Political Economy. April 1931, pages 133-75. U.S. Department of Energy, Energy Information Administration. Amwal Report to Congress 1979, Volume 3. Washington, D.C.: U.S. Government Printing Office, 1980. Kalt, Joseph. Federal Regulation of Petroleum Prices: A Case Study of the Theory of Regulation. Unpublished Ph.D. dissertation, University of California, Los Angeles, 1980. U.S. Department of Energy. Monthly Energy Review. Washington, D.C.: U.S. Government Printing Office. Kimmel, David. The Price-Responsiveness of Petroleum Supply: A literature Review. Washington, D.C.: American Petroleum Institute, 1977. U.S. Geological Survey. Geological Estimates of Undiscovered Recoverable Oil and Gas Resources in the United States. Geological Survey Circular 725 Washington, D.C.; June 1975. Lewin and Associates. Enhanced Oil Recovery Potential in the United States. Study prepared for the U.S. Congress, Office of Technology Assessment. Washington, D.C.; Office of Technology Assessment, 1978. "U.S. Windfall Tax Bonanza Based on $75 Oil Price in 1990," Petroleum Intelligence Weekly. March 10,1980, pages 3-6. MacAvoy, Paul W., ed. Federal Energy Administration Regulations. Report of the Presidential Task 28 Adrian W. Throop* Increasingly volatile financial markets have put a premium on accurate forecasts of interest rates. To the extent, however, that market participants base their decisions on available interest-rate forecasts, the value of these forecasts to an individual investor is diminished because security prices would already tend to reflect that information. At the extreme, the efficient-market hypothesis asserts that the market efficiently utilizes all available information in the pricing of securities, so that market participants generally can not profit from more accurate forecasts than those already incorporated in security prices. This article examines the degree to which the Treasury-bill market efficiently utilizes all available information so as to incorporate the best possible interest-rate forecast into current market prices. In other words, it evaluates the applicability of the efficient-market hypothesis to the Treasury-bill market. Two types of independent forecasts are examined: an autoregressive forecasting equation based on the past history of the bill rate, and the forecast of a selected panel of market professionals. Statistical tests are used to determine whether all useful information contained in these two forecasts is efficiently incorporated into Treasury-bill market prices. The market's forecast is derived from the "forward rate" implied by the term structure of yields. If the market is not efficient, a group of investors could improve their returns by altering the maturity of their investments in light of superior interest rate forecasts. Near a peak in the business cycle, for example, if such forecasts correctly foresee a larger decline in interest rates than anticipated by the market, then investors should buy securities with a maturity longer than their expected investment periods. The yields obtained by investors would be greater than those obtainable if they had chosen maturities equal to their investment period, due to larger capital gains than had been anticipated by the market. Alternatively, if the available forecasts correctly foresee interest rates falling more slowly than the market does, the investors utilizing those forecasts should buy maturities shorter than their investment periods. This is because investors would obtain a higher return from "rolling over" a series of short-term securities than from buying maturities equal to their investment periods. Whether or not investors can profit by "speculating" on interest rates, through holding other than their normally preferred maturities, depends critically on whether available interest-rate forecasts are more accurate than the market's. If the market is efficient, the information in these forecasts would already be incorporated into the prices of securities, and therefore nothing would be gained. In that case, investors would be better off by simply "hedging" their positions with maturities equal to their planned investment periods, thereby avoiding possible risk. This study shows that professional analysts' prediction of the Treasury-bill rate two quarters ahead are significantly more accurate than market predictions. This indicates that the market does not efficiently utilize all available information in making bill-rate forecasts. By 'Senior Economist. Federal Reserve Bank of San Francisco. Coleman Kendall provided research assistance for this article. 29 scribed and performed, on the basis of these three forecasts. Both the analysts' forecast and the forecast from the autoregressive forecasting equation are found to contain useful information which is not fully incorporated into the market's forecast indicating market inefficiency. Section III shows that the analysts' forecast contains information similar to that in the autoregressive forecasting equation, plus other useful information which is also not fully incorporated into the market's forecast. Investors could have traded on both types of information to improve their returns in the period examined. Section IV provides a summary and some further conclusions. making use of the information contained in the analysts' forecast, an investor in Treasury bills could have improved his return. We show that the analysts' ability to "beat the market" stems from a better utilization of information about past movements in the bill rate, and also from a more efficient utilization of other sorts of information. Section I explains the concept of market efficiency and examines the forecasting accuracy of the market, compared with that of both a panel of professional analysts and a simple autoregressive forecasting equation based on the past history of the bill rate. In Section II, statistical tests for market inefficiency are de- I. The Concept of Market Efficiency Efficient financial markets exist when the prices, or yields, of securities fully reflect all available information relevant for their valuation. In the case of riskless fixed-income securities, such as Treasury bills, the relevant information consists of expectations about the future course of interest rates. Investors then bid the prices of securities to the point where expected holding-period yields for securities of different maturities are roughly the same, given these expectations. For example, given the current six-month bill rate, the price and yield of a nine-month bill depends upon the expected three-month rate for six months ahead. New information can develop, but when it does it is rapidly reflected in revised expectations and in the prices of securities. Consequently, in an efficient Treasury-bill market, investors would not have significant opportunities for making profits on the basis of information about future interest rates l . The hypothesis of an efficient market is an extreme one, and therefore could not be expected to be literally true. Past tests of the efficient-market hypothesis have thus attempted to pinpoint the level of information at which the hypothesis breaks down. In these tests, all available information can be separated into three distinct types, or subsets, as shown in Table 1. The first subset consists of the past history of rates of return, or prices. A test of whether the market efficiently utilizes information in the past history of rates of return, or prices, is called a test of weak-form efficiency. The second information subset consists of any other information publicly available at little or no cost - such as government statistics on other relevant variables. Tests of whether the market efficiently utilizes this kind of information in securities pricing are called semistrong-form tests. The third information subset consists of information that is privileged or available only at significant cost. Tests of whether the market efficiently utilizes this kind of information, so that profits cannot be made from trading on it, are called strongform tests. Table 1 Types of Market Efficiency Subset of Information Test of Whether Particular Information Subset Is Efficiently Utilized by Market I. Past History of Prices, or Rates of Return Weak-Form Test 2. All Other Publicly Available Information Semistrong-Form Test 3. Privileged or Costly Information 30 Strong-Form Test of professional analysts, and compiled by the Goldsmith-Nagan Bond and Money Market Letter since Septem ber 1969. 6 The forecast period is again 1970-1 through 1979-III. This period begins with the first Goldsmith-Nagan sampling of professional forecasts and continues through the quarter just prior to the Federal Reserve's October 1979 shift in operating procedures, which emphasized controlling bank reserves for achieving its monetary objectives. We excluded later data on the ground that both the market and professional forecasters had to go through a learning experience which reduced the forecasting accuracy of each by perhaps differing amounts. Professional analysts use, either directly or indirectly, information in the past history of the bill rate - but they undoubtedly use other sources of information as well. So the analysts' forecasts contain information relevant to all three kinds of efficiency. However, it is not possible here to distinguish between a semistrong and strong test of efficiency. Some interest-rate forecasts in the Goldsmith-Nagan sample are not widely circulated, being privileged or costly to obtain, while other information may be publicly available. Since we cannot discriminate between these two types of information in the analysts' forecasts, we simply use the term "strong-form efficiency" to refer to efficient market use of all types of information besides the past history of the rate. In order to test the bill market's efficiency, we need a measure of the market's forecast of the 3-month bill rate for two quarters ahead. This can be obtained from the term structure of Treasury bill rates - specifically, from the market's two quarter ahead "forward rate." This is the interest rate on a 3-month Treasury bill two quarters ahead that would be required to equalize expected returns on 6- and 9month bills over a 9-month holding period. 7 The forward rate also contains a "liquidity premium," which compensates investors in the longer-term security for their sacrifice of liquidity. So an adjustment for that premium must be made to provide an estimate of the market's forecast of the bill rate. A n Appendix develops the concept of the forward rate more The three subsets of information exhaust the universe of all available information; thus, a market is fully efficient only if it passes all three kinds of tests. According to previous studies, the Treasury-bill market tends to be efficient in the weak-form sense of utilizing information in the past behavior of the bill rate. However, the evidence has generally been confined to very near-term market forecasts of only up to a few months ahead. 2 Also, little evidence is available on the bill market's efficiency in the strong or semistrong form sense of incorporating other available information useful for forecasting. 3 In this study, we consider two specific types of information that the Treasury-bill market could utilize in formulating a two-quarter ahead forecast of the 3-month bill rate. The first is simply an autoregressive forecast based on the past history of the bill rate. The yield on Treasury bills may vary somewhat predictably over time, due to the business cycle and/or predictable patterns in monetary policy. To pass the weak-form test, the market's forecast of the future bill rate needs to take into account any systematic behavior evidenced by its past history. 4 For the period 1951-IV through 1969-III, we estimated a simple autoregressive forecast that could have been used by the market. In this period, quarterly movements in the 3-month bill rate followed a significant autoregressive pattern, ie., past movements of the rate were significantly related to future movements. An equation explaining quarterly changes in the bill rate contained significant lags at 1, 2, 3, and 6 quarters, as well as a significant constant term indicating a positive time trend. We then obtained autoregressive forecasts for the period 1970- I through 1979-III by reestimating this equation on a growing sample of available observations and computing two quarter ahead forecasts from the latest coefficient estimates at each point in time. 5 Whether or not the market efficiently utilized the information contained in this autoregressive forecast is a test of weak-form efficiency. The second type of information is the average interest-rate forecast made by a panel 31 precisely, and details the technique used to estimate the liquidity premium. For a preliminary view of market efficiency, we can compare the market's two-quarter ahead forecast of the 3-month Treasury bill rate (F;':2) with the professional analysts' forecast (F~: 2) and the forecast from the autoregressive equation(F:l~2)' To measure forecast accuracy, we use the mean squared error (MSE), the arithmetic average of the squares of the forecast errors - or better still, the root mean squared error (RMSE), since it measures the error in the same units as the forecasted variable (Table 2) . As a baseline for comparison, we use the RMSE for a forecast of no change. After extraction of the estimated liquidity premium, the RMSE of the market's forecast, 1.24 percentage points, is only slightly lower than that for the forecast of no change - but substantially higher than the RMSEs of the analysts' and autoregressive forecasts. Moreover, the approach used to estimate the market's liquidity premium is more likely to have understated rather than overstated the true difference between forecast errors, as shown in the Appendix. These differences thus suggest the possibility of market inefficiency. If the estimated difference between the RMSEs of the autoregressive forecast and the market's forecast were statistically significant, then the condition of weak-form efficiency would not be met. An autoregressive forecast based on available past information on the Table 2 Accuracy of Forecasts Root Mean Squared Error (RMSE) 19701 - 1979111 (percentage points) Forecast of No Change (it) 1.25 Market's Forecast (F:"t-2) (Adjusted for Liquidity Premium as Estimated from Appendix Table 1, Eq. 3) 1.24 Analysts' Forecast (Fg n ) 1.10 Autoregressive Forecast (Ff't) t+2 .94 Treasury- bill yields would have provided a means for investors to "beat the market". By altering the maturity of their holdings in light of a superior forecast they could have improved their returns. The RMSE of the analysts' forecast is lower than the market's, but higher than that of the autoregressive forecast. Thus, the analysts may not have made full use of the available autoregressive information. Still, the analysts' forecast could contain other information besides that embodied in the past history of the rate. If an investor could have profited from trading on such other information, the market would also be inefficient in a strongform sense. 8 The next two sections provide tests of forecast accuracy that distinguish between strong-form and weak-form inefficiency. II. Evidence of Market Inefficiency Il1 We first test for weak-form inefficiency by examining whether the autoregressive forecast could have been systematically used to reduce the error in the market's forecast. This could have been done if the market's forecast error is at least partially explained by the difference between the autoregressive forecast and the market's forecast. Symbolically, this relationship is: i 1+ 2 - F:: 2 = B I (F;"t2 - F;':2) + e'+2' where i'+2 - F 2 = the market's forecast error for two quarters ahead; F;t~2 = the autoregressive forecast for quarter t+ 2 made at time t; F;': 2 = the market's forecast for quarter t+ 2 made at time t; (1) e t+ 2 = a random error term. 32 Note that this equation can be rearranged to give the optimal forecast'of the interest rate as a weighted average of the two forecasts. That is, it implies: i t +2 = BIF:~2 + (l - B I ) F:: 2+ e t +2 A similar test can be performed with the forecast of the Goldsmith-Nagan panel of analysts, F~:2' as the alternative to the market's forecast. When the difference between the analysts' and the market's forecast is used to explain the market's forecast error, we obtain the following estimated equation: (2) In the case of weak-form inefficiency, the autoregressive forecast would receive a significant weight. On the other hand, if there were no weak-form inefficiency, the weight of the autoregressive forecast would be insignificantly different from zero, and the weight of the market's forecast would be close to one. Thus, the test of weak-form inefficiency is that the estimated value of B I be significantly different from zero, so that the autoregressive forecast could have been used to improve upon the accuracy of the market's forecast. F:~2 and F~:2 are apt to be highly correlated in equation (2), tending to increase the standard errors of the estimated coefficients. So it is preferable to test for weak-form inefficiency by estimating equation (1); in addition, this equation constrains the weights to add up to unity. We estimated this and other equations using ordinary least squares, with a correction for a moving-average pattern of serial correlation in the error term. 9 The estimate of equation (l) is: S.E. = .893 The estimated value of the B I coefficient, at 1.15, is over six times its estimated standard error. It is clearly significantly different from zero and also not significantly different from one. Thus, only the analysts' forecast should be used in an optimal forecast combining both, because the market's forecast provides little or no additional information. This test also supports a finding of market inefficiency. Whether it is of the weak or strong form, or both, depends upon the type of information embodied in the analysts' forecast. If the analysts' forecast were based only on the historical behavior of the bill rate, only weak-form inefficiency would be confirmed. But if it contains other information as well, both weak and strong-form inefficiency would be indicated. Two additional tests can be made for the presence of strong-form inefficiency. The first explains the forecast error of the autoregressive equation by the difference between the analysts' forecast and the autoregressive forecast. The significant coefficient estimated in this equation, given below, indicates that the analysts' forecast can be used to explain an important part of the error in the autoregressive forecast. Therefore, that forecast contains other useful information besides the past history of rates. (3) (.163) 2 "R = .467 S.E. = .900 The value of B b at .664, is more than four times its estimated standard error, given in the parentheses, indicating that it is indeed significantly different from zero. Thus, the autoregressive forecast could have been used to reduce the market's forecast error, confirming the existence of weak-form inefficiency. In fact, in an optimal forecast combining the two, the autoregressive forecast would be given a larger weight (.664) than the market's forecast (.336). Also note that the unexplained forecast error has been reduced to .90 percentage points (equals the equation's standard error, S .EJ from 1.24 percentage points (equals RMSE in Table 2). i'+2 - F:~2 = .685 (F~:2 - F;1~2) (5) (.187) "R 2 = .155 S.E. = .862 A second test for confirming the existence of strong-form inefficiency involves generaliz- 33 ing the form of the original test. That is, error in the market's forecast might be plained by both the difference between autoregressive and the market's forecast the difference between the analysts' and market's forecast. In symbols, the exthe and the i'+2 - F~:2 = B I (F;"1-2 - F~+2) (6) -I- Q I {pgn _ U2 \.1·1+2 \ than their standard errors and significantly different from zero at the 5-percent level, indicating both weak- and strong-form inefficiency. To summarize the results so far, we have found that two different types of information could have been used to improve upon the market's forecasts of the 3-month Treasury bill rate, as embodied in the forward rate. The significance of an autoregressive equation in explaining the market's forecast error indicates weak-form inefficiency, and the added significance of the analysts' forecast confirms strong-form inefficiency. By incorporating the information contained in an autoregressive model and the analysts' forecast, the RMSE for a two-quarter ahead forecast of the 3month Treasury bill rate could have been reduced from the market's 1.24 percentage points to .87 percentage points (equal to S.E. of equation (8). In fact, when all three forecasts are combined into an optimal forecast, as in equation (7), the weight attached to the market's forecast is not significantly different from zero, at .022 (equals 1 .657 - .321). Thus, the autoregressive forecast and the analysts' forecast contain all the useful information embodied in the market's forecast, plus other useful information as well. -'- ~ l+2JT C;t+2 or in the alternative form, i'+2 = BIF;I~2 + (l + B2F~:2 - BI - (7) B 2) F:: 2 + e'+2 If both B I and B 2 were significantly different from zero, then both weak and strong-form inefficiency would be confirmed. We again choose the first form of the equation for estimation, in order to reduce the extent of multicollinearity among the independent variables. The estimate of the equation in this form is: (8) (.337) + .321 (F:~2 - F:: 2) (.201) R2= .498 S.E. = .873 Both of the estimated coefficients are larger III. Information in the Analysts' Forecast used by the professional analysts was efficiently incorporated into the market's forecast. In fact, we shall see that the autoregressive information contained in the analysts' forecast is substantially the same as that in our autoregressive equation. To approach this question, we decomposed both the professional analysts' forecast and the market's forecast into the portion related to current and past bill rates (extrapolative component) and the remainder that is not so related (autonomous component) .10 The difference between the two forecasts can then be decomposed into the difference between The previous section tested whether information from a simple autoregressive model that could have been used to predict the Treasury-bill rate was fully incorporated into the market's forecast, as embodied in the forward rate. However, some judgment enters into the construction of even such a simple au toregressi ve forecast. Specifically, if the time pattern of rate movements is unstable, the forecasting power of the estimated autoregressive equation could depend importantly upon the period of estimation chosen. Therefore, we should also test whether the autoregressive information that was actually 34 the extrapolative components of the two forecasts plus the difference between the autonomous components. If the difference between the extrapolative components were statistically significant in explaining the market's forecast error, this would indicate that autoregressive information actually incorporated into the analysts' forecasts was useful in predicting the interest rate, but nevertheless was not fully incorporated into the market's forecast - an instance of weak-form inefficiency. Similarly, if the difference between the two autonomous components were significant in explaining the market's forecast error, this would mean that the market did not incorporate other useful information available to the analysts - an instance of strong-form inefficiency. The extrapolative components of the two forecasts were estimated by regressing the difference between each forecast and the current rate on lagged quarterly differences in the bill rate. Such components can be divided conceptually into three parts (Figure 1). The first is simply the current interest rate, or a prediction of no change corresponding to a randomwalk hypothesis. If there were no significant coefficients in the above regression, the estimated extrapolative component would contain only the current interest rate. The second component is a time trend, or drift factor, indicated by a significant constant term in the above regression. The third component is the part of the forecast related to past changes in the interest rate, indicated by any significant coefficients on past changes in the rate. The best equation explaining the difference between the market's two-quarter ahead forecast of the bill rate and the current rate, chosen on the basis of a minimum standard error, contained no constant term and no lagged changes in the bill rate. Thus the extrapolative component of the market's forecast, E:: 2, is estimated to be simply a Figure 1 Elements of a Forecast Realized interest rate - - - - - -....."'. Forecast error Forecasted interest rate ---,-----_. ",' j Related to past changes !'" Autonomous component of forecast '" ." Time-Trend [ Current interest rate _ .1$! Extrapol ative component of forecast Forecast of no change 1 1+2 Time Note: For simplicity, this illustration assumes that elements of the forecast other than the current interest rate all contribute to reducing the forecast error. Of course, this need not be true in general. For example, the time trend, the part of the extrapolative component related to past changes, and the autonomous component could all be negative, instead of positive as assumed here. In that case, and given the same realized interest rate, the forecasted interest rate would be below, rather than above, a forecast of no change; and the forecast error would be increased, rather than reduced, by these three elements. 35 forecast of no change, equal to the current interest rate (i t). Symbolically, shows that the analysts made better use of information in the bill rate's past history than did the market. In addition, the significant coefficient on the difference between the two autonomous elements indicates that the analysts made superior use of other information besides past rates. The first finding confirms weak-form inefficiency, and the second substantiates a strong form of inefficiency. A final point of interest is whether the extrapolative component of the analysts' forecast utilizes available information on past bill-rate movements as efficiently as possible. To examine this question, we simply added to the right-hand side of equation (12) the difference between the forecast from the autoregressive equation and the market's forecast, and then reestimated the equation. This gives (9) In contrast, the extrapolative component of the analysts' two quarter-ahead forecast, E~:2' estimated in the same fashion, was found to be more complex. It contains a constant term, indicating a positive time trend in the interest rate of 54 basis points per year, and significant lags at 1, 3, 4, and 5 quarters on past changes in the bill rate. Specifically, the extrapolative component of the analysts' forecast is estimated to be: E~:2 = it + .270 - .196(it - it-I) + .186 (it-J + iH (10) ) .306(iH - i t - s ) + .200(i t • s - i H ) The autonomous components, denoted by 2 and A~:), are simply the difference be- A: tUloan tho t"ocn.a.r"ttUA fArAf"''lC'f -:lnrl i t +2- F:: 2 = .650 (A~:2 - A:) ite' p-vtr<.:lnf\lo:l_ ,.,,\.:1. ""'"p ....... ~~ tive element. It follows that the difference between the analysts' forecast and the market's forecast equals the sum of the differences between the autonomous and extrapolative components; or F~:2 - 1+2 (A~:2+ E~:2) - (A: 2 + + .766 (E~:2 - E: 2) + .323 (.584) + .323 (F:I~2- F~:2) (.215) (1 I) E:: R 2 = .484 2) E:':2) To test for both weak and strong-form inefficiency in terms of the autoregressive information actually used by analysts in the Goldsmith-Nagan survey, we substitute equation (1 I) into equation (4) and reestimate the latter. The resulting estimated equation is: i t + 2- F: 2= 1.10 (A~n~2 - A:: 2) (12) (.200) (,377) R2 = .464 S.E. = S.E. = .885 The estimated standard error of the equation is reduced somewhat by the addition of the difference between these two forecasts. However, neither the coefficient on this variable nor the coefficient on the difference between the extrapolative elements is significantly different from zero at the 5-percent level. Thus, neither the extrapolative component of the analysts' forecast nor the autoregressive equation's forecast significantly reduces the market's forecasting error once the other is already being utilized. Both contain a positive time trend and extrapolate from past changes in the bill rate, in contrast to the prediction of no change in the extrapolative element of the market's forecast. On the basis of this evidence, we conclude that the analysts have utilized the available information on past movements in the bill rate about as efficiently as possible in their forecast. = (A~:2 - A::) + (E~:2 - (13) (.348) L YV \,.1\,;1.1 Ll1'-' .l \,.IJP,",,"""-1 V ...... J. VI. ",,-,,"-,U'-'\. Ul.1. ..... l .. oJ .... .902 The significant coefficient on the difference between the two extrapolative components 36 These analysts also efficiently incorporated other publicly available information into the autonomous component of their forecast. As Friedman (1980) has shown, the forecasting errors of the Goldsmith-Nagan panel for shortterm interest rates have not been significantly related to costlessly available information on past values of macroeconomic series affecting the interest-rate outlook. These series include the unemployment rate, industrial production, price inflation, the money stock, and the Federal government's deficit. IV. Summary and Conclusions We have investigated whether the Treasurybill market has efficiently utilized all available information, so as to incorporate the best possible two-quarter ahead forecast of the 3month interest rate into its pricing of Treasury bills. A forecast by a panel of professional analysts was used as a measure of all available information, and the subset of information relating to the bill rate's past history was estimated by a simple autoregressive forecasting equation. We found that the analysts' forecast contained a similar extrapolative component, based on past movements in the bill rate. Either this extrapolative component of the analysts' forecast or a forecast from the autoregressive equation could have been used to reduce the error in the market's forecast significantly; but both contained substantially the same information. In addition, the analysts' forecast contained other useful information for explaining a portion of the market's forecast error. Altogether, the tests performed indicate the existence of both weak-form and strongform inefficiency. Earlier studies of the efficiency of the Treasury-bill market, focusing on shorter term interest-rate forecasts, have usually indicated weak-form efficiency. In contrast, our results for a two-quarter ahead forecast of the bill rate show the existence of weak-form inefficiency. This reflects the positive time trend found in both the extrapolative component of the analysts' forecast and the autoregressive forecasting equation, but not in the market's forecast. The market's forecast of the 3-month bill rate failed to incorporate the upward drift in the bill rate attributable to rising inflation in the forecast period, even though this drift could have been extrapolated from past data. In testing for a stronger form of efficiency in the bill market, earlier studies have generally used current and past values of relevant macroeconomic variables to measure other available information besides the past history of the bill rate. This study used the nonextrapolative component of the professional analysts' forecast for this purpose. We found that this component also contributed significantly to reducing the market's forecast error, indicating strong-form inefficiency as well. The difference between the overall forecast error of the analysts and the market in the 1970's was a modest 14 basis points, as measured by the root mean square error. Nevertheless, our results indicate that the above types of market inefficiency would have allowed an investor to trade on the information contained in the analysts' forecast. An investor could have improved his overall returns with a strategy of shortening maturities when the analysts' forecast was above the market's and lengthening them when the opposite was true. 37 Appendix Estimation of the Market's Forecast are generally unbiased, so that the realized interest rate is approximately equal to the anticipated interest rate plus a random error: The market's two-quarter ahead forecast of the 3-month Treasury bill rate is embodied in the differential between yields on 6- and 9month bills. To see this, first consider that market arbitrage makes the yield on a 9-month Treasury bill equal to the expected return on a 3-month Treasury bill that is rolled over twice, plus premiums for the sacrifice of liquidity. In algebraic terms, the yield on the 9-month Treasury bill is: (4) The market's forward rate (f) is composed of an anticipated interest rate and a liquidity premium (p), or: (5) Subtracting (4) from (5), we obtain the liquidity premium as equal to the difference between the market's forward rate and the realized interest rate plus the random (expectationaO error: 0) where i = yield at a quarterly rate, p = liquidity premium, left subscript = maturity of security in quarters, right subscript = time at which investment in security begins in quarters, superscript "e" = interest rate expected by the market as of time t. Similarly, for a 6-month Treasury bill: 0+ 2it)2= 0 + lit) 0 + li e t + l + IPt+l) (6) If the liquidity premium were constant, it could be estimated simply from the average difference between the forward rate and subsequently realized interest rate. But previous research indicates that liquidity premiums are in fact somewhat variable. Three not mutually exclusive hypotheses have received support in the literature. The first hypothesis views short-term securities as better substitutes for money balances than longer-term securities. If this is so, the liquidity premium would tend to be affected by the level of interest rates. When interest rates are high, the foregone interest income in holding money is larger; and so the public desires to exchange part of its money balances for securities. However, if the public prefers short to long securities in this exchange, prices of short-term securities would be bid up relative to those on longerterm ones. This would increase long-term interest rates relative to returns on short-term securities, or in other words increase the liquidity premium. Thus, if short-term securities are better substitutes for money than longer-term ones, liquidity premiums would vary positively with the level of interest rates. II (2) Dividing (0 by (2) and rearranging terms, we obtain equation (3) for the market's twoquarter ahead "forward rate." This is the expected return (including a liquidity premium) on a 3-month Treasury bill two quarters ahead that is required to bring about equality between the expected returns on 6and 9-month bills over a 9-month holding period. o+ 3i ) 3 To evaluate the accuracy and information content of the market's two-quarter ahead forecast, it is necessary to separate the liquidity premium from the forward rate. The approach used here assumes that market expectations 38 n, Two other hypotheses relate to the fact that liquidity premiums on longer-term securities compensate investors for exposure to the risk of capital losses. The greater the probable variation in interest rates, the larger would be such risk. If the variation in interest rates is expected to approximate its recent variance, liquidity premiums ought to vary positively with that variance. Moreover, at times when interest rates are abnormally low (high), there is a large likelihood of unanticipated increases (decreases) producing unanticipated capital losses (gains) for holders of longer-term securities. Therefore, liquidity premiums might also be expected to vary inversely with the height of interest rates relative to recently experienced levels. 12 These three hypotheses were tested with available data for the period 1963-IV through 1979-III (Table A.1). Following equation (6), we regressed the difference between the twoquarter ahead forward rate and the realized Treasury-bill rate on the current level of the Treasury-bill rate (j), the difference between its current level and a weighted average of its and the value over the previous 2 years (i standard deviation of the bill rate over the same period (SD).13 When no explanatory variables other than a constant term are included, the constant term is significantly pQ$itive - indicating a positive average liquidity premium of 54 basis points. The level of the bill rate (reflecting the effect of better substitutability between shortterm securities and money) and the deviation of the bill rate from its recent trend (proxying for the probability of unanticipated capital gains on long-term securities) show significant effects when entered together. So also does the moving standard deviation of the bill rate when entered alone. But when all three variables are entered together, only the standard deviation retains at least marginal significance. Also the equation containing only the moving standard deviation of the bill rate explains more of the variation in the liquidity premium than does the equation containing the other two variables. Mainly on statistical grounds, we chose this equation ((3) in Table A.1) for separating out the liquidity premium from the Table A.1. Estimation of liquidity Premium in the Two-Quarter Ahead Three-Month Forward Rate (percentage points) 19631V - 1979111 Constant (I) .540 (.134)** (2) -1.30 (.565)** (3) -1.56 (.565)** (4) -1.49 (.578)** (5) -1.49 (.582)** (6) -1.58 (.560)*- 3.54 (I.03)** D.W. 1.28* S.E. .148 1.67** 1.08 .162 (.0424)** .175 1.53** 1.07 .195 (.0611)** .169 1.62** 1.07 -.0746 (.126) .158 (.0431)** .166 1.62** 1.07 .545 (.675) .409 (.273) .165 1.51 ** 1.07 -.443 (.153)** -.0909 (.120) -.602 (.645) R2 .000 SO i-i 117 Note: Standard errors are in the parentheses .•• indicates a regression coefficient that is significantly different from zero at the 1- percent level on the basis of a single-tailed test; and' indicates significance at the 5-percent level. The same symbols on the Durbin-Watson statistic (OW.) indicate the absence of significant serial correlation in the residuals at the 5- and 1-percent levels, respectively. 39 expectational error, because the additional data allowed more of an ex ante estimation. A purely ex post fit of the forward rate's forecasting error against the liquidity-premium variables would have generated a larger bias. Even so, our estimate of the liquidity premium appears to have picked up some of the true expectational error in the market's forecast. This was tested for by removing from both the analysts' forecast and the market's forward rate the portion of the respective forecast error associated with the moving standard deviation of the bill rate (and a constant term) during 1971-1 thru 1979-IIl. These adjusted forecasts were obtained by regressing each forecast error on the moving standard deviation of the bill rate, and subtracting the values predicted by these regression equations from each respective forecast. The procedure removes a similar amount of true expectational error from both forecasts, but also an estimated liquidity premium from the market's forward rate. The RMSE's of these adjusted forecasts are 1.01 and 1.18 percentage points for the analysts' and the forward rate, respectively. The difference of 17 basis points exceeds the 14- basis-point difference between the RMSE's of the analysts' forecast and our estimate of the market's forecast, suggesting that our estimate of the latter tends to underestimate the size of its forecasting error. The procedure used for estimating the liquidity premium has thus tended to understate the difference in true expectational errors between the market's forecast and the two alternative forecasts. A possible bias in the opposite direction could result from the omission of an important variable for explaining the liquidity premium. However, we have already considered all such possibilities. On balance, it is more likely that our estimate of the market's liquidity premium has understated, rather than overstated, the true difference between the forecast errors of the market and those of alternative forecasts. forward rate to arrive at an estimate of the market's anticipated 3-month bill rate. Because this equation has a minimum standard error, the resulting estimate of the market's forecast has a minimum forecast error. However, equations incorporating the other variables did about as well; and our results are not particularly sensitive to this choice. The procedure used to estimate the liquidity premium assumes that the market's true expectational error, v, is uncorrelated with the variables used to model the liquidity premium. To the extent that this condition is not met, the estimated liquidity premium captures a portion of the market's true expectational error. This portion of the true expectational error would then be removed from the estimate of the market's forecast when the estimated liquidity premium is subtracted from the forward rate. As a simple example of the problem, if the true expectational error is positively biased, the procedure would overestimate the liquidity premium and correspondingly underestimate the size of the market's forecast errors. We investigated the seriousness of this problem by regressing the difference between the forecast of the Goldsmith-Nagan panel and the realized interest rate on the variables used in Table A.l to explain the liquidity premium. The estimated coefficients from these regressions carried the same signs as those in Table A.l and were frequently statistically significant, indicating that these variables are capable of proxying for a portion of the analysts' forecast errors. However, the size of these coefficients was generally considerably smaller than those in Table A.I., as would be true if these variables were also related to the market's liquidity premium. To minimize inclusion of expectational errors, we estimated the liquidity-premium models on all available data going back to 1963, rather than just on the forecast period of 1971-1 thru 1979-111. This reduced contamination of the estimated liquidity premium with 40 FOOTNOTES 1. Tests of market efficiency were first applied to the stock market. Surveys of this literature are contained in Cootner (1964), Fama (1970), and Lorie and Hamilton (1973, Ch. 4). market prices. Any systematic pattern in expected short-period returns would quickly be eliminated as investors bid prices up or down in attempts to profit from them. In contrast, the returns on 3-month Treasury bills held to maturity cannot be affected by this kind of speculation, because the price at the end. of the 3 months is fixed contractually. Therefore, even a fully anticipated time pattern in 3-month Treasury bill yields is not likely to be arbitraged away in an efficient market. However, an efficient market would take such a pattern into account in its bill-rate forecasts. 2. The weak form of the efficient-market hypothesis has been tested in several ways in the bill market. One approach is to determine whether the forward rate applicable to any particular point in time follows a random walk. If the market's adjustment to new information is virtually instantaneous, as in an efficient market, successive changes in the forward rate applicable to any particular period should be random. This approach is followed in Shiller (1973) and Roll (1970). 5. The data sample was extended back to 1951-IV using Salomon Brothers, An Analytical Record of Yields and Yield Spreads. The best autoregressive equation for explaining quarterly changes in the 3month Treasury bill rate during 1951-IV thru 1969-111 was found to be: A second approach to testing the weak form is to determine whether the market reacts appropriately to any autocorrelation in the bill rate. While changes in the forward rate applicable to any particular time period should be random, it does not follow that for weak-form efficiency the spot rate should also follow a random walk. Indeed, in an efficient market the forward rate should extrapolate any systematic autocorrelation that tends to occur in the spot rate. Evidence of weak-form efficiency in market forecasts of up to a few months ahead is contained in Hamburger and Platt (1975), Fama (1975b), and Fildes and Fitzgerald (1980). Li t = .118 + .239 Li t _1 - .241 Li t _2 (.120) (.055) (1.19) - .126 Li t _3 - .338 Li t_6 (.124) (.122) f'l2=.193 S.E. = .449 DW.=1.93 Standard errors are indicated in the parenthesis. A two-step ahead forecast from this autoregressive model of quarterly changes was found to be considerably more accurate than a one-step ahead forecast from past changes over six-month periods. Hence, the two-step ahead forecast (with coefficients reestimated at each point in time) was used for Ff~2' Fama (1975a, 1977) takes a rather different approach by focusing on the relationship between the nominal Treasury-bill rate and the subsequently observed inflation rate. For weak-form efficiency, the nominal interest rate would summarize all the information about future inflation rates contained in the time series of past inflation rates, so long as the expected real return is constant. Fama argues that the Treasury-bill market is efficient in this sense, and also that expected real returns on Treasury bills are approximately constant. Carlson (1977), Joines (1977), and Nelson and Schwer! (1977) present contrary evidence on the constancy of the expected real rate, but their evidence with respect to weak-form efficiency is not conclusive. 6. The author wishes to thank Mr. Peter Nagan of The Goldsmith-Nagan Bond and Money Market Letter for permitting use of this data. 7. A large body of literature exists on the term structure of interest rates. Useful surveys are contained in Dodds and Ford (1974), Malkiel (1966), and Van Horne (1966). An alternative measure of the market's forecast can be obtained from yields on Treasury-bill futures contracts. We did not examine the accuracy of this type of forecast because a futures market in 3-month Treasury bills has existed only since January 1976, reducing the number of observations by more than half. For a comparison of yields on futures contracts and implied forward rates, see Lang and Rasche (1978) and Poole (1978). 3. Hamburger and Platt (1975) conduct a test of the semistrong form of efficiency by investigating whether variables other than current and past levels of interest rates are better than forward rates in predicting future bill rates. They considered such potential predictors as the current and past values of personal income and three alternative monetary aggregates, but found that the root-mean-squared error for such forecasts was no smaller than from a forecast using the forward rate, consistent with the semistrong form of efficiency. 8. Prell (1973) compared the accuracy of forecasts by the Goldsmith-Nagan panel of analysts with those of no change and an autoregressive model for 1970 through 1973. In that period, the RM8E of the Goldsmith-Nagan panel for a two-quarter ahead prediction of the 3-month bill rate was smaller than for both of these forecasts. However, for long-term bond yields, the panel's forecasts were no more accurate than a forecast of no change. Prell (1973) did not compare the accuracy of the analysts' forecast with those of the market implied by the forward rate, as would be necessary for a test of bill-market effi- 4. For the market to be considered efficient, the Treasury-bill rate does not necessarily have to follow a random walk, in which the change in the return from one period to the next is completely random. A random walk would be consistent with weak-form efficiency in stock and bond markets because shortperiod returns on these securities are dominated by capital gains or losses resulting from changes in 41 quarter. The tw o-quarter ahead 3-m onth forward rate, and the 3-m onth spot rate, were converted to a bonde q uiva le nt yie ld (as was the fo re c a s t by the G oldsm ith Nagan panel). The form ula for converting the T reasu ry-bill rate on a discount basis into its bond-equivalent yield is: ciency. For another study of the accuracy of interestrate forecasts by market professionals, see Fraser (1977). 9. A moving-average pattern follow s from overlap ping of forecasts, i.e. quarterly observations on twoquarter ahead forecasts. The Chi Square statistic indicated an absence of significant serial correlation after this correction. 10. This approach is sim ilar to one used in Diller (1969). '_ 11. Cagan (1969), Conard (1966, Ch. 7), Friedman (1 979) and Kessel (1 965) find evidence of a positive relationship between estim ated liquidity premiums and the level of interest rates, consistent with the th e o ry th a t s h o rt-te rm s e c u ritie s are c lo s e r substitutes for money than long-term securities. ~d m 360 1 - d-H5___ 360 365 m where i is the bill rate on a bond-equivalent yield basis, d is the bill rate on a discount basis, and m is the m aturity of the bill in days. From equation (3) in the Appendix, the tw o-quarter ahead forw ard rate on a bond-equivalent basis is seen to be: 12. Evidence for this is contained in Fildes and Fitzgerald (1980), M alkiel (1966, Ch. 7), Nelson (1972, Ch. 6) and Van Horne (1 966). Studies in dica t ing a positive relationship between liquidity premiums and recent variance in the interest rate include Fildes and Fitzgerald (1980), M cElhattan (1975), M odigliani and Shiller (1973), and Talley (1979). '3 ° t 1 ft+ 2 “ 1 - 2 dt In estim ating a liquidity premium for the Aaa corpor ate bond rate, M cElhattan (1975) found a response to the recent variation in inflation, in addition to the recent variance in the nominal interest rate. We experim ented with this form ulation, but could find no significant response of the liquidity premium in the tw o-quarter ahead forward rate to the variance in inflation, after allowing for the impact of the variance in the interest rate. 273 360 365 91 182 360 where 2dt and 3dt are the 6- and 9-m onth Treasurybill rates on a discount basis. Besides the current spot rate, the variables in the equation explaining the liquidity premium are the standard deviation of the spot rate about its mean over the past two years (SD), and the deviation of the current spot rate from a weighted average of its value over the same period (i - i). W eights in this average of past spot rates were declining, with: 1 3. We obtained interest-rate data from daily closing quotations tabulated by the Federal Reserve Bank of New York. The original data from this source con sisted of bid-side interest rates on a discount basis for 3, 6, and 9-m onth Treasury bills at the end of each ( i - i ) t= i t- S n=1 n_ it-n 28 REFERENCES Dodds, J.C. and Ford, J.L. Expectations, Uncertainty and the Term Structure of Interest Rates. New York: Barnes and Noble, 1974. Cagan, Philip. “ A Study of Liquidity Premiums on Federal and M unicipal Government S ecurities,” in Jack M. G uttentag and Phillip Cagan (eds.), Essay s on Interest Rates, Volume. I. New York: National Bureau of Economic Research, 1969. Fama, Eugene F. "E fficie n t Capital M arkets: A Review of Theory and Em pirical W ork,” The Journal of Finance, May 1 970, pp. 383-41 7. Carlson, John A. “ Short-Term Interest Rates as Pre dictors of Inflation: Comment,” American E co nomic Review, June 1977, pp. 4 6 9-7 5. ____“ Short-Term interest Rates as Predictors of Inflation,” American Economic Review, June 1975, pp. 26 9-8 2. Conard, Joseph W. The Behavior of Interest Rates, National Bureau of Econom ic Research, 1966. ____“ Forward Rates as Predictors of Future Spot Rates,” Journal of Financial Economics, July 1976, pp. 36 1 -7 7 . Cootner, Paul H., ed. The Random Character of Stock Market Prices. Cambridge: M.l.T. Press, 1964. ____“ Interest Rates and Inflation: The Message in the Entrails,” American Economic Review, June 1977, pp. 48 7 -9 6 . Diller, Stanley. “ E xpectations in the Term S tructure of Interest Rates,” in Jacob Mincer, (ed.), Economic F o re c a sts and E xp e ctatio n s: A n alyse s of Forecasting Behavior and Performance. New Fraser, Donad R. “ On the Accuracy and Usefulness of Interest Rate Forecasts,” Business Economics, September 1977, pp. 38 -44 . York: National Bureau of Econom ic Research, 1969, pp. 112-66. 42 Nelson, Charles. The Term Structure of Interest Rates. New York: Basic Books, Inc., 1972. Fildes, Robert A. and Fitzgerald, M. Desmond. “ E ffi ciency and Premiums in the Short-Term Money M arket,” Journal of Money, Credit and Banking, November 1980 (Part I), pp. 61 5 -2 9 . ___ and Schwert, G. W illiam. "S hort-Term Interest Rates as Predictors of Inflation: On Testing the Hypothesis that the Real Rate of Interest is Con stant,” American Economic Review, June 1977, pp. 4 7 8-8 6. Friedman, Benjamin M. “ Interest Rate Expectations Versus Forward Rates: Evidence from an Expec tations Survey,” The Journal of Finance, Sep tember 1979, pp. 96 5 -7 3 . ____“ Survey Evidence on the Rationality of Interest Rate E xpectations,” Journal of Monetary E co nomics, O ctober 1980, pp. 45 3 -6 5 . Poole, W illiam . “ Using T -B ill F utures to Gauge Interest Rate Expectations” , Federal Reserve Bank of San F rancisco, Econom ic Review, Spring 1 978, pp. 7-1 9. Hamburger, M ichael J. and Platt, E lliott N. “ The Expectations Hypothesis and the E fficiency of the Treasury Bill M arket,” Review of Economics and Statistics, May 1975, pp. 190-99. Prell, Michael J. “ How Well Do the Experts Forecast Interest R ates?” Federal Reserve Bank of K an sas City, Monthly Review, Septem ber-O ctober 1973, pp. 3-13. Joines, Douglas. “ Short Term Interest Rates as Pre dictors of Inflation: Comment,” American E co nomic Review, June 1977, pp. 467-77. Roll, Richard. The Behavior of Interest Rates: An Application of the Efficient Market Model to U.S. Treasury Bills. New York: Basic Books, Inc., 1970. Kessel, Reuben A. The C yclica l Behavior o f The Term S tru ctu re o f In te re s t Rates. New York: National Bureau of Economic Research, 1965. Sargent, Thomas J. “ Rational Expectations and the Term S tructure of Interest Rates,” Journal of Money, Credit, and Banking, February 1 972, pp. 74-97. Lang, Richard W. and Rasche, Robert H. “ A Com parison of Yields on Futures C ontracts and Implied Forward Rates,” Federal Reserve Bank of St. Louis, Review, December 1 978, pp. 21 -30. Shiller, Robert J. “ Rational Expectations and the Term S tructure of Interest Rates: A Comment,” Journal of Money, Credit, and Banking, August 1973, pp. 85 6 -6 0 . Lorie, James H. and Hamilton, Mary T. The Stock M arket: Theories and Evidence. C h icag o: Richard D. Irwin, Inc., 1973. Talley, Ronald J. “ M arket Interest Rate Expectations as Reflected in the Term S tructure,” Business Economics, September 1979, pp. 10-18. Malkiel, Burton G. The Term S tru c tu re of Interest Rates: Expectations and Behavior Patterns. Van Horne, James C. “ Interest Rate Risk and the Term S tructure of Interest Rates,” Journal of Po litical Economy, December 1966, pp. 6 2 9 -3 5 . Princeton: Princeton U niversity Press, 1966. M cElhattan, Rose. “ The Term S tructure of Interest R ates and In fla tio n U n c e rta in ty ,” F ederal Reserve Bank of San Francisco, Economic Review, December 1975, pp. 27-35. ____Financial Market Rates and Flows. New York: Prentice-Hall, 1 978. Modigliani, Franco and Shiller, Robert J. “ Inflation, Rational Expectations and the Term S tructure of Interest Rates,” Economica, February 1973, pp. 1 2-42. 43 Herbert Runyon* In published polls, inflation generally tops the list of problems troubling the public. Indeed, television news programs are full of discussions of the rampancy of inflation. But most audiences remain unaware of how to measure inflation - or of what it really is. Inflation may best be described as a substantial and continued rise in the general price level. However, the general price level is hard to define in practice and presents a number of measurement problems, since it is a single number that represents the average behavior of a great many prices during a given period of time. In a system of freely functioning markets, there can be a great deal of disparity in the movement of prices of individual commodities or services. This is to be expected, as shifting demand-and-supply conditions for specific items become reflected in their prices. As demand presses against supply for some goods, their market prices may rise relative to other goods. Of itself, this does not constitute inflation, because other prices may be falling - witness color television sets or hand-held calculators. Relative price movements of this sort are a normal manifestation of functioning markets for resources and final goods. But inflation exists only if the prices of most goods are rising, or if increases in the prices of some goods consistently outweigh declines in the prices of other goods. As inflation has accelerated - with consumer prices doubling over the last decade policymakers have attempted to offset its impact on living standards by indexing incomes to the cost of living. Workers, still active in the labor market, have tried to minimize their inflation-caused loss of economic welfare by negotiating cost-of-living *Research Officer, Federal Reserve Bank of San Francisco. Steve Kamin provided research assistance. 44 adjustments (COLAs) into bargaining agreements. Other groups now outside the labor market - such as the retired or disabled instead have depended upon the political process to ensure income maintenance. These two methods of adjusting to inflation - the market process and the political process - are not necessarily comparable, and consequently have sometimes produced inequities. The official measures of the cost of living playa significant role in public-policy decisions affecting both wages and income-maintenance programs. About two-fifths of the Federal budget consists of expenditures that are tied, or "indexed," to some such measure. As a result, inflation significantly affects the Federal budget. And in the private sector, COLAs are imbedded in most large union contracts. Altogether, about 80 million persons are affected by indexed payment of wages or nonwage benefits. Thus, policymakers must use an index that accurately reflects change in the price level of the pattern of consumer expenditures. The choice of a measure raises fiscal-policy questions. If the growth of the official index exceeds the increase in the actual expenditure pattern of individuals (designated here as the cost of living), real government expenditures will rise and contribute to a Treasury deficit. Moreover, if the index is upwardly biased, many wage and benefit recipients will be overcompensated. However, many cost-of-living adjustments (at least in government transfer programs) are based on the inflation rate in some earlier period (generally the preceding year). Hence, when inflation accelerates, the amount of actual overcompensation may be less than might appear; on the other hand, when inflation decelerates, the overcompensation increases. I. Measuring Changes in the Cost of Living base period. Suppose now that the prices of both commodities double, as in the move from P to P". In such a case we can say unequivocally that the cost of living has doubled, since clearly the cost of obtaining the same level of satisfaction has exactly doubled. But suppose the price of commodity 1 more than doubles while that of commodity 2 less than doubles, as in the movement of prices from P to P'. The total cost of the consumption bundle may have exactly doubled, but this does not mean a doubling of the cost of living as we have defined it. In fact, it will probably have less than doubled. The consumer can probably obtain his previous level of satisfaction at less than double the cost, by buying less of commodity 1 (whose price has more than doubled) and more of commodity 2 (whose price has less than doubled). However, we cannot tell precisely how much the cost of living has risen, because we do not know how much commodity-substitution the household will undertake in any particular price situation. Statisticians have developed index numbers to deal with the problem of measuring changes in the cost of living. Their calculations utilize the following quantities: Po = price ofa given commodity in the consumer's expenditure pattern in time period to qo = quantity of the given commodity purchased by the consumer in to PI = price of the commodity purchased in t l q I = quantity of the commodity purchased in t l . One approach - the Laspeyres index maintains a fixed composition of the pattern of goods and services consumed in the base period. (The index was developed by the I9thcentury French-German economist, Etienne LaspeyresJ Here, the amount spent on an individual commodity is represented by the price multiplied by the amount purchased, or Po qm for the base year. Hence total expenditures of consumers in the base period to are simply the sum of these products of price and quantity, or Po qo. In order to compute a Before attempting to measure the cost of living, we must first define what we wish to measure. The term "living" refers to individuals' consumption patterns - the sum of commodities and services they consume within a given period. We assume that consumers plan their expenditure patterns so as to achieve the maximum amount of pleasure or well-being within the income fixed for a given time period. I Hence an increase in the cost of living may occur when the income needed to secure a given level of satisfaction increases from one period to another. In principle, then, the change in the cost of living between two periods may be represented as the ratio of two incomes - with the denominator being the income in the first or "base" period, and with the numerator being the smallest income required in the second period to buy the group of commodities that affords the base period's level of satisfaction. 2 However, this concept is not amenable to direct measurement, since we cannot observe degrees of individual satisfaction, and hence cannot know whether an individual is maintaining the same level of satisfaction. 3 This basic problem may be illustrated by considering a simple situation in which a household consumes only two commodities with prices PI and P 2 (Figure 1) . In this illustration, the point P represents the level of prices of both commodities in the Figure 1 P" P 45 Laspeyres index for a later period, t j, we must calculate the quantities purchased in the base period to at the market prices prevailing in the later period tj, or pj qo. Total consumer expenditures are then represented by PI qo and the entire index by = :l,Pjqo L :l,Poqo while the denominator represents how much that bundle would have cost in the past period. How Do Indexes Differ? Because of their difference in approach, the two indexes may differ considerably as measures of the cost of living. The base-period consumption bundle is fixed for the Laspeyres index, so that it makes no allowance for substitution in the consumption pattern. However, as relative prices of goods change, consumers may find it advantageous to switch from higher-priced goods to lower-priced substitutes. The classic example is the relative price of beef and chicken. As the price of beef rises, the consumer can substitute chicken in his diet and expenditure pattern. But the Laspeyres index, being a base-period fixedweight index, assumes that consumers will continue to buy the same quantity of beef purchased in the base period rather than turn to chicken. With substitutions disallowed in the base-year consumption pattern, expenditures for the more expensive beef receive too great a weight and chicken too small a weight. Hence the index overstates the true increase in the cost of living. The Paasche index, being based upon the In this formula, the denominator represents the amount actually spent in the base period, while the numerator represents how much that same bundle of commodities would have cost in the second period. An alternative approach - the Paasche index - weights prices at current-period quantities. (This index was developed by Laspeyres' German contemporary, Hermann Paasche.) Where the Laspeyres index projects the consumer-expenditure pattern forward from the base period til the Paasche index projects the pattern backward from the current period t j to the past period to' The formula for the Paasche index is therefore = :l,Pjqj P :l,Poqj In this formula the numerator represents the amount actually spent in the current period, Figure 2 Paasche (B) Laspeyres (A) Expenditures Expenditures E Prices Prices 46 current consumption-expenditure pattern, does allow for substitution - but in a way that understates the rise in the cost of living (Figure 2). Consumers may substitute on the basis of changes in relative prices, and thus avoid part of the burden of generally rising prices, but in doing so they lose the satisfaction of consuming the more expensive good which they originally desired. In each panel of Figure 2, the curve EE' is the consumer's expenditure function, or the amount of money necessary to maintain the same level of satisfaction in periods to and t I, assuming unchanged tastes. The consumer faces different sets of relative prices in the two periods, with some prices rising, some falling and others remaining unchanged. However, on average, in a period of inflation, the total expenditure necessary to maintain the same standard of living (in terms of personal satisfaction) will rise. The curve EE' traces out the amount of expenditures needed at different times to maintain a given standard of living. In Panel A (Laspeyres index), the expenditure line EE denotes the total cost of the base-period consumer-expenditure pattern at the prices prevailing in period to (as expressed by the vector PoqJ and at the prices prevailing in t l (as expressed by the vector PIqoJ Although no allowance is made for substitution, the consumer will adjust his expenditures if faced with a different set of relative prices in period t i . He will substitute lower-priced for higher-priced goods in order to minimize total expenditures while maintaining the same level of satisfaction, thereby moving along expenditure function EE' rather than EE. The quantity PIq' I represents the expenditures required to maintain the same level of satisfaction in period t l as in period to' In panel B (Paasche index), we project the consumer-expenditure pattern PIqI in t l backwards along EE to period to' where the expenditure vector is PoqI' However, this ignores the substitutions made in the expenditure package between to and t j, which would make the actual expenditure line EE'. The quantity Poq' represents the expenditure which would have been required in period to to achieve the same level of satisfaction attained in period t I' Substitution makes a considerable difference between the measured and "real" change in the cost of living. In the case of the Laspeyres index, the measured change is the ratio of p Iqo to p oqo' If the consumer makes substitutions in his consumption pattern on the basis of changes in relative prices, the true change in the cost of living will reflect these substitutions and will be the ratio of PIqi to Poqo' The Laspeyres index thus overstates the change in the cost of living. Empirical studies of the substitution phenomenon suggest that the effect, while negligible with stable prices, becomes significant as the inflation rate rises. 4 The expenditure schedule in panel B can be considered in much the same way. Here again, the schedule EE' represents substitution in the expenditure pattern between to and t i . The Paasche index assumes that at base-period prices Pm consumers would have made expenditures Poq I' which is not what consumers would have spent given the change in relative prices. The pattern of q I consumption is heavily weighted with goods and services which were relatively expensive in the earlier period. Thus, by overstating base-period expenditures, the Paasche index understates the rise in the cost of living. 5 Durable Goods and Price Indexes When dealing with price indexes, we assume that the goods and services purchased during a specified time period are consumed in the course of that period. This is obvious in the case of nondurable goods and services, such as a hamburger, an opera performance or a haircut. None of these items can be used more than once. But problems occur with durable goods, such as houses, autos, furniture and appliances. By definition, such items provide services for at least three years, and some of them much longer. We expect a new house to provide shelter for up to 80 years, and an auto to provide transportation for (say) ten years. 6 However, the full cost of a durable good is picked up in the first period rather than being allocated over 0 47 its useful service life. The present value of an asset is determined by the value of the stream of services that it is expected to yield during its service life, as well as by the market rate of interest. For this reason, the inclusion of durable goods in a cost-of-li ving index distorts the pattern of actual consumption expenditures over a single period. By including the actual market price of such an asset in the index at the time of purchase, we greatly overstate the price of the services of this durable good for that period. Moreover, the value of durable goods II. belongs in the consumer's balance sheet as part of his stock of wealth, rather than in his consumption-expenditure pattern. To the extent that ownership of a home or other durable good generates capital gains, it effectively reduces the cost of the services yielded by that good. The effectiveness of any price index thus depends, to a great extent, upon the way that statisticians handle these two attributes of durable-goods purchases - the stream-of-services attribute and the investment-good attribute. Two Indexes of the Cost of Living quarterly over a IS-month period. For less expensive day-to-day purchases, the agency utilized diaries of actual expenditures kept over a t\:'1o-\veek period. These t\VO efforts were supplemented by a point-of-purchase survey conducted in 1974, and updated on a regular schedule. 7 BLS includes almost 400 categories of goods and services in its statistical market basket, pricing them on a monthly basis. Interviewers contact a sample of about 18,000 retail establishments, such as supermarkets, cleaning establishments, repair shops and professional offices. Questionnaires provide other data such as utility rates, transportation fares, and information not requiring personal visits and Federal agencies and private research organizations add further information. The PCE deflator is widely used as a cost-ofliving index, although it was not designed for that purpose. It results from the procedure used to deflate personal-consumption expenditure values into constant (I972) dollars, in order to obtain a measure of change in the physical volume of consumption. With its current-period reference weights, the PCE is a Paasche index 8 with the form: The most widely used price index, which has official sanction in the indexing of labor agreements and retirement benefits, is the Consumer Price Index (Cpn. The Bureau of Labor Statistics (BLS) has compiled this index ever since World War I, when it was developed to help determine wage rates in the shipbuilding industry. The other major index is the implicit price deflator for the personal-consumption expenditure (PCE) sector of the nationalincome accounts. The two indexes differ in several respects, such as population coverage. The CPI covers the expenditures of urban consumers, and represents about 80 percent of the population; the personal-consumption deflator covers "persons" as defined in the nationalincome accounts, chiefly individuals and nonprofit institutions. The indexes also differ in terms of items covered; the CPI regularly covers a selected list of about 400 items, while the PCE includes all goods and services currently consumed. The CPI is a straightforward Laspeyres index with the form CPI, = P Q1972-73 P oQ 1972-73 (3) BLS chose the base-year weights on the basis of a 1972-73 survey of expenditures by about 20,000 family units. For most durable-goods purchases, the agency utilized interview panels in which consumer units were interviewed PCE, = P,Q, (4) P 1972Q, The two indexes differ in commodity composition, as well as in population coverage and 48 statistical form. About three-quarters of the components of the CPI and the PCE indexes are comparable, largely because of the deliberate use of CPI components in generating the PCE index. Most of the differences occur in the components of homeownership, autos and allied services, and hospital charges and health insurance. The homeownership cost in the III. PCE is based upon the imputed rental cost of owner-occupied homes, whereas the comparable item in the CPI is based on home purchase prices and new-home mortgage rates. Both indexes treat durable goods on the basis of current purchase prices, rather than the cost of the stream of services which they yield. 9 Relative Performance of the CPI and peE above that of the PCE, a Paasche index. For the entire 20-year period, however, the average difference amounted to only 0.6 percentage points (Table 1). In the first half of the 1960's, when prices were reasonably stable, the CPI perversely showed the lower rate of change - but the difference was hardly significant, especially in view of the low rate of inflation prevailing at that time. Again, the two indexes showed much less correlation than would normally be expected, in view of the heavy use of CPI In comparing the historical performance of the CPI and PCE, we should keep in mind the conceptual differences of the indexes and the manner of their construction. Neither captures the "real" cost of living; rather they form an upper bound (Laspeyres) and a lower bound (Paasche) to the cost of living. This can be seen from an analysis of the past two decades, which contained one period of relative price stability and two episodes of high inflation (Chart 3). For most of this period, the rate of change in the CPI, a Laspeyres index, was Chart 1 Consumer Price Movements Annual Change (0/0) 14 12 Consumer price index ~ 10 8 6 ... 4 Personal consumption expenditure deflator 2 0 1960 1964 1972 1968 49 1976 1980 prices in the generation of the PCE. The much lower correlation in this period reflects the fact that both indexes moved within a fairly narrow range,given the .low level of inflation. Because the general price level was comparatively stable,. variations in relative. prices thus tended to cause greater variation about the general price level. The 1973-75 and 1978-80 episodes of inflation were different, both other and in relation to the earlier period of stability. As expected, the.CPImt>asured mOre inflation than thePCE in both episodes. But meanwhile, the inflation of 1973-75 was more general in scope than the 1978-80 episode. In 1973 -7 5, the U.S . economy felt the impact of world-wide inflation in the prices of internationally traded commodities, partly food but especially petroleum. Food and energy prices also contributed to the 1978-80 run-up in inflation, but the divergence in the two indexes in this later period largely reflected a sharp rise in the price of houses and in homemortgage rates (Chart 4). The divergence thus could be explained by the fact that the PCE incorporates only the CPI's relatively slow-rising rental component, rather than the CPI's fast-rising home-ownership component. measures (Table 2). The fixed-weight PCEis a Laspeyres index - like the CPI but with a 1972 expenditure pattern. At the same time, the fixed-weight PCE weights specific consumer items in the same way that the PCE does. Consider the weight, or relative importance, of three major PCE components on a fixedweight vs. a current-weight basis. In 1972, housing expenditures accounted for 15.3 percent of total consumer spending, but by 1980, this had increased to 17.6 percent of the total in terms of 1972 dollars. At the Same time, the weight of food in the index dropped from 20.5 percent in 1972 to 19.4 percent in 1980, while gasoline and oil dropped from 3.4 percent to 2.8 percent. Thus, in terms of 1980 consumption patterns, the fixed weight PCE underrepresented housing and over-represented food and gas and oil. Of course, the composition of weights can change in a currentweighted index as changes occur in relative prices, and in consumer tastes and income. Weighting differences were not significant in 1977-78, but they began to tell in 1979. Home prices and mortgage costs began to soar, leading to a spread of 1.9 percentage points between the CPI and the fixed-weight PCE (Table 2, line 4). But at the same time, a significant amount of substitution took place in the consumer-expenditure package because of sharp changes in relative prices, leading to a spread of 2.8 percentage points between the fixed-weight PCE and the current PCE (line 5). An increase of more than 50 percent in retail gasoline prices resulted in a decline of 11 percent in real purchases of gasoline and nearly 10 percent in real purchases of fuel oil Weighting Problem The differences in the CPI and the PCE are rooted in their basic conceptual natures, including differences in weights and the effects of substitution in the consumer-expenditure package. To understand these differences, we can compare the rates of change in CPI, the PCE, and also the fixed-weight PCE, which has some of the features of the other two Table 1 Average Annual Rates of Change in the CPI and PCE Indexes, and Coefficients of Determination between the Indexes (change in percent) 19601-19801V 19601-19651V CPI 5.29 1.35 9.59 11.81 PCE 4.69 1.50 8.34 9.35 .975 .667 19731-19751V .969 50 19781-19801V .947 and coal, not to mention an 11-percent cut in auto spending. As a result of these changes in current consumption, the composition of weights changed substantially between the current-weight and fixed-weight PCE. In 1980, the situation was reversed, with the largest difference occurring between the CPI and the fixed-weight PCE (line 4), reflecting the different treatment of homeownership. The homeownership component of the CPI rose by nearly 17 percent from the end of 1979 to the end of 1980, as compared with a 9-percent increase in the PCE in the same period. Because of this difference, and the heavier weighting of homeownership in the CPI, that single component accounted for roughly half of the difference in the overall indexes in 1980. The much smaller difference between the two PCE indexes was due to declines in real expenditures for food and gas and oil, which changed the composition of PCE expen ditures for the year. Homeownership Problem BLS’ problem with measurement of homeownership costs stems partly from its treat ment of home prices. The agency computes the weight from the purchase prices of homes bought in the survey year, minus the prices of homes sold in that year, plus transaction costs accompanying the purchase or sale. It then derives the index from price data supplied by the Federal Housing Administration. But the FHA sample is a small and unrepresentative segment of the market. The coverage is not geographically uniform; also, with its ceiling cutoff on mortgage loans, higher-priced homes are effectively eliminated from the sample. Problems also occur with the computation of mortgage cost. BLS assumes that the mortgage Chart 2 Rental, Homeownership and M ortgage Costs Annual Change (%) Year-over-year rate of change 51 borrower pays interest equal to the sum of all interest payments over the first half of the life of the mortgage, each year's interest cost being equal to the first. But this results in overcounting, in comparison with discounting the value of future house services (as reflected in the future stream of interest payments) to their present value. Moreover, many analysts suggest using the actual interest paid on all mortgages instead of the current new-home rate, to diminish the impact from sharp fluctuations in mortgage rates. The CPI's weighting of homeownership overstates its importance in the cost of living. Homeownership accounts for nearly onequarter of the total weight of the index. With that weight, it has five times the importance of the rental component. In the PCE, homeownership is only about 2 1/2 times as important as rental housing - a more reasonable figure, in view of the 2-to-l relationship of owners to renters in the national housing stock. In response to these and other criticisms, BLS is currently publishing five experimental measures of housing costs. These range from a rental index to various "shelter" and "asset" concepts. Conceptually, much can be said for a rental-equivalence measure, such as is used in the PCE index, since it directly measures the price of the services of a home. However, BLS does not at this time possess a true rentalequivalence sample; that is, one made up of housing units comparable in type and location to its sample of owner-occupied units. The rental index, as presently constituted, is based upon market observations of rental payments. The units involved in this sample generally differ from owner-occupied dwellings in such terms as age and income of inhabitants, as well as age, location and size of dwelling. Thus, the current CPI rental index cannot serve as proxy for the imputed rent of the different and larger population of owner-occupied units. The concentration of the rental sample in inner-city areas - many of them rent-controlled - also tends to bias the sample downwards, so that this is one of the slowest rising components of the entire CPI. Other considerations must be kept in mind when choosing an inflation index. The CPI in its present form substantially overestimates the rate of inflation - even more than might be expected from a Laspeyres index. However, a Laspeyres (Cpn index number seldom needs to be revised, being based on actual prices taken from primary sources and on fixed base-year quantities. On the other hand, the PCE index depends heavily on estimation procedures and thus is subject to frequent revisions, which can create problems of interpretation for policymakers. However, over the 1960-80 period, revised estimates of PCE rates of change were only slightly higher than the initial observations. Both the average rate of change and its variation were smaller for the PCE than for the CPI over this period. Table 2 Average Annual Rates of Change in the CPI, PCE and Fixed-Weight PCE Indexes (percent) 1960-80 1977 1978 1979 1980 mCPI 5.29 6.5 7.7 11.3 13.5 (2)PCE 4.69 5.8 7.3 8.5 9.0 (3)PCE (fixed weight) 47] 63 7.6 9.4 9.6 (4)CPI - PCE (fw) 0.58 0.2 0.1 1.9 39 (5)PCE (fw) - PCE 0.02 0.5 0.3 2.8 06 52 IV. Indexes and Public Policy The current Congressional debate on the Federal budget has focussed attention on the role of indexation in increasing expenditures. By linking increases in certain spending categories to changes in the cost of living, Congress originally sought to assure that benefit recipients would be able to cope with increases in the cost of living without further Congressional action. In other words, recipients of Federal payments would almost automatically preserve their "real" income. Payments to individuals account for more than half of Federal-budget expenditures, and 90 percent of such programs are indexed. Thus, nearly two-fifths of Federal budget outlays, whether paid out as wages or transfer payments, are linked to a price index (Table 3). In the case of social-security benefits, Congress ironically adopted an escalator approach as a means of capping the extraordinarily generous benefits it had adopted in the early 1970s. (Between 1970 and 1974, benefits increased by about 70 percent - just about double the increase in the cost of living,) The current indexation formula essentially provides for a benefit adjustment equal to the annual percentage increase in the CPI Cfirstquarter to first-quarter). The benefit formula thus called for a 14.3-percent increase in July 1980 and an 11.2-percent increase in July 1981. If the cost of living had been measured by the PCE instead, the increase in benefits would have amounted to 8.2 percent in 1980 and to about 9.4 percent in 1981. Congress recently has begun to consider several less costly indexing alternatives. One of these would delay the first payment of increased benefits from July to October each year. This lag would lessen the extent of overstatement of the previous year's inflation rate if inflation were accelerating, but it would worsen the overstatement if inflation were decelerating. Another suggestion would peg the increase in benefits to the increase in either the CPI or average wages, whichever is smaller. Still another suggestion would put an 85-percent "cap" or ceiling on the increase in the CPI. According to the General Accounting Office, inflation accounted for approximately half of the increase in expenditures for indexed programs over the 1970-77 period. The GAO study indicated that such spending increases automatically by $15-25 billion at a 10-percent inflation rate, and increases by $1.5 billion to $2.5 billion more for each additional percentage point of measured inflation. If the PCE index had been used in place of the CPI during the 1970-77 period, roughly 11l/2 percent ($12.5 billion) of the cumulative spending increase could have been saved. A Congressional Budget Office study argued that CPI-based indexation could account for three-fourths of a $200-billion increase in Federal payments to individuals projected for the 1980-85 period. But again, a shift from the CPI to the PCE index could mean savings of $11 billion through 1986 for the social-security program alone. 11 Table 3 Major Indexed Programs Directly indexed Social security Supplemental security income Railroad retirement Veterans' pensions Civilian retirement and disability Military retirement Black lung Food and nutrition assistance Subtotal, directly indexed As percent of total outlays Outlays, FY 1980 (billions) $117.1 6.4 4.7 3.6 14.7 11.9 1.8 13.3 $173.5 29.9% Indirectly indexed Medicare Medicaid Subtotal, indirectly indexed As percent of total outlays $ 35.0 14.0 $ 490 8.5% Total indexed programs Total outlays As percent of total outlays $222.5 38.4% Source: Office of Management and Budget, American Council of Life [nsurance 53 Compensation and Equity Table 4 Annual Change in the CPI, PCE and Average Wages* (percent) When indexation overcompensates for inflation, questions of equity arise. Federal payments to individuals are made only to certain individuals, while taxpayers who pay for benefits may be falling behind the increase in living costs. The cost-of-living adjustments (COLAs) that are written into many labor contracts are generally capped, so that workers fail to receive full compensation for their higher living costs (Table 4). And those whose income is not indexed at all may fall even further behind. Increases in social-security benefits have been tied to the CPI since 1975. From 1976 through 1978, the yearly increase in wages exceeded the annual indexed increase in social-sec uri ty benefi ts. However, the difference was smaller than indicated, because social-security benefits are not taxed as wages are. When annual CPI increases ran ahead of wage increases, from 1979 through 1981, the V. CPI PCE Average Wages 1975/76 6.4 5.7 7.2 1976/77 5.8 5.3 7.7 1977/78 6.6 6.3 7.7 1978/79 9.8 9.1 8.3 1979/80 14.3 8.2 8.4 1980/81 11.2 9.4 9.8 Annual Average 9.0 7.3 8.2 *First quarter to first quarter taxability factor made the gap even wider. In contrast, the PCE index - though rising more slowly than wages - would have placed wage earners and benefit recipients on a reasonably equitable basis during this period. Summary and Conclusions Many analysts have used the term "inflation" rather loosely in describing the recent sharp rises in prices of certain individual commodities, such as oil and food. Nonetheless, the inflationary process primarily involves an increase in the general price level. Within this context, the relative prices of individual goods may rise or fall according to market forces. Since numerous goods and services are bought and sold daily in the markets, an index number represents the only feasible method of describing the general movement of prices through time. An index number designed to measure inflation should give an accurate representation of changes in living costs. The true cost of living cannot be measured directly, since it is a matter of personal satisfaction or well-being. However, it can be measured indirectly by market observations as consumers reveal their individual preferences by purchasing certain goods and services. The Consumer Price Index has long been used as the common indicator of the cost of living, and thus as a basis for indexing cost-of-living adjustments. The CPI, as a Laspeyres index, uses quantities purchased in a certain base year as a reference point from which to measure changes in prices of a basic - presumably unchanging - consumer-expenditure pattern. The unchanging nature of the base-year consumption pattern ignores substitution of less expensive for more expensive goods in the reference expenditure package as relative prices change in succeeding periods. This imparts an upward bias to the index, as the increasingly expensive base-period expenditure pattern (in current prices) overstates what consumers actually bought and paid for at the checkout counter. The Personal Consumption Expenditure index (or deflator) has come to be used as an alternative index, although it was not specifically designed to measure changes in the 54 cost of living. The PCE, as a Paasche index, uses the current-period expenditure pattern as a reference base for comparison with expenditures in earlier periods. This approach thus allows for substitution of goods in the expenditure pattern. However, when that pattern is projected backward from the current year, the substitute goods of the current period may not have been as desirable to consumers in earlier periods. The consumer's loss of satisfaction, relative to the current period, thus causes a PCE (Paasche) index to understate changes in the "true" cost of living. As a practical matter, the CPI and the PCE were quite close in their measurement of living costs from 1960 through 1966. But from 1978 through 1980, the CPI rose at a much faster rate. This disparity resulted not so much from the indexes' different statistical composition as from their different treatment of sharply rising homeownership costs. The housing component of the CPI overstated the inflation in housing costs, because it included the full cost of a house, which includes its value as an asset, rather than merely the cost of shelter services. The treatment of mortgage interest also contributed to this overweighting of the costs of homeownership. The Federal government uses the CPI as a standard index in its efforts to offset the impact of inflation on benefits paid to individuals. But since individuals receive roughly half of all total budget outlays, and since 90 percent of these payments contain a cost-of-Iiving adjustment, the indexing formula should measure living costs as closely as possible. In the past several years, most evidence has suggested that the use of the CPI overcompensates benefit recipients. This overcompensation has raised two public-policy questions. First, overcompensation leads to unwarranted increases in Federal expenditures and in the Treasury deficit. Beyond that, it introduces inequities relative to those individuals not receiving indexed benefits, and thus amounts to an unintended redistribution of income. Indexed transfer payments are not taxable, and this widens the gap between benefit increases and wage increases. A comparison of index movements since the Federal government's widespread adoption of indexing suggests that the PCE index is a more equitable choice for determining cost-of-living adjustments. FOOTNOTES 1. Allen. RD.G., Index Numbers in Theory and Practice. Aldine: Chicago, 1975. 70 period. However, an examination of their results indicates a steady increase in the percentage error from 1964 (when the CPI rose by 1.2 percent) to 1970 (when it rose by 6.1 percent). For a class of 20 expenditure categories, the percentage error increased from 0.005 to 0.326, or at a compound annual rate of 100.6 percent - more than three times the annual rate of increase in the CPI in this period. Steven D. Braitwait, ("The Substitution Bias of the Laspeyres Price Index", The American Economic Review, March 1980, vol. 70, no. 1, p. 71), using 1958-70 data, found that the substitution bias increased significantly the longer that weights were held constant, rising from 0.36 for the 1958-63 period to 1.5 for the 1958-73 period. As Braitwait points out, relative prices change more in a high-inflation period than in a low-inflation period. His index of relative price dispersion increased from .11 in 1956-63 to 3.5 in 1958-73. And it is this dispersion of relative prices that affords greater opportunity for commodity substitution. Alan S. Blinder ("The Consumer Price Index and the Measurement of Recent Inflation", 2. Klein, Lawrence and Rubin, Alan, "A Constant Utility Index of the Cost of Living ", The Review of Economic Studies, 1949, p. 84. 3. Houthakker, Hendrik, "Compensated Changes in Quantities and Qualities Consumed", The Review of Economic Studies, Vol. XIX, no. 50, 1952-53, pp. 156159. 4. Consumer substitution with a Laspeyres index can make a difference between the measured cost of living and the "true" or perceived change in the cost of living as measured by an index. A number of studies have indicated that the amount of error involved increases rapidly with the rate of inflation. Nicholas N. Noe and George von Furstenburg, ("The Upward Bias in the Consumer Price Index due to Substitution", Journal of Political Economy, November/December 1972, Vol. 80, p. 1283) found the percentage error in the CPI due to substitution to be "trivial" in the 1963- 55 Brookings Papers on [;conomicActivity, 1geO:2,p. 545J,.found.thatdifferences in·weiQhting accounted foraboUlone-.half of the difference betweenthe CPI andPCEin.t979. Indeed,thepost- 1978 period,an almostcontinuous perioqofhighinflation, "ha~b~en one .of • tho~erare .timesinwh.ich. sUbstantial differences. between the Laspeyres and Paasche indexe.sareexpected to arise." 5. Blinder, Alan $., "The Consumer Price Index and the Measurement of Recenllnflation", Brookings Papers on Economic Activity, NO.2,igeO, p. 542. 6.• "Fixeq Nonresidential and Residential Gapital in the UnitedStates, 1925-1975", U.s. Depwtment of Commerce, Washington, June 1976, pp. T-6, T-7. not berlavi,or to period. Katz, Arnold J. and Value of Services Provided by the Durables, 1947-1977; An uooortumtv Measure", Survey of Current Business, 22 n. 7. U.S. Department of Labor, Bureau of Labor Statistics, Consumer Expenditure Survey: Interview SurveY,1972-73,Bulietin 1997, u.S. Government Printing Office, 1978 and Consumer Expenditure Survey: Diary Survey, July 1972-June 1974, Government Printing Office, 1977. 10. "An Analysis of the Effects of IndexinQ for Inflation on Federal Expenditures," General Accounting Office, August 15, 1979, p. 40. 8. The PCE is not a perfect Paasche index in the theoretical sense, because it cannot follow every price for the commodities and services that it covers. The Bureau of Economic Affairs combines the 107 personal-consumption categories, expressed in nominal dollars, into 70 categories which are 11. "Reducing the Federal Budget: Strategies and Examples", Congressional Budget Office, February 1981, p.145. P.O. Box 7702 San Francisco. California lJ4120 BULK RATE MAIL U.S. POSTAGE PAID PERMIT NO. 752 SAN FRANCISCO. CALIF 56