The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
R V Vol. 73, No. 6 November/December 1991 3 T h e 1990 Oil P ric e H ike in P e r s p e c tiv e 19 A lte rn a tiv e M e asu res o f the M o n e t a r y Base: W h a t A r e the D iffe r e n c e s and A r e T h e y Im portant? 36 A M ic r o e c o n o m e t r ic A p p r o a c h to Estim ating M o n e y D em an d: T h e A s y m p to tic a lly Ideal M o d e l 52 M ic r o s tr u c tu r e T h e o r y and the F o re ig n E xch an ge M a r k e t THE FEDERAL J RESERVE JU BANK of ST.IjOI IS 1 Federal R eserve Bank of St. Louis R e v ie w November/December 1991 In This Issue . . . When oil prices doubled in 1990 as a result of Iraq's invasion of Kuwait, the economic effects of oil price shocks became central to dis cussions o f the economic outlook and economic policy. In the first arti cle in this Review, “The 1990 Oil Price Hike in Perspective,” John A. Tatom discusses several popular but misleading perceptions about the changing effects of oil price shocks on the U.S. economy and policymak ers’ response to such changes in the past. The author explains that the principal channel of influence of higher energy prices is a loss in economic capacity and productivity. Other channels of influence arise through the oil import bill or changing ener gy efficiency. Tatom compares the recent experience with the two previous oil price shocks and finds that there was little reason to expect the effects of the 1990 oil price shock to be much smaller. The most significant difference in the latest shock, says the author, is that it was more clearly temporary, so that the doubling of oil prices from July to October 1990 was virtually eliminated by offsetting move ments from October to March 1991. As a result, Tatom concludes, the adverse effects of the oil price shock should be reversed almost as quickly as they occurred. * * * In the second article in this issue, “Alternative Measures of the Mone tary Base: What Are the Differences and Are They Important?” Michelle R. Garfinkel and Daniel L. Thornton explore the differences in two measures of the adjusted monetary base, one constructed by this Bank, the other constructed by the Federal Reserve Board. Noting that these two indicators of monetary policy can and frequently do give conflicting impressions of monetary policy, the authors briefly review the basic idea behind the adjustment for reserve requirement changes. They then discuss differences in the construction of the two measures and go on to explore the importance of these differences empirically. Because the differences in their construction are arbitrary and technical in nature, there is little reason to prefer one measure over the other. The authors suggest that, when the difference is important, the measure that ap pears to perform best for the problem at hand should be used. * * * In the third article in this Review, “A Microeconometric Approach to Estimating Money Demand: The Asymptotically Ideal Model,” Piyu Yue presents an advanced approach to estimating money demand called the “Asymptotically Ideal Model” or AIM. She briefly reviews alternative microeconometric approaches to money demand, then estimates the AIM using U.S. quarterly monetary aggregate data. Dynamic simulations NOVEMBER/DECEMBER 1991 2 of the growth rate of the various aggregates and consumption suggest that the model performs well. Among other things, she concludes that the failure of conventional money demand equations may result from the inability of linear equations to approximate the behavior of non linear demand functions. * * * In recent years, a number of articles discussing the theory of securi ties market microstructure have appeared in economics and finance journals. The study of market microstructure deals with the behavior of participants in securities markets and the effects o f information and in stitutional rules on the economic performance of securities markets. Although a large and growing number of such articles have appeared, relatively few have attempted to model the foreign exchange market. In the fourth article in this issue, "Microstructure Theory and the For eign Exchange Market," Mark D. Flood reviews the theoretical literature on market microstructure to see what lessons it holds for the foreign exchange market. Microstructure theory is of interest to students of the foreign exchange market, says the author, because it can yield insights into dealers’ behavior and the impact of institutional arrangements. Con versely, the foreign exchange market is o f interest to students of microstructure, because it combines two very different methods of matching buyers and sellers—bank dealers trade both directly and through brokers. * FEDERAL RESERVE BANK OF ST. LOUIS * * 3 John A. Tatom John A. Tatom is an assistant vice president at the Federal Reserve B ank o f St. Louis. James P. Kelley provided research assistance. The 1990 Oil Price Hike in Perspective J . ' HE ECONOMIC EFFECTS of the sharp rise in oil prices in 1990 were, for a while, the cen tral issue in discussions of the economic outlook for 1990 and 1991. Iraq’s maneuvers to raise the world price of oil late in July 1990 and their invasion of Kuwait less than a week later led to a doubling of oil prices. As a result, oil price shocks and the appropriate economic policy response to such shocks became subjects of renewed speculation. One of the most popular hypotheses to emerge at the time was that, since the economy was different in 1990 than it had been when previ ous large oil price increases occurred, the 1990 price rise should not affect the economy to the same extent.1 It still was widely believed, how ever, that the principal and most immediate effect would be the onset o f a recession. In response, many analysts believed that the Fed eral Reserve would ease monetary policy be cause they thought it had done so at the outset of previous oil shocks. This article outlines the potential channels of influence o f a rise in the price of oil and the ex tent to which the purported differences in eco nomic conditions in 1990 could account for differences between the economic effects of the 1990 oil price surge and those in earlier, com parable episodes. W HY DO OIL PRICES MATTER? One usually encounters two principal argu ments in assessing how oil and energy price changes affect the economy. First, since energy resources are used to produce other goods and services, a change in their price affects how much of the goods are produced as well as the mix of resources that will be used to produce them. This argument focuses on the supply side of the markets for goods and services. It sug gests that the output losses associated with higher energy prices are permanent, so that changing economic policies or shifting market prices cannot replace the loss. A second argument focuses on the effects on the demand for a country’s output. It suggests that output losses are cyclical or transitory, so that adjustments in wages and prices, or in eco nomic policy, can reverse the loss in output. 1Fieleke (1990) was one of the first to develop this a rg u ment. Am ong the reasons he cites are differences in the size of the shock, the sensitivity of oil consum ers to oil price changes, the state of the econom y before the oil shock and differences in available policy options. The Council of Econom ic Advisers (1991) provides a more ex tensive discussion consistent w ith this view. NOVEMBER/DECEMBER 1991 4 Each argument suggests which characteristics of the economy determine the effects of an energy price shock, as well as how changes in these characteristics would alter these effects. Each also provides a different conclusion about the potential for economic policy to ameliorate the adverse influences of energy price shocks. Energy Prices and Econom ic Ca pacity: The Permanent Effects o f an Energy Price Shock Energy resources are used to produce most goods or services. As such, a rise in their price will (1) raise the total cost of an efficient pro ducer’s output, (2) alter the most efficient means for producing output, (3) lower the profit-maxi mizing level of output, (4) raise the long-run equilibrium price of output and (5) reduce the capacity output of each firm ’s existing stock of capital.2 Capacity output declines when energy prices rise because firms reduce their use of energy and energy-using capital, some capital becomes obsolete, and firms use labor and capi tal to economize on energy costs—that is, they generally switch to less energy-intensive p ro duction methods. The shaded insert on pages 6 and 7 b riefly explains the microeconomic foundations of this capacity effect. The economy’s aggregate supply is the sum of the supply decisions o f the nation’s firms. Thus, the effect of energy prices on the typical firm’s economic capacity determines the effect on the economy’s natural output and its aggregate sup ply. The influence of a rise in the price of ener gy on aggregate supply is shown in figure 1. The aggregate supply curve indicates the output that producers will supply at various levels of the aggregate price level, given other factors influencing this decision. The supply curve typi cally is derived from a given production func tion, which relates output to the employment of resources such as labor and capital. An initial level of nominal wages, the supplies of labor 2T his discussion draw s upon Rasche and Tatom (1977a) and (1981); Karnosky (1976) was one of the first to argue that a rise in the price of energy reduces capacity and raises th e price level. H ickm an, H untington and Sweeney (1987) sum m arize th e sim ila rities and differences o f e m piri cal estim ates of the effects of energy price shocks in 14 prom inent m acroeconom ic m odels. All of these m odels show a perm anent o utput loss due to an oil price increase; in six of these m odels, th is loss is e xp lictly cited as a decline in potential output. The Council of Econom ic Advisers (1991) suggests that any effect on capacity is transitory. O thers who have been FEDERAL RESERVE BANK OF ST. LOUIS and capital goods and the relative price of energy resources are assumed to be given in deriving a particular aggregate supply curve. Suppose that the price level, P0 in figure 1, results in a real wage (nominal wage deflated by the price level) at which a given supply of labor resources is fully employed. At this level of employment, which often is referred to as natural employment, the economy produces its capacity or natural output level, X°. Given the nominal wage level, the real wage is lower when prices are higher than P0, so firms would desire to produce more output and demand more employment. Workers would be unwilling to work more at a lower real wage, however, so neither output nor employment could rise. Indeed, to maintain output and employment, the nominal wage must rise proportionately with the price level to keep the real wage unchanged. Thus, the aggregate supply curve is vertical at X" for prices above P0. At a lower price level than P0, the real wage is too high for firms to employ as much labor or produce as much out put as at X"; output and employment are below their natural counterparts along this upwardsloping portion of the aggregate supply curve. A rise in the relative price of energy, given the short-run supply of capital and labor re sources, will reduce capacity output from X" to X^, and raise the aggregate level o f prices as sociated with this output from P° to P1. The percentage decline in capacity output and the rise in price level associated with each 1 per cent rise in the relative price of energy gener ally are equal and proportional to the share of energy in the cost of outpu t.3 In this case, al though real output has fallen, the level of nomi nal spending on output at point B in figure 1 will be the same as at point A. Thus, if output is measured by the nation’s real GNP, then real GNP is lower at point B than at point A, but nominal GNP is the same. Aggregate output and the price level are de termined by the interaction o f aggregate supply critical of the significance o f the capacity e ffect include Berndt (1980), Berndt and W ood (1987), Denison (1979) and (1985), D arby (1984) and O lson (1988). 3The conditions required to obtain the eq u ality of these out com es are discussed in Rasche and Tatom (1977a) and derived in Rasche and Tatom (1981). The shaded insert to this article provides a sum m ary of th e analysis. 5 Figure 1 The Effect of a Higher Price of Energy on Output and the Price Level and demand. Aggregate demand indicates the quantity o f output demanded at various price levels and is inversely related to the general price level. The aggregate demand curve in figure 1 passes through both points A and B. At these points, nominal GNP (the product o f the price level and output) is the same, indicating that a rise in the price level is associated with an equal proportionate decline in real output. Thus, the nominal value of aggregate demand is unaffected by the price level. This assumption simplifies the analysis with out reducing its generality. The higher price level reflects the permanent decline in natural output, with no cyclical loss of output or em ployment; the smaller natural output level is produced by an unchanged level of natural em ployment. Only a further reduction in output would fit the notion o f a cyclical loss associated with cyclical unemployment. For cyclical output and employment losses to arise from an energy price increase, either (1) 4Tatom (1981) indicates th a t tem porary cyclical effects oc cur for the third reason above; that is, they are short-run dynam ic variations as the econom y m oves from point A to point B. In this analysis, sticky prices keep the price level from rising instantaneously. Inventories and increased em ploym ent initially are used to m eet unchanged sales and partially offset the productivity loss. W ithin a short tim e, however, firm s begin to reduce output because sales fall aggregate demand must be more responsive to a rise in the price level (flatter than that drawn in figure 1), (2) an increase in the relative price of energy must cause the aggregate demand to shift to the left, or (3) there is some short-run dynamics o f price and output adjustment not shown in the movement from A to B. For exam ple, if the price level adjusts upward slowly be cause of temporary rigidities in the prices of goods and services, then a rise in energy prices will lead producers to reduce employment tem porarily, reducing output by more than the decline in natural output. When output prices rise sufficiently to reduce real wages by the extent of the permanent decline in labor pro ductivity, employment will be restored to its natural level and output will have fallen only to the extent of the capacity loss.4 Thus, even if the principal effects o f an energy price rise are a permanent decline in capacity and a rise in the price level, some transitory recessionary declines in output and employment are likely to occur. Energy Prices and Aggregate Dem and The second channel of influence above indi cates that a rise in the relative price o f energy would shift aggregate demand to the left, reduc ing output and/or the level of prices. These ef fects are transitory, or cyclical, however, in contrast to the permanent output loss arising from reduced capacity. When output is less than its natural level, employment is as well. Thus, wages and rental prices of capital goods will tend to fall, shifting the upward-sloping portion of the aggregate supply curve and the price level down until output is restored to its natural level. Aggregate demand will fall if a rise in oil prices raises expenditures on oil and total im ports and thereby lowers net exports. In effect, the rise in the oil import bill acts like a tax on domestic income, reducing aggregate demand. For such a shift in aggregate demand, the de cline in output and employment are propor- m ore as prices begin to rise; cyclical losses in output and em ploym ent occur. Em pirical evidence indicates that, after about a year, the price level has adjusted fu lly (to P, in fig u re 1), so producers step up production and em ploy m ent to th e ir natural levels (point B). NOVEMBER/DECEMBER 1991 6 The Effect of a Higher Price of Energy on Eco nomic Capacity The effect of a rise in the price of energy on a firm ’s cost structure is illustrated in the accompanying figure, which shows the longand short-run average cost of output and how they are affected by a rise in the price o f energy. The long-run average cost curve (LACq) indicates the minimum cost per unit of output for the firm. This curve is derived from the least-cost combination of resources that produces the indicated quantity of out put, given available technology and the prices of the resources used to produce the firm ’s output. The long run refers to a period over which the firm is free to vary the quantity of all resources used in production. In the figure, the long-run average cost curve is horizontal, or unaffected by the level o f output. The long-run average cost could decline over some range of output, indicating what are called “economies of scale,” or it could rise over some range indicating “dis economies of scale.” When the curve is hori zontal, as in the figure, the firm ’s production exhibits constant returns to scale so that, for any initial output level, proportional increases or decreases in output can be obtained from equiproportional changes in the employment o f each resource. In this case, long-run cost varies equiproportionately with output, so that the long-run average cost is unaffected by the output level the firm chooses to pro duce. With constant returns to scale, the long-run average cost also indicates the longrun marginal cost, the minimum additional total cost o f producing an additional unit of output.1 The short run is characterized by the inabil ity to vary the use of some resources. In par ticular, firms have difficulty in varying their capital stock—their plant and equipment—to produce more or less output in the short run. Thus, a given size of capital stock would be freely chosen to allow least-cost produc tion at only one level of output. At a larger ’ The m ost general case is often illustrated with a Ushaped long-run average cost curve, w hich exhibits in creasing returns to scale over the range of relatively low o utput levels and decreasing returns to scale at FEDERAL RESERVE BANK OF ST. LOUIS The Effect of a Higher Price of Energy on a Firm’s Cost, Capacity and Price X1 X0 O u tp u t p e r p e rio d (smaller) output, more (less) capital would be used to minimize the cost o f production. The output level at which the existing stock of capital would be selected is called the eco nomic capacity o f the firm ’s capital stock. At this output, the long-run and short-run total and average cost of output are the same. Should the firm desire to produce more or less output, it could not do so as cheaply in the short run as it could in the long run be cause the capital stock cannot be varied in the short run. Higher-cost methods of pro duction, which use relatively more labor or other variable resources, must be used until the capital stock can be altered. Since the total cost of producing any level o f output other than the capacity level (XQ) is higher in the short run than the long run, the shortrun average cost (SAC0) is also higher. When the price of energy rises, the longrun and the short-run average cost of output rise (LACi and SACt, respectively). The size of relatively high levels of o utput. A t the m inim um longrun average cost, there are constant returns to scale. 7 the rise depends, in part, on the size of the increase in energy prices and the share of energy resources in total cost. The effect on the firm's economic capacity depends on how the optimal long-run mix o f resource employ ment changes. The higher price of energy will cause the firm to reduce its use of energy to produce a given level o f output (say X0) and increase the use of some re sources whose prices have not changed. The use of some other resources could also be reduced along with energy.2 As drawn in the figure, the capacity output of the firm falls to X , since the short-run average cost rises more than the long-run average cost at out put X0. 2T he effect of a rise in the price of one resource on the e fficient level of em ploym ent of another is assessed by the “ e lasticity of su b stitu tio n ” between the two resources, w hich m easures the percentage change in one resource associated w ith each 1 percentage-point rise in the price of the other, holding o utput constant. If the e lasticity o f substitution between energy and some o ther resource is positive, a rise in the price of energy raises the em ploym ent of th e other resource. In this case, energy and the other resource are said to be substitutes. W hen this e lasticity of substitution is nega tive, energy and the other resource are “ com ple m e n ts.” 3Some analysts describe production as “ p utty-clay,” m eaning that, in the short run, the capital stock re qu ire s fixed proportions of oth er resources. In this case, the elasticity of substitution between energy and capital is zero in the short run, so that capacity is un affected. O ver tim e, a capacity loss w ould be realized as obsolete capital is replaced w ith capital that uses tional to the rise in the oil price and the share of oil imports in GNP. In this case, the net oil import status of a country determines the ef fects o f an energy price shock.5 Countries that export oil face larger aggregate demand when oil prices rise; world aggregate demand and out put are redistributed from oil importers to ex porters when oil prices rise, and conversely when oil prices fall.6 5Feldstein (1990) and the Council of Econom ic Advisers (1991) provide recent restatem ents of th is shift in ag gregate dem and and the price-level-induced m ovem ent along the aggregate dem and curve as the central chan nels of influence of an oil price hike. The Council of Eco nom ic Advisers also em phasizes a decline in real consum ption expenditures as a result of an oil price hike. Perry (1991) argues that the oil price hike had little effect on the econom y in 1990, because it did not reduce real in com e m uch (operating through the aggregate dem and channel above), nor did it induce the Fed “ to raise interest Whether capacity output falls, rises or is unaffected by a rise in the price of energy depends on the relationship between capital and energy resources in production. When these resources are substitutes, a rise in en ergy prices would lead the firms to substitute capital for energy in producing XQ. Thus, the existing capital stock would be chosen to produce a smaller output level (XJ.3 If capital and energy are independent or complements, the higher short-run and long-run average cost level will be identical at the same or higher level of output, respectively. Energy and capital are substitutes, however, which implies that economic capacity will fall when energy prices rise.4 less energy. See C orcoran (1990), for exam ple. Evi dence cited in the text suggests that putty-clay con siderations are not dom inant in the short run. 4The size of the fall in econom ic capacity is proportional to the share o f energy in fa cto r cost and th e elasticity of substitution and inversely proportional to the expen diture elasticity of the capital stock (or fixed resources generally). The ela sticity of capacity equals ( —ake/rjk) se, w hich reduces to - se, th e share of energy in total cost, if the expenditure e lasticity of capital, >jk, and the e lasticity of substitution, oke, are one; these elasticities are each one for a broad class of fu n ction s used in em pirical analysis. The form er e lasticity is the response of desired capital use to an increase in expenditures on all resources. For increases in the relative price of energy, instead of the nom inal price, the share of ener gy in the expressions here and above is replaced by (se/1 - s e). Monetary P olicy and Oil Price Shocks The appropriate monetary policy response to a rise in the relative price o f energy depends on its dominant channel of influence. If the higher energy price only lowered aggregate demand, policymakers could take offsetting actions to neutralize this shift by increasing the money rates to fig h t in fla tio n ” as, he argues, it had in th e past. 6A change in the relative price of energy could also affect a ggregate dem and by altering investm ent in plant, equip m ent and housing. Such an effect can account for a d ecline in the real interest rate, which is incom patible with a conventional m odel of aggregate dem and. Reinhardt (1991) d iscusses the effects of energy price shocks on in terest rates. NOVEMBER/DECEMBER 1991 8 supply, which would shift the aggregate de mand curve back to the right. If an energy price increase affects aggregate supply, however, both raising the price level and reducing natural output, policymakers could attempt to offset the price level rise by reducing the money stock to reduce aggregate demand. This would result in a cyclical loss in output and employment as the economy’s output fell short of its lower natural output level until the price level declined sufficiently. Alternatively, policymakers could attempt to offset the reduction in output by raising ag gregate demand. Raising demand could not re store the economy’s natural output, however; it would not replace the energy and capital re sources that firms can no longer afford to pur chase or use. Instead, it would further raise the aggregate level of prices associated with the smaller level of capacity output.7 Thus, there is no real policy dilemma posed by oil price increases. Raising the money stock cannot offset a loss in natural output, while reducing the money stock can only offset a price level increase at the cost of a further loss in output and a cyclical rise in unemployment. Moreover, it is virtually impossible to alter mon etary policy enough to fully offset the price level surge because of the time it takes for a change in the money stock to affect the price level and because of the relatively small size of the initial price response to changes in mone tary policy.8 An unchanged growth rate for the money stock is a policy that accepts the perma nent output and price level consequences de scribed above without compounding one or the other loss. HAVE THE ECONOMIC EFFECTS OF OIL PRICE SHOCKS CHANGED? Many analysts argued that the rise of oil prices in 1990 would have substantially less im 7Kahn and Ham pton (1990) contrast three m onetary policy options, which include tightening to offset the price level effect, easing to offset the cyclical effects and a neutral policy which “ m aintains constant m onetary or nominal GNP g ro w th .” Feldstein (1990) endorses the third option, nom inal GNP targeting, and he also equates this w ith un changed m oney stock growth. 8See Tatom (1981) and (1988a), for exam ple, for evidence on the relative size and lag lengths for energy price and m onetary policy effects on prices and output. FEDERAL RESERVE BANK OF ST. LOUIS pact on the U.S. economy than earlier oil price hikes. There were two versions of this argu ment. The first was that the adverse effects of an oil price rise are proportional to the share of oil imports in the economy and that this share had fallen since the earlier oil price shocks. The second argument was that the effects of an oil price rise are proportional to the use of energy per unit of output and that this dependence on energy also had fallen.9 D oes a Smaller Im port Share Reduce the Adverse Effects o f an Oil Price Hike? If the share of oil imports in GNP has fallen, then the first argument above implies that the economy’s aggregate demand and output have become less sensitive to a rise in oil prices. Figure 2 shows expenditures on petroleum im ports as a percent of nominal GNP since 1970. In mid-1990, this share was about 1 percent, less than half its level in early 1979, but above its 0.6 percent share in 1973. Thus, the share had fallen below its level preceding only one of the previous two oil price shocks. The import share argument has other short comings. First, it suggests that oil-exporting countries, including Canada in 1974 or the United Kingdom in 1979, should gain when oil prices rise, because net exports and aggregate demand should rise. In each instance, however, output did not rise nor was there other evi dence of a cyclical expansion following the previous oil price shocks. The argument also suggests that countries that import a relatively small share of their oil, like the United States, will be less affected than countries that import relatively more o f their oil, like Germany or Japan. The earlier experience with oil price shocks indicates that, especially in 1973-74, both the temporary rise in inflation and the perma nent loss in output were larger in Japan than in 9See C ouncil of Econom ic Advisers (1991), Kahn and Ham pton (1990), Anderson, Bryan and Pike (1990), Brinner (1990), “ How Big An Oil S hock” (1990), "S h o cke d A g a in " (1990), May (1990), Y anchar (1990) and Fieleke (1990) for analyses that em phasize one or both of these argum ents. Fieleke, Kahn and Ham pton, May and Y anchar emphasize, to varying degrees, that the expected effects also are sm aller because of a sm aller expected rise in the price of oil. 9 Figure 2 Petroleum Imports as a Percent of GNP the United States, but that these effects were smallest in Germany.10 There are three other major difficulties with the import share argument. First, it is difficult to reconcile the relatively large economic effects of oil price hikes with the relatively small size o f the petroleum import share. Second, an ag gregate demand reduction in the face of an oil price hike implies only a cyclical decline in out put, not a permanent one. The failure of real GNP per worker and real wages to return to their previous growth trends in virtually all na tions after the two previous OPEC price hikes is not consistent with the pattern expected for a purely cyclical loss. Third, the trade-based ag 10See Rasche and Tatom (1981), Tatom (1987) and Tatom (1988a) for reviews of this international evidence. In 1973, the share of petroleum im ports in GNP equaled 1.7 per cent in G erm any and 1.6 percent in Japan, m uch more than the 0.6 percent in the United States. S im ilarly, in 1978, this share was 2.7 percent in Japan, 2.5 percent in G erm any and 1.9 percent in the United States. 11See Tatom (1988b) for a discussion of the theory and evi dence supporting such contrary effects. C onsistent with gregate demand story predicts a decline in net exports and the currency value of a large oil importer after an oil price shock. At least for the United States, however, exports rose rela tive to imports so that both net exports and the exchange rate rose after each earlier oil price shock. Indeed, the only periods of positive net exports since 1970 occurred in 1974-75 and 1979-82, following the earlier oil price hikes.11 D oes Increased Energy Efficiency Reduce the Adverse Effects o f an Oil Price Hike? The second argument for less adverse effects of the 1990 price hike is based on a decline in th is rise in net dem and for U.S. goods, th e trade-weighted value of the dollar rose in IV/1973 and 1/1974, and was higher over the rest of 1974 than it had been in the two quarters preceding the oil price rise. In the second quarter of 1979, the value of the d o lla r also rose slightly. Over the next four quarters, the value of the dollar was only 0.6 per cent lower than in the two quarters before the oil price hike. NOVEMBER/DECEMBER 1991 10 Figure 3 E nergy Use per Unit of Real GNP (Thousands of BTUs per dollar of real GNP, 1982 prices) 25 0 1970 energy use per unit of output. According to this argument, energy is less important to a firm ’s production than in the past, so a rise in oil prices is expected to have a smaller effect on prices and production today than in the past. Figure 3 shows total U.S. energy use per unit of output (measured in BTUs per unit of real GNP) from 1970 to 1988, the latest year availa ble on this basis.12 Energy use per unit o f out put has fallen sharply since 1973: BTUs used per unit of real GNP were about 31 percent lower in 1988 than in 1973 and about 22 per cent lower than in 1979. This rise in output per unit of energy is not surprising given the rise in the relative price of energy since 1973, but it is not relevant in assessing the importance of energy as a resource or in assessing whether 12T he energy expenditures and quantity data used for figures 2 and 3 are from the Energy Inform ation A dm inis tration, State Energy Price a nd Expenditure Report, 1988 (Septem ber 1990). FEDERAL RESERVE BANK OF ST. LOUIS 1988 the effects of an energy price boost have declined in magnitude. While energy use per unit of output is lower than earlier, the responsiveness of prices or out put to a change in a resource’s price are pro portional to the share of the resource’s cost in total cost, not to the share of its quantity in out put. Consider the familiar case of labor produc tivity. Labor employment per unit of output in the business sector declined by nearly one-third from 1955 to 1973, as output per worker rose from $21,084 to $31,142 (1982 prices). Thus, the economy became less dependent on labor over these 18 years—in exactly the same sense and to nearly the same extent as some have sug gested about energy resources over the past 18 years. Nevertheless, the share o f labor in total 11 Figure 4 Energy Expenditures as a Percent of GNP Percent cost was about the same: 65.3 percent in 1973 and 64.8 percent in 1955. For a given share of labor in cost, a percentage point rise in the wage rate will raise the cost of an additional unit of output and price in proportion to this share.13 Analysts who emphasized the increased pro ductivity of energy are unlikely to espouse the equivalent view that a 10 percent rise in wages has a smaller effect on unit costs or product prices today than in 1973 or 1955. As discussed previously, the response of capacity and price to changes in a resource’s price depends on the share of the resource in cost, not on its produc tivity or output per unit. Figure 4 shows how the share of energy ex penditures as a percent of GNP has changed from 1970 to 1988. Following each energy price hike, expenditures rose sharply relative to GNP; as energy prices fell beginning in 1982 (on an 13A typical discussion of the relationship between wages, productivity and prices can be found in Fischer, Dornbusch and Schm alensee (1988), pp. 566-67. Shin (1991) annual basis), the share fell. By 1988, the share nearly had returned to its 1970-73 level. These data suggest that the share of energy in the cost of the economy’s output has not fallen be low its level before the earlier oil price changes, especially the 1973-74 rise. Thus, these data do not support the view that a doubling of the price of oil should be expected to have smaller effects in 1990 than it had earlier, especially in 1973-74, because the share of energy in total cost has not declined. RECENT OIL AND ENERGY PRICE DEVELOPMENTS The economic effects of an energy price shock depend on the size of the price change as much as they depend on the responsiveness of mea sures o f economic performance to a given discusses o ther shortcom ings of using the energy-output ratio for analytical or policy purposes, NOVEMBER/DECEMBER 1991 12 change in energy prices. Table 1 shows the monthly average price o f oil purchased by re finers since June 1990. Following the Iraqi inva sion of Kuwait and the subsequent U.N. embargo of crude oil exports from both countries, the price of oil doubled within three months. The 1990 oil price rise was comparable in magnitude to the two earlier OPEC price hikes in 1973-74 and 1979-80. In each of these previous cases, oil prices nearly doubled. In the second instance, oil prices rose again sharply in the first quarter of 1981. A rise in the price of oil is likely to raise the cost of production of competing energy sources and raise the demand for competing forms of energy, as consumers substitute other fuels for oil. For both reasons, the prices of competing sources of energy change along with the price of oil. Thus, an oil price shock can be consi dered more generally an energy price shock. Figure 5 shows the relative price of crude petroleum—measured by the producer price index for crude petroleum deflated by the busi ness sector implicit price deflator—and the rela tive price o f energy—the producer price o f fuel, power and related products relative to the same deflator.14 The relative price affects economic performance because producers of goods and services assess the cost o f energy relative to the goods and services produced using it. From the third quarter o f 1973 to the third quarter of 1974, the relative price o f crude oil nearly dou bled. Measured in 1990 prices, the composite refiner acquisition cost of crude oil rose from $10.67 per barrel in 1973 to $21.28 in 1974, or 99.4 percent.15 In the second OPEC oil price shock, from early 1979 to the second quarter of 1980, this relative price of oil nearly doubled 14A logarithm ic scale is used because differences in logarithm s show percentage changes; an equal-sized in crease or decrease in figure 5 reflects equal percentage changes. For exam ple, a rise from 50 to 100, or 100 to 200 represents a doubling of the relative price and the respective distance in each case is the sam e in figure 5. 15The rise in the relative price of oil shown in the figure ac tu a lly begins in early 1973, but this earlier increase largely reflects partial and tem porary relaxation of U.S. price con trols on dom estic crude oil prices. The m uch larger OPEC price increases followed the Yom Kippur W ar in O ctober 1973. The 1947 oil price shock is not discussed here. The producer price for crude petroleum m easures prices paid to dom estic producers, w hich were controlled from 1971 to early 1981. O ver m ost of th is period, the com posite refiner acquisition cost was higher, but was representative of oil p rices paid by dom estic purchasers. ,6The total output o f Kuwait and Iraq fell about 4 m illion bar rels per day in A ugust 1990 from its M ay-July 1990 aver- FEDERAL RESERVE BANK OF ST. LOUIS Table 1 The Composite Refiners Acquisition Cost of Crude Oil (dollars per barrel) Date Price Date Price June 1990 $14.98 Ja n u ary 1991 $22.90 July 16.15 February 19.02 A ugust 23.57 March 17.89 Septem ber 30.01 April 18.43 O ctober 33.18 May 18.60 Novem ber 30.61 June 17.98 Decem ber 26.21 again, rising from $22.35 per barrel to $41.82 per barrel. A further surge in early 1981 put the price up to $50.75 per barrel. From the second quarter o f 1990 to the fourth quarter of 1990, the price of oil rose from $16.10 per barrel to $30.00 per barrel, an 86.3 percent rise that is almost as large as the near-doubling in the previous two oil shocks.16 If the effects of oil price hikes are proportional to their size, then the effects of the 1990 in crease should be about the same as in the two previous instances. The relative price of energy rose about 50 percent during the previous two energy price shocks. From the second quarter of 1990 to the fourth quarter o f 1990, however, the rela tive price of e n e rg y rose 29.6 percent, about 60 percent o f the earlier magnitudes.17 Thus, on this basis, the recent energy price shock is somewhat smaller. age; by Novem ber and D ecem ber 1990, it was down 4.6 m illion from th e e arlier average. The latter reduction equaled 7.6 percent of w orld production and 19 percent of OPEC output. In com parison, the reductions in the total of Iran and Iraq production from 1978 to its lowest annual average level in 1981 was som ew hat larger, 5.4 million barrels per day, but this was 18.2 percent of O PE C ’s 1978 production. 17Em pirical estim ates suggest th a t the relative price of ener gy adjusts contem poraneously and w ith a one-quarter lag to changes in the relative price o f oil; thus, one reason for the relatively sm aller rise in the energy price is the fact that the relative price of crude oil fell 20.6 p ercent in the second quarter of 1990. W hen expressed in logarithm s, each 1 percentage-point rise in the relative price o f crude oil is estim ated to result in about a one-half percent rise in the relative price of energy. See Tatom (1987b). 13 Figure 5 Relative Price of Energy and Crude Petroleum Index (1975 to 1978 average equals 100) 250 50 1950 55 60 65 70 75 80 85 1990 There were two other important differences between the recent rise and the previous two. First, the recent rise occurred much more quickly—in two quarters instead of four or six. Second, the recent increase did not persist. Nevertheless, producers did not know at the time whether, or by how much, oil prices might decline in the future. This article assumes that producers treat price changes as permanent, in the sense that the expected price they use for economic decisions is the current price. It also focuses only on the effects of the recent price increase. To the extent that producers did not anticipate having to face the price increase, the effects of the price shock should be smaller. as a temporary acceleration in inflation and a temporary reduction in output growth. More over, temporary rigidities in nominal prices and lags in the adjustments that firms and con sumers make in response to large price changes were likely to give rise to temporary movements in employment, including a recessionary decline in employment, although past experience sug gests that such a change occurs with a delay of about one year. These effects should be expected to have been somewhat smaller than those fol lowing previous oil shocks, because the rise in the relative price of energy in the 1990 episode was only about 60 percent as large as the previ ous increases. THE EXPERIENCE IN PREVIOUS OIL PRICE SHOCKS Following the sharp rise in the relative price of energy in 1973-74 and 1979-80, the loss in ca pacity and adjustment to a higher price level, as discussed earlier in reference to figure 1, were reflected in a temporary acceleration in the in flation rate. In each case, output growth slowed, reflecting both the permanent decline in natural output and a transitory loss in output. Produc The previous discussion of energy price ef fects indicates that the 1990 oil price hike should be associated with a lower level of natural out put and productivity and a rise in the price level. These changes were likely to be revealed NOVEMBER/DECEMBER 1991 14 tivity (and real wages) fell.18 Generally, the per manent loss in output and productivity and the rise in prices were experienced first, with the temporary surge in inflation (as measured by the GNP deflator) delayed about two quarters.19 Employment declined much later and for only a few quarters. Cyclical unemployment associated with an oil price rise peaked about six quarters later, before quickly dissipating. Table 2 shows these developments for the three most recent large energy price hikes. For periods surrounding each oil price hike, the ta ble provides real GNP growth, productivity (bus iness sector output per hour) growth, the rate o f increase in the GNP deflator, civilian employ ment growth, the average unemployment rate for the civilian labor force and money stock (M l) growth. Each measure is provided for the year before and the first four consecutive twoquarter periods following the shock. Two-quarter periods are used to simplify the data presen tation, although the timing of energy price effects facilitates the usefulness o f this proce dure. OPEC1 refers to the first oil price shock which began in IV/1973. OPEC2 begins in 11/1979 and IRAQ begins in III/1990. As table 2 indicates, real GNP growth slowed following the two previous oil price hikes, but did not become negative on a two-quarter basis until after the first two quarters (OPEC1) or after a year (OPEC2). The slowing in output growth reflects both the decline in natural out put and, principally later, a temporary cyclical loss in output. Table 2 also shows that the ex pected productivity decline (negative growth) oc curred more quickly than the decline in real GNP in the previous two cases; it began in the first two quarters of the energy price shock in 18These developm ents were observed in nearly all countries. The notable exception was that incom e policies im peded the reductions in real wages (and, therefore, in labor productivity) in some countries, especially in 1973-74, so that the effective supply of natural em ploym ent fell, further reducing natural output. See Rasche and Tatom (1977a), (1977b) and (1981), Tatom (1988a) and (1987). Ham ilton (1983) also provides em pirical evidence supporting the perm anent effect on U.S. real GNP. Helliw ell, Sturm , Jarrett and Salou (1986) provide international evidence on the e ffect on natural output. 19See, for exam ple, Tatom (1981) and (1988a). The lag for the PCE d eflator and CPI is shorter (one quarter) and the m agnitude is larger for these consum er price series, be cause the share o f energy cost in expenditures is larger for consum er expenditures than for GNP as a whole. Thus, the effect of a given rise in oil prices is larger for consum er price inflation m easures. The effects on producer prices o ccu r even fa ste r and are even larger. FEDERAL RESERVE BANK OF ST. LOUIS each case. Both productivity and output growth show a sharp cyclical acceleration in the last two-quarter period. The most recent energy price shock, like the earlier two, was accompanied by an immediate decline in productivity and a slowing in output growth.20 Output growth became negative earlier than in the previous two cases. Since the recent energy price hike occurred over only two quar ters, the period of decline in productivity and output growth should be correspondingly shorter than in the previous two instances. The slight rise in productivity growth in the second twoquarter period is consistent with this expectation. In the previous two instances, the decline in productivity and natural output was reflected, with about a two-quarter lag, in a sharp and temporary acceleration in the rate o f price in crease as measured by the GNP deflator. Thus, in the second two-quarter period in OPEC1, in flation accelerated sharply and only temporarily, reflecting the one-time adjustment in the price level. The same acceleration occurs in OPEC2, but with a one-quarter lag; the data for the two-quarter period ending one quarter later are shown in parentheses. As table 2 shows, how ever, in the first two-quarter period, the rate of increase in the GNP deflator rose (OPEC1) or was unchanged (OPEC2); in the latest instance, it declined.21 In the previous two cases, the delayed acceler ation in the rate of price increase persisted for about four quarters (five quarters for OPEC2), about as long as the period of sharp increases in energy prices. There is also an acceleration in the recent second two-quarter period (1/1991 and 11/1991). Since the latest price hike occurred over half as many quarters as in the previous “ Productivity growth had declined m ore rapidly in the year before the recent oil price shock than it did in the initial tw o-quarter period, so pro d u ctivity grow th did not actually slow in the second half of 1990. 21The initial decline in the rate of price increase in the first tw o-quarter period is not out of line. In each of the previ ous initial tw o-quarter periods, th is rate was m uch lower in at least one of the two quarters. In particular, in the first quarter of 1974, the rate of increase of the d eflator fell to a 5.6 percent rate; in 1979, it fell from 9.5 percent in the first quarter to a 9.2 percent rate in the second quarter and to 8.5 percent in the th ird quarter of 1979. 15 Table 2 Economic Performance Surrounding Three Energy Price Shocks Previous Year Real GNP growth rate OPEC1 OPEC2 IRAQ Productivity growth rate OPEC1 OPEC2 IRAQ 4 .4% 5.3 1.0 1.5 0.6 -0 .6 First TwoQ uarter Period Second TwoQ uarter Period 0.7% 1.6 - 0 .1 - 2.0% 1.6 -1 .7 -1 .3 -2 .6 -0 .2 - 2 .1 0.4 0.2 Third TwoQ uarter Period - Fourth TwoQ uarter Period - 5.6% -4 .5 5.5% 6.6 0.9 0.4 6.8 3.7 Rate of price increase OPEC1 O PEC 21 IRAQ 7.1 8.9 4.0 Civilian em ploym ent growth rate OPEC1 OPEC2 IRAQ 3.5 3.9 0.9 3.3 1.6 - 1 .1 0.9 1.7 -1 .0 -3 .9 -1 .9 1.8 2.7 Average unem ploym ent rate OPEC1 OPEC2 IRAQ 5.0 5.9 5.3 5.0 5.8 5.7 5.4 6.1 6.7 7.4 7.5 8.7 7.4 M1 growth rate OPEC1 OPEC2 IRAQ 7.0 7.4 4.0 6.0 10.3 3.6 3.7 5.1 (1.5)1 6.8 3 .7 ' 5.8 6.9 8.0 7.7 8.9 (8.4) 3.2 11.5 8.5 (9.1) 4.9 10.9 9.4 (10.7) 7.8 11.4 (8.7) Period OPEC1 OPEC2 IRAQ 111/1972 to 111/1973 1/1978 to 1/1979 11/1989 to 11/1990 II and 111/1974 IV/1979; 1/1980 I and 11/1991 IV/1973; 1/1974 II and 111/1979 III and IV/1990 IV/1974; 1/1975 II and 111/1980 II and 111/1975 IV/1980; 1/1981 1Data in parentheses are for the two quarters ending one quarter later. two, the acceleration would be expected to be reversed in the third two-quarter period, even without any effect from the decline in energy prices in 1/1991 and 11/1991. It remains to be seen whether inflation will decline as abruptly as it did following earlier oil price shocks.22 The delayed cyclical response to an energy price hike is seen most clearly by looking at the 22The rate of increase in the CPI rose from a 3.8 percent rate in the second quarter of 1990 to about a 7 percent rate in the third and fourth quarters of 1990. Sim ilarly, the rate of increase of the producer price index rose from a 0.3 percent rate in the second quarter of 1990 to a 6.6 percent rate and a 10.8 percent rate in the third quarter growth o f civilian employment. In the two pre vious instances, employment growth slowed, but did not become negative until a year after the energy price shock began. Moreover, this de cline occurred in only one two-quarter period (the third one), when employment fell at a rela tively rapid pace. Thus, the typical recessionary characteristic o f falling employment did not ocand fourth quarters of 1990, respectively. The rate of in crease of the latter two price m easures fell sharply in the first half of 1991, reflecting the quicker response of these m easures to a rise in energy prices as well as to th e ir sub sequent decline. NOVEMBER/DECEMBER 1991 16 cur until a year after the onset of the two pre vious energy price hikes. The unemployment rate also did not rise im mediately after the two previous adverse energy price shocks. In 1973-74, it fell slightly in the fourth quarter of 1973, rose only 0.8 percen tage points by the third quarter of 1974, then peaked 3.3 percentage points higher three quar ters later.23 The unemployment rate peaked six quarters after the initial surge in energy prices, in the last period shown in the table. In the second quarter of 1979, the initial quarter of OPEC2, the unemployment rate also fell slightly, then rose gradually for the next three quarters so that it was only 0.4 percentage points higher in 1/1980 than it was before the energy price shock. The unemployment rate then rose 1.4 percentage points to a peak in III/1980, six quar ters after the initial energy price surge.24 In the most recent case, the unemployment rate rose immediately, climbing from 5.5 per cent in July 1990 to 7 percent in June 1991. Such a rise is substantially different from the pattern in the initial stages of the previous en ergy price shocks. Its behavior might better be understood in the context of the slowing in U.S. economic activity that began in 1988. For example, civilian em ployment actually began declining sharply in March 1990, five months before the energy price hike; civilian employment fell at a 0.9 per cent rate from March to July 1990 and declined further at a 0.5 percent rate from July to Oc tober 1990, when energy prices peaked; from October 1990 to August 1991, such employment fell at a 1.3 percent rate. Thus, the path of eco nomic activity downward into recession had begun well before energy prices rose.25 23One explanation for the initial decline in the unem ploy m ent rate when oil prices rise relies on the capacity loss and “ stic k y ” prices. The initial fall in productivity and ini tial absence of a price-related decline in aggregate de mand when oil prices rise require that producers raise em ploym ent to offset some of th e output loss and avoid larger-than-desired depletion of inventory. See Tatom (1981) and O tt and Tatom (1986) for discussions of this ef fect. Rasche and Tatom (1977a) show that em ploym ent rose during th e first three quarters of the 1973-74 oil shock and did not fall until five quarters later. 24ln th is second instance, a further rise in energy prices late in 1980 and early in 1981 contributed to a fu rth e r rise in the unem ploym ent rate about a year later, from IV/1981 to 11/1982. 250 th e r analysts have em phasized this point. See W eidenbaum (1990) and Erceg and Leovic (1990), for exam ple. 26After late 1982, m onetary policym akers placed relatively m ore em phasis on M2 instead of M1. Another m easure, FEDERAL RESERVE BANK OF ST. LOUIS A Comparison o f Changes in M o n e tary P olicy Actions Each of the two previous oil shocks were fol lowed by changes in monetary policy actions. There is no clear initial pattern, as money growth slowed in the initial two quarters in 1973-74 but accelerated in 1979. As shown at the bottom of table 2, however, in each case, M l growth then slowed sharply during the se cond two-quarter period, at the same time that the rate of price increase temporarily accelerat ed.26 Then, in each instance, M l growth acceler ated sharply in the fourth two-quarter period following the sharp rise in the unemployment rate. The expectation that the economy would quickly experience a recessionary rise in unem ployment because of the 1990 oil price rise was widespread. There were equally widespread warnings against repeating the "typical” policy response o f easing monetary policy to combat this unemployment.27 While there is evidence of rising unemployment and subsequent accelera tions in M l growth following previous oil price surges, these changes came more than a year after the initial oil price rise. These changes also occurred after the substantial slowing of M l growth and the transitory inflation rate hike that are more closely associated with the oil price increases. In the most recent case, money (M l) growth slowed from a 4.8 percent rate from IV/1989 to 11/1990 to a 3.7 percent rate in III/1990 and to a 3.5 percent rate in IV/1990. Money growth quickly reversed course, however, accelerating to a 6.8 percent rate, as the unemployment rate continued to rise in the first half of 1991. This the adjusted m onetary base, is often a convenient sum m ary m easure of m onetary policy actions. H igher energy prices sig n ifica n tly raise relative currency dem and one q uarter later, reducing m onetary aggregates relative to the adjusted m onetary base; see Tatom (1990). Thus, m one tary base growth is less useful as an indicator of m onetary policy d u ring energy price shocks. Bullard (1991) d iscuss es these and o ther indicators o f m onetary policy and the potentially co n flictin g signals they offer. 27For exam ple, according to T rehan (1990), “ Researchers have generally concluded that the Fed eased policy to overcom e the reduction in o utput caused by the oil em bar g o ” and " ... the Fed’s initial response to the second oil shock also was sim ilar to its response to the first oil sh o ck.” See also, C ouncil o f E conom ic A dvisers (1991), w hich indicates that policy was excessively stim ulative p ri or to the previous oil shocks so that it lacked credibility, m aking e fforts to ease ineffective. The C ouncil of Econom ic Advisers (p. 80) suggests such te m porary actions would be appropriate and effective today. 17 acceleration in M l growth occurred earlier than it had following the previous oil price hikes, although it did follow both a previous slowing in M l growth and a recessionary rise in the un employment rate, just as had similar accelera tions in M l following the two previous energy price increases.28 growth that had been under way since late in 1988. Thus, the expected productivity decline and temporary surge in inflation were accompa nied by a continuing decline in employment and cyclical output loss. While these developments were uncharacteristic of the initial effects of previous oil price hikes, monetary growth slowed in the second half of 1990 anyway. CONCLUSION There were other distinguishing features as sociated with the 1990 oil price hike. Foremost among them was its brevity: it occurred over a three-month period and was nearly reversed in another five months. Thus, while the response of output, productivity and prices appears con sistent with the capacity-loss-induced effects as sociated with previous oil price doublings, the subsequent decline in oil prices from October 1990 to March 1991 can be expected to result in offsetting price, output and employment movements. The rise in oil prices from August to October 1990 set in motion renewed concern and confu sion over both the effects o f oil price hikes and the appropriate monetary policy response. Three views achieved widespread acceptance. First, the economy was believed to be less sensitive to oil price hikes than it had been earlier. Second, it was widely believed that the principal and most immediate effect would be a cyclical decline in output and employment. Third, analysts believed that the Fed would ease policy, as it had when faced with this problem in the past. These views are at odds with previous ex perience. In 1990, the share of oil imports in GNP and energy per unit of GNP had not fallen to the level before the first oil price shock in 1973. Moreover, the relevant parameter, the share of energy in cost, had not fallen below its 1973 level either. Thus, U.S. economic perfor mance should not have become less sensitive to oil price shocks than it was before. In addition, negative employment growth and an accelera tion in money growth had not characterized the initial year of previous energy price shocks. Earlier evidence suggests that the principal cost of an energy price hike is the loss in capac ity output and productivity. A counterpart of this loss is a one-time surge in the general level of prices, which follows the energy price hike relatively closely. The adverse cyclical conse quences of past shocks occurred later. The prin cipal policy response following previous oil price hikes was a slowing in money growth. Later, when inflation declined and the unemployment rate rose sharply, money growth accelerated. The 1990 oil price rise occurred against the backdrop of a slowing in money and output 28M2 shows the same pattern. It grew at a 2.5 percent rate from 11/1990 to IV/1990, down from a 4 percent rate in 11/1990 or the 5.1 percent rise in the tw o-quarter period ending in 11/1990. In the first half of 1991, M2 growth also rose, but only to a 4.2 percent rate. Bullard (1991) indi cates that Fed decisionm akers were keenly aware of the policy dilem m a and chose to pursue a course o f neither REFERENCES Anderson, G erald H., M ichael F. Bryan, and C hristopher J. Pike. “ Oil, the Econom y and M onetary Policy,” Federal Reserve Bank of Cleveland Econom ic Com mentary (Novem ber 1, 1990). Berndt, Ernst. “ Energy Price Increases and the Productivity Slowdown in United States M anufacturing,” The Decline in Productivity Growth, Federal Reserve Bank of Boston Con ference Series, no. 22 (1980), pp. 60-89. Berndt, Ernst R., and David O. Wood. “ Energy Price Shocks and Productivity Growth: A Survey,” in Robert C. Gordan, et. al., eds., Energy: M arkets a nd Regulation, Essays in Honor o f M.A. Adelman (MIT Press, 1987), pp. 305-67. Brinner, Roger E. “ Finding Keys to Turn,” U.S. Forecast Sum mary (DRI/M cG raw Hill, O ctober 1990). Bullard, Jam es B. “ The FOM C in 1990: O nset of Reces sion,” this Review (May/June 1991), pp. 31-53. Corcoran, Patrick J. “ ‘P utty’ and ‘C lay—A Parable for Potential GNP,” Prudential Economics, The Prudential (February 15, 1990), pp. 6-9. Council of Econom ic Advisers. “ Oil Price Shocks and Eco nomic Policy,” chapter 3 in Econom ic Report o f the Presi dent (1991), pp. 79-110. Darby, M ichael R. “ The U.S. Productivity Slowdown: A Case of Statistical Myopia,” Am erican Econom ic Review (June 1984), pp. 301-22. Denison, Edward F. Trends in Am erican Econom ic Growth, 1929-1982 (The Brookings Institution, 1985). ________ . “ Explanations of Declining Productivity Growth,” Survey o f Current Business (August 1979, Part II), pp. 1-24. Erceg, John J., and Lydia K. Leovic. “ The O utlook After the Oil Shock: Between Iraq and a Soft Place,” Federal Reserve Bank of Cleveland Econom ic Com m entary (O c tober 15, 1990). easing nor tightening. He indicates that there was a con cern for a ctual inflationary pressures late in 1990, but con cern for the cyclical consequences of the oil price hike was fram ed only in term s of the potential risk. NOVEMBER/DECEMBER 1991 18 Feldstein, M artin. “ The Fed Must Not Accom m odate Iraq,” Wall Street Journal, August 13, 1990. Fieleke, Norm an S. “ Oil S hock III?” New England Econom ic Review (Septem ber/O ctober 1990), pp. 3-10. Fischer, Stanley, R udiger Dornbusch, and Richard Schm alensee. Economics, 2nd ed. (M cGraw Hill Book Company, 1988). H am ilton, Jam es D. “ Oil and the M acroeconom y since World War II,” Journal of Political Economy (April 1983), pp. 228-48. Helliwell, John, Peter Sturm , Peter Jarrett, and Gerard Salou. “ T he Supply Side in the O EC D ’s M acroeconom ic M odel,” OECD Econom ic Studies (Spring 1986), pp. 75-131. Hickm an, Bert, Hillard G. H untington, and Jam es L. Sweeney. M acroeconom ic Im pacts o f Energy Shocks (North H olland Press, 1987). “ How Big An Oil Shock,” The Econom ist (August 11, 1990), pp. 12-13. Kahn, G eorge A., and Robert Ham pton, Jr. “ Possible M one tary Policy Responses to th e Iraqi Oil Shock,” Federal Reserve Bank of Kansas C ity Econom ic Review (Novem ber/D ecem ber 1990), pp. 19-32. Karnosky, Denis S. “ The Link Between M oney and P rices— 1971-76,” this Review (June 1976), pp. 17-23. May, Todd Jr. “ Oil Prices W on’t Bring a Recession,” For tune Forecast/Special Report, Fortune (Septem ber 10, 1990), pp. 51-52. Olson, Mancur. “ The P roductivity Slowdown, The Oil Shocks and The Real Cycle,” Journal o f Econom ic Perspectives (Fall 1988), pp. 43-69. Ott, M ack, and John A. Tatom. “Are E nergy Prices Cyclica \T 'E n e rg y Econom ics (O ctober 1986), pp. 227-36. Perry, G eorge L. “ The Anom alous Recession,” The Brook ings Review (Spring 1991), p. 51. Rasche, Robert H., and John A. Tatom. “ Energy Price Shocks, A ggregate Supply and M onetary Policy: The The ory and the International Evidence,” in Karl B runner and Allan H. Meltzer, eds., Supply Shocks, Incentives a n d Na tional Wealth, C arnegie-R ochester C onference Series on Public Policy (N orth Holland, 1981), pp. 9-93. FEDERAL RESERVE BANK OF ST. LOUIS ________ . “ The Effects of th e New Energy Price R egim e on Econom ic Capacity, Production and Prices,” this Review (May 1977a), pp. 2-12. ________ . “ Energy Resources and Potential GNP,” this Review (June 1977b), pp.10-24. Reinhart, Vincent. “ Reading the Effects of an Energy Shock in Financial Markets,” Journal o f Econom ics and Business (1991), pp. 115-32. Shin, David. “ International C om parisons of Energy-G ross Na tional Product Ratios,” A m erican Petroleum Institute Dis cussion Paper 68 (June 1991). “ Shocked Again,” The Econom ist (August 11, 1990), p. 23. Tatom, John A. “ The Effects of Financial Innovations on C heckable Deposits, M1 and M2,” th is Review (July/August 1990), pp. 37-57. ________ . “Are T he M acroeconom ic Effects of Oil Price C hanges S ym m etric?” in Karl B runner and Allan H. M elt zer, eds., Stabilization Policies a nd Labor Markets, C arnegie-R ochester C onference Series in Public Policy (North Holland, 1988a), pp. 325-68. ________ . “ M acroeconom ic Effects of the 1986 Oil Price De cline,” Contem porary Policy Issues (July 1988b), pp. 69-82. ________ . “ T he M acroeconom ic Effects of the Recent Fall in Oil Prices,” th is Review (June/July 1987), pp. 34-45. ________ . “ Energy Prices and Short-Run Econom ic Perfor mance,” this Review (January 1981), pp. 3-17. Trehan, Bharat. “ Lessons from the Oil Shocks of the 1970s,” Federal Reserve Bank of San Francisco Weekly Letter (Novem ber 9, 1990). W eidenbaum , Murray. “ Recessionary Signs Were Evident Even Before Oil C risis,” Christian Science Monitor, October 16, 1990. Yanchar, Joyce. “A Look at Oil Market Fundam entals,” Spe cial Study, DRI/M cG raw-Hill U.S. Review (August 1990), pp. 25-31. 19 Michelle R. Garfinkel and Daniel L. Thornton M ichelle R. Garfinkel, an assistant professor at the University o f California at Irvine, was a senior econom ist a t the Federal Reserve Bank of St. Louis while this p a p er was written. Daniel L. Thornton is an assistant vice president at the Federal Reserve Bank of St. Louis. R ichard I. Jako provided research assistance. Alternative Measures of the Monetary Base: What Are the Differences and Are They Im portant? HE MONETARY BASE, adjusted for changes in reserve requirements, is a measure intended to summarize the net effect of all Federal Re serve actions on the money stock.1 As such, it serves as an indicator of the effects o f monetary policy actions on the money stock. Because of the Federal Reserve Bank o f St. Louis' long-standing interest in monetary policy, it began publishing a series on the adjusted monetary base in August 1968.2 Nearly 11 years later, the Federal Reserve Board began publish ing an alternative series. While the objective of 1The idea of adjusting th e m onetary base to reflect changes in reserve requirem ents was proposed initially by Karl B runner (1961) in an effort to fo rm u la te an “ e m pirically significant th e ory” of the m oney supply process. B runner called this adjustm ent “ liberated reserves.” He was the first to com pile data on the adjusted m onetary base and em pirically investigate the relationship between his m eas ure and the m oney supply. T his research agenda was pur sued by both him and Allan M eltzer in a num ber of articles dealing with the m oney supply process and m onetary policy. 2See Andersen and Jordan (1968). 3One of the m ost dram atic changes in the structure of reserve requirem ents occurred with the im plem entation of the Board’s series is the same, it has always differed from the series constructed by the St. Louis Fed in a number of respects. These differ ences have changed over time with changes in the structure of reserve requirements and, thus, changes in the methods of calculating the re spective series.3 The two series have been used separately and, on occasion, jointly to address a number of issues of importance in monetary theory and policy. Occasionally, they have yielded signifi cantly different results.4 Moreover, at times the M onetary Control Act of 1980 (MCA). G arfinkel and Thornton (1989) discuss the changes in the structure of reserve requirem ents brought about by the phase-in o f the MCA. G arfinkel and T hornton (1991) discuss the effects of the MCA on the m oney supply process and the usefulness of the m onetary base as an indicator of th e effect of m on etary policy on the m oney stock. See Burger and Rasche (1977) and G ilbert (1987) for tw o significant changes in the calculation of the St. Louis adjusted m onetary base. “ See Friedm an (1988), M cCallum (1988a,b), Haslag and Hein (1990) and M uelendyke (1990) for some exam ples of differential perform ance. NOVEMBER/DECEMBER 1991 20 Figure 1 Levels of the STL-AMB and BOG-AMB, Seasonally Adjusted Billions of dollars Billions of dollars Data plotted from January 1959 thru April 1991. they have presented conflicting pictures of mon etary policy. Because of changes in their calculation and the recent conflicting results, now seems to be an appropriate time to re-examine these series to see how and why they differ, both historically and currendy. W e can also investigate whether they are likely to continue to behave differently in the years ahead and whether their differential performance is attributable to fundamental differ ences or merely to arbitrary differences in their construction. Finally, we will provide preliminary evidence on whether the existing difference is potentially important for money stock control. THE DIFFERENCE BETWEEN THE TW O ADJUSTED MONETARY BASE MEASURES The adjusted monetary base series constructed by the St. Louis Fed (hereafter labeled STL- FEDERAL RESERVE BANK OF ST. LOUIS AMB) and that constructed by the Board of Governors of the Federal Reserve System (here after labeled BOG-AMB) were designed for the same purpose. Each is intended to isolate the ef fects of monetary policy actions—that is, changes in the supply of reserves and changes in reserve requirements—in a single measure. Neverthe less, the two series are quite different. H o w D o They Differ? The difference in their behavior can be seen in figure 1, which shows both adjusted mone tary base measures, seasonally adjusted, from January 1959 to April 1991. Although the two measures behave similarly throughout the sam ple period, STL-AMB is always larger than BOGAMB and the spread between them increases over time. As will be discussed later, much of this spread can be attributed to differences in the reserve requirement measure used to calcu late the two series. Currently, this difference is 21 Figure 2 Difference Between the Growth Rates of STL-AMB and BOG-AMB Percent 0.10 Percent 0.10 0.05 0.05 0.00 0.00 - 0 .0 5 -0 .0 5 - 0.10 - 1960 62 64 66 68 70 72 74 76 78 80 82 84 86 88 0.10 1990 Data plotted from February 1959 thru April 1991. due to the weight each assigns to the level of transaction deposits. As a consequence, the re cent widening in the spread between the two AMB series is driven by the growth of deposits.5 Although the difference in the level of the two series gets larger over time, figure 2 shows that the difference in the monthly compounded annual growth rates of the two adjusted mone tary base measures does not exhibit a signifi cant trend. Indeed, although the monthly dif ference in the growth rates ranges from -8.4 percent to 8.6 percent, the average difference 5Thus, the series should be strongly cointegrated. Haslag and Hein (1990), using data from 1959 through 1989, find that the two adjusted m onetary base m easures are coin tegrated. T heir results, based on a procedure suggested by Engle and G ranger (1987), indicate that th e null hypoth esis of no cointegration can be rejected at the 5 percent level. This hypothesis, however, cannot be rejected at the 1 percent level. Nevertheless, alternative tests for coin tegration, using a procedure developed by Johansen (1988), add fu rth e r support to the notion th a t the two ser in their growth, 0.18 percent, is not significantly different from zero at the 5 percent significance level (the t-statistic is 1.35). Thus, over a suffi ciently long period, the growth rates of the two series are nearly identical. Over shorter periods, however, the differ ences in the growth rates of these aggregates persist, as illustrated by a six-month moving average of the difference between the growth rates of the two series—presented in figure 3. The six-month moving average, which ranges from -3.6 percent to 3.4 percent, shows that, ies are cointegrated [This procedure and others are dis cussed in Dickey, Jansen and T hornton (1991).] The chisquare statistic for the null hypothesis that there is one cointegrating vector is 74.5, com pared w ith a critical value at the 1 percent significance level o f 22, w hich provides evidence that the two m onetary base series are cointegrat ed, as expected. NOVEMBER/DECEMBER 1991 22 Figure 3 A Six-Month Moving Average of the Difference Between the Growth Rates of STL-AMB and BOG-AMB Percent Percent 1960 62 64 66 68 70 72 74 76 78 80 82 84 86 88 1990 Data plotted from July 1959 thru April 1991. for periods as long as six months, the two ad justed monetary base measures can give con siderably different pictures of monetary policy. In fact, as an examination of figure 3 shows, the growth rates of the two series can differ significantly even for fairly long periods o f time. For example, the six-month moving average of the difference in the growth rates was strictly positive before April 1969 and nearly always negative after August 1984. The difference in the monthly growth rates averaged 0.60 percent during the former period and -0.32 percent during the latter, and both differences are statistically significant (the t-statistic is 7.96 in the former period, 7.04 in the latter). 6O nly the STL refers to th is adjustm ent as RAM — short for reserve adjustm ent m agnitude. For expository conve nience, th is article also w ill refer to th e BO G ’s adjustm ent for reserve requirem ents as RAM. FEDERAL RESERVE BANK OF ST. LOUIS Why D o They Differ? Much of the difference in the two series is at tributable to the method of adjusting for reserve requirement changes. Each adjustment creates an index o f reserves that would have been held during some “base period.” The magnitude of this adjustment, hereafter referred to as RAM, reflects the reserves that are absorbed (released) if the reserve requirements w ere higher (lower) than those in effect during the base period.6 When the STL series was first created, a bank’s reserve requirements depended on its lo cation, the type of deposits it held (transaction vs. non-transaction) and whether it was a mem- 23 ber of the Federal Reserve System.7 The adjust ment for reserve requirement changes was made by multiplying the deposits in a given reserve category by the difference between that cate gory’s reserve requirement in the base period and its corresponding reserve requirement in the current period. The initial base period used was from August 1935 to July 1936.8 In contrast, the BOG has always used the cur rent period as the base period in calculating its reserve adjustment. Thus, each time reserve requirements change, the BOG revises the his torical data to reflect the “current” system of reserve requirements. For example, when re serve requirements are reduced, the BOG calcu lates the amount of reserves that would have been required had the lower reserve require ment been in effect previously. The actual level of required reserves in past periods are multi plied by the ratio of the new average reserve requirement to the old, thereby creating a new, counterfactual “adjusted reserve” series. In this example, the new historical series for adjusted reserves would be lower than the previous historical series. The lower level of the "new ” historical series relative to the current period’s levels reflects a hypothetical release of reserves in the past brought about by the decrease in re quirements in the current period. In essence, both the STL and BOG methods cre ate counterfactual series for adjusted reserves. For STL, the counterfactual series is the reserves that would have been held if the historical re serve requirements w ere in effect today. Hence, if reserve requirements have been reduced (raised), actual reserves would be lower (higher) than adjusted reserves. For the BOG, the coun terfactual series is the reserves that would have 7Prior to MCA, non-m em ber banks did not have to m aintain reserves w ith the Federal Reserve System . At the tim e the Federal Reserve A ct was passed, reserve requirem ents were diffe re n t for “ central reserve c ity ,” “ reserve c ity ” and “ c o u n try” banks. This d istinction was not based ex p licitly on the size of the institution, but on the location of the bank at the tim e the Federal Reserve A ct was passed. O ver tim e, th is classification system becam e less m eaning ful, with m any large banks classified as country banks. 8See Burger and Rasche (1977), G ilbert (1980) and Tatom (1980). been held in the past if the current reserve re quirements w ere in effect. Actual reserves held in the past would be higher (lower) than ad justed reserves if reserve requirements were reduced (raised) in the current period. With the STL approach, it is not necessary to revise the historical monetary base series each time reserve requirements are changed. As not ed above, the BOG series, in contrast, requires that all historical data be revised, which means there is a delay in the availability of data. Con sequently, the STL approach has a comparative advantage for empirical research over the BOG approach because it produces a series that is readily available at the time it is most needed— when there is a change in reserve requirements.9 Beyond this advantage, the base-period distinc tion between these approaches appears to be unimportant. A Recent Change in the Construc tion o f the STL Series In 1987 the STL adjusted monetary base ser ies was revised in response to fundamental changes in the structure of reserve require ments associated with the Monetary Control Act (MCA) of 1980. Currently, the series is obtained by splicing two adjusted monetary base series with RAMs based on different systems of reserve requirements. Before November 1980, RAM is based on average ratio of reserves to deposits, for transaction and non-transaction deposits, during the period from January 1976 to August 1980. After November 1980, RAM is based on the structure of marginal reserve re quirements on transaction deposits in effect un der the MCA. These series are spliced together at the first reserve maintenance period (Novem- d ifficu lt or im possible to continue using the o riginal base period. A t these tim es, the historical data were revised. The first of these changes occurred in 1977 when Burger and Rasche (1977) altered the series by both changing the m ethod used to calculate RAM and adjusting the series to account for th e sig n ifica n t ch ange in the structure of reserve requirem ents in Novem ber 1972. The next oc curred in D ecem ber 1980 w hen the base period was changed to the period from January 1976 to A ugust 1980. The m ost recent, discussed next, was in 1987. 9Although reserves absorbed or released by changes in reserve requirem ents typ ica lly are offset through open m arket operations so that there is no m arked change in the adjusted m onetary base, RAM does change sig n ifica n t ly. It should be noted, however, th a t the base period for the St. Louis series has changed when fundam ental changes in the structure o f reserve requirem ents m ade it NOVEMBER/DECEMBER 1991 24 ber 19, 1980) under the new reserve require ments imposed by the MCA.10 At the splice date, the second part o f the ser ies is calculated under the assumption that the marginal reserve requirements on reservable categories of time and savings deposits are zero.11 Reserve requirements on all non-transaction deposits w ere eliminated in December 1990. Consequently, after November 1980, the STL series utilizes the present structure of re serve requirements for its base period. Because the base period for the BOG’s series is always the current one, the "base-period” distinction between the two series has been virtually elimi nated for the period since November 1980.12 This distinction remains relevant for the preNovember 1980 data, however. Another distinction remains, however. STL calculates its RAM using the marginal reserve requirement on transaction deposits, 12 percent, while the BOG uses the average reserve require ment ratio at the time of the last change in reserve requirements. Currently, this ratio is about 8 percent for transaction deposits.13 As a result, the level of STL-AMB is larger than that of BOG-AMB. DETERMINING THE SOURCE OF THE DIFFERENCES BETWEEN THE TW O SERIES Although the difference in adjustment methods for reserve requirement changes accounts for much of the difference exhibited by the two monetary base series, it is not the entire source of their differential behavior. There are two other potentially significant sources: the treat 10The procedure scales th e “ o ld e r” part of the series down to the level of the “ new er” part of the series to reach the level consistent w ith the post-M CA base period. The growth rates of the data before the splice date are unaffected by the change in the base-period for the level data. See G il bert (1987), p. 26, for a detailed discussion o f the procedure. " T h is was done because the data necessary for m aking a RAM adjustm ent for non-transaction deposits were not available. See G ilbert (1987). 12The only base-period d istinction that rem ains is due to the fact that the G arn-St. G erm ain Depository Institutions Act of 1982 requires that a certain level of transaction deposits at each depository institution be subject to a reserve re quirem ent of zero and that this rate be adjusted upward with the rise in total reservable liabilities. 13Because the average ratio of reserves held against tran s action deposits to transaction deposits for th e period from FEDERAL RESERVE BANK OF ST. LOUIS ment of vault cash and the seasonal adjustment methods used. Each of these three sources and their empirical importance is discussed below. (A more detailed discussion of the current con struction o f the two series and their differences appears in the appendix.) The Treatment o f Vault Cash Currently, the two adjusted monetary base series start with slightly different “raw ” data. Both unadjusted monetary base measures, roughly speaking, are the sum o f reserve balances held by depository institutions at Fed eral Reserve Banks and currency in circu lation—in other words, currency held by the public, including depository institutions. The differences in these raw data lies primarily in the treatment of vault cash—that is, cash held by depository institutions in their vaults. The BOG, in contrast to STL, adjusts its series for the timing of reserve requirements as satisfied by depository institutions with vault cash. To get a better understanding o f this differ ence, it is helpful to review briefly the Federal Reserve’s system of reserve requirements. Un der the current system, depository institutions are required to hold reserves in the form of vault cash and/or reserve balances at the Feder al Reserve equal to a fixed percentage of their reservable deposit liabilities—specifically, trans action deposits held by the public, government and foreigners.14 A depository institution’s re quired reserves are calculated on the basis of the transaction deposits it holds during a twoweek period ending every other Monday. An in stitution can satisfy its requirements with de posit balances at the Federal Reserve during the two-week reserve-maintenance period ending January 1976 to A ugust 1980 was 0.12664, the use of the m arginal reserve requirem ent, rather than the average ra tio of reserves to deposits previously used, m inim izes the difference between the " o ld e r ” and “ new er” series at the splice date. 14There are low er reserve requirem ents on a tranche of deposits for each depository institution. Also, depository in stitu tio ns are required to m aintain reserves against th e ir net checkable deposits with o ther institutions. Aggregated over all institutions, however, the net deposits are zero. Consequently, in the aggregate, no reserves are held against checkable deposits w ith other depository in stitu tions. Because reserve balances held against such deposits do not net out to zero for individual institutions, however, reserves held against net “ in te r-b a n k” checkable deposit lia bilities affects the d istribution of required reserves am ong these institutions. 25 two days after the period used in computing its reserves, and with vault cash held during a two-week period ending 30 days before the end of the maintenance period. Vault cash used to satisfy statutory reserve requirements is called "applied vault cash.” The BOG's adjustment for vault cash involves a distinction between "bound" institutions, whose statutory reserve requirements exceed their holdings o f vault cash, and "non-bound” institu tions, whose vault cash exceeds their statutory reserve requirements. Another important dis tinction is between weekly reporting institutions, called EDDS, and quarterly reporting institu tions, called QEDS. In the BOG’s adjustment for vault cash, the difference between current vault cash and lagged (applied) vault cash of bound EDDS (see appendix) is excluded from the BOG's raw monetary base series. Hence, the STL and BOG unadjusted, not-seasonally-adjusted series effectively differ by the change in vault cash of bound EDDS.15 What is the empirical importance of this differ ence? Because of limited data availability, an exact measure o f the magnitude of the BOG's adjust ment to vault cash cannot be obtained over the entire sample. As a proxy, the difference be tween the STL source base and the BOG’s notbreak-adjusted monetary base (with no seasonal adjustments) is used.16 This measure, denoted here by ATVC, is depicted in figure 4 for the full sample period from February 1959 to April 1991. As the figure shows, the difference in the treatment of vault cash fluctuates between -$2.3 billion to $1.7 billion, but is positive for most of the sample period. Indeed, ATVC aver ages $.15 billion over the period. While this average is small, both in absolute terms and relative to the difference in the two base meas ures, it is statistically significant from zero at the 5 percent level (the t-statistic is 7.82). As discussed in more detail below, the differ ence in the treatment of vault cash has a pro 15They also d iffe r because of “ as-of” adjustm ents, w hich the BOG m akes but STL does not, and because the BOG in cludes “ required clearing ba la n ces” in its not-breakadjusted series. The rationale for the BO G ’s adjustm ent to vault cash is not entirely clear. A recent Board m em o states that the ad justm ent is made on th e belief that current vault cash con strains the lending activities of non-bound institutions, w hile lagged vault cash is the relevant constraint for bound institutions. 16As noted in the appendix, required clearing balances and “ as-of” adjustm ents are included in the B oard’s not-break- nounced seasonal pattern, which has become more amplified over time. Thus, although the difference in the treatment of vault cash con tributes relatively little to the "low-frequency” (quarterly or annual) variation in the difference between the two series, it contributes somewhat to the "high-frequency” (monthly) variation. The R eserve Adjustment Before November 1980, there are two basic differences between the adjustments that STL and the BOG use on the raw data for reserve requirement changes. First, as noted previously, STL uses a fixed, historical period, while the BOG uses the current period as the base period. Because the average ratio of required reserves to deposits was substantially higher during this period than it is today, the STL adjustment be fore November 1980 is significantly larger than the BOG’s. Second, the BOG makes an additional adjustment for changes in reserve requirements on applied vault cash (see appendix for details). These two differences continue to be relevant after November 1980. Because STL uses the marginal reserve requirement o f 12 percent, which is larger than the average ratio of re serves to deposits used by the BOG to calculate its RAM, the levels o f the series are quite differ ent even after this date. As in the period before November 1980, the BOG makes, as part of its break-adjustment procedure, a separate adjust ment to applied vault cash. In addition, since February 1984, with the switch to contempora neous reserve accounting, the BOG makes a separate break adjustment to its adjustment for lagged vault cash of bound EDDS. The difference between the STL adjusted monetary base and the BOG’s break-adjusted monetary base can be used to measure the em pirical magnitude of differences in the method of adjusting for reserve requirement changes. ATVC is added to this measure to isolate the efadjusted m onetary base series, but not in the source base. (See table A1 of the appendix for details). Required clear ing balances and these “ as-of” adjustm ents, however, are not included in the B o a rd 's break-adjusted series and, thus, play no role in explaining the d ifference between the two adjusted m onetary base measures. To isolate the ef fect of the d ifference in the treatm ent of vault cash on the difference between the two series, required clearing balances and the as-of adjustm ents (when these data are available) are rem oved from the difference between the two unadjusted base series. NOVEMBER/DECEMBER 1991 26 Figure 4 Difference in the Treatment of Vault Cash ATVC Billions of dollars Billions of dollars 2 2 1960 65 70 75 80 Data plotted from January 1959 thru April 1991. feet of the difference in RAM from that o f the BOG’s different treatment of vault cash. The resulting series, denoted here by ARAM, and the difference between the two seasonally adjusted, adjusted monetary base measures, denoted by AAMBSA, are presented in figure 5. The figure shows that most of the difference between the levels of the two seasonally adjusted bases is, in fact, explained by differences in the method used to adjust for reserve requirement changes. Seasonal Adjustment To remove regular variations in the AMB ser ies due to seasonal factors, both series are sea 17See Zeller (1972). Also, see G ilbert (1985) fo r a discussion of the change in seasonal adjustm ent associated with the sw itch to contem poraneous reserve accounting. FEDERAL RESERVE BANK OF ST. LOUIS sonally adjusted. For monthly and quarterly data, STL adjusts its monetary base series by simply applying the standard X -ll seasonal ad justment program to its not-seasonally-adjusted series. Weekly data are seasonally adjusted with a separate program that inputs unadjusted weekly data and seasonally adjusted monthly data.17 The BOG seasonally adjusts weekly data using a model-based approach; it then obtains season ally adjusted monthly and quarterly data from the seasonally adjusted weekly data.18 In con trast to STL, the components o f the base are seasonally adjusted separately: break-adjusted required reserves against transaction deposits, the break-adjusted measure of surplus vault 18See Pierce, G rupe and C leveland (1984) and Farley and O ’ Brien (1987) for a discussion of the seasonal adjustm ent procedures used by the Board. 27 Figure 5 Difference Between STL-AMB and BOG-AMB and the Difference Between Their Reserve Adjustments, AAMBSA and ARAM Billions of dollars Billions of dollars Data plotted from January 1959 thru April 1991. cash used by the BOG, and currency held by the nonbank public.19 The difference in the two series due to differ ent seasonal adjustment procedures is presented in figure 6. This difference, denoted by ASAM, is measured by subtracting the BOG’s seasonal adjustment (the difference between the seasonally adjusted and the not-seasonally-adjusted, breakadjusted monetary base) from STL’s seasonal ad justment (the difference between the seasonally adjusted and the not-seasonally-adjusted, adjust ed monetary base). Not surprisingly, the average difference in the series due to differences in seasonal adjustment is essentially zero over the sample period.20 Nevertheless, ASAM ranges 19AII com ponents of the m onetary base, excluding excess reserves, are adjusted as a whole before the switch to contem poraneous reserve accounting. For the series after the switch, the w eekly series is adjusted by com ponent us ing a m odel-based procedure and then is m odified to be m ade consistent with the m onthly series. from -$1.6 billion to $2.6 billion, suggesting that differences in the seasonal adjustment are a source of high-frequency variation in the dif ference between the two series. The Relative Im portance o f These Differences O ver Time As indicated by figures 4 and 6, the contribu tion of the differences in the treatment of vault cash and the seasonal adjustment have become more variable starting shortly after the MCA, especially around 1984. Moreover, after 1984, both differences have large seasonal compo nents. ATVC becomes larger and more variable, perhaps because the BOG's adjustment for lagged 20The average difference is $.025 billion, with a t-statistic of 1.14 for the test of the null hypothesis th a t the difference is zero. NOVEMBER/DECEMBER 1991 28 Figure 6 Difference in the Seasonal Adjustment ASAM Billions of dollars 3 Billions of dollars 3 -1 -1 -2 1960 65 70 75 80 85 1990 Data plotted from January 1959 thru April 1991. vault cash at bound EDDS does not change with the Fed's move from lagged to contemporaneous reserve accounting. Separate break adjustments are made to this series and to lagged vault cash at QEDS starting in February 1984, however. The difference due to seasonal adjustments becomes larger with higher frequency variation about this same time. This may be the result of the Board’s changing its seasonal adjustment method in February 1984.21 The seasonal variations in these two series tend to offset each other so that the difference between STL-AMB and BOG-AMB does not have 21As noted above, the BOG seasonally adjusts its breakadjusted m onetary base as a whole for data before the sw itch to contem poraneous reserve a ccounting in February 1984. Thereafter, it has used a m odel-based approach to FEDERAL RESERVE BANK OF ST. LOUIS a large seasonal component. Indeed, ATVC and ASAM are highly negatively correlated after 1984—the simple correlation between changes in the two series after January 1984 is - .84. As shown in table 1, which reports the vari ances of AAMBSA, ARAM, ATVC, ASAM and (ATVC + ASAM) for various subperiods, the varia bility in AAMBSA has also increased over the entire sample period. The subperiods corre spond to various reserve requirement regimes. The full sample is broken at the time of the im plementation of the MCA and the time of its ef fective completion, which coincides with the switch to contemporaneous reserve accounting. seasonalize the com ponents of the base separately for w eekly data. In addition, as discussed by G ilbert (1985), th e seasonal factors used by the St. Louis Fed also changed around that tim e. 29 Table 1 The Variances of Changes in the Difference between STL-AMB and BOG-AMB and Changes in its Sources_________________ AAMBSA ARAM 1959.02-1991.04 .093 1959.02-1980.11 1980.12-1991.04 1980.12-1984.01 1984.02-1991.04 Period ATV C + ASAM ATVC ASAM .034 .245 .290 .104 .033 .216 .023 .058 .016 .725 .019 .858 .025 .269 .120 .258 .072 .051 .069 1.011 .119 1.181 .094 .346 Table 2 Simple Correlations between Changes in the Difference between STL-AMB and BOG-AMB and Changes in its Sources________ Period ARAM ATVC ASAM ATV C + ASAM 1959.02-1991.04 .206 .175 .334 .826 1959.02-1980.11 1980.12-1991.04 .563 .022 .233 .177 .488 .333 .614 .887 1980.12-1984.01 1984.02-1991.04 .528 -.1 5 6 .227 .180 .421 .334 .669 .925 As the table shows, the variance of changes in the differences of the two seasonally adjusted AMB series has increased throughout the sam ple period, especially since February 1984, when the variances of both ATVC and ASAM sharply increase. Because these series are negatively correlated, however, this increased variability is not reflected in a similar rise in the variance of these combined series (ATVC + ASAM). Neverthe less, the simple correlations between changes in AAMBSA and both changes in ARAM and changes in (ATVC + ASAM) presented in table 2 show that more of the month-to-month changes in AAMBSA is attributable to changes in ATVC + ASAM since January 1984. ALTERNATIVE MEASURES OF THE MONETARY RASE: ARE THE DIFFERENCES IMPORTANT? 22G arfinkel and Thornton (1991) have shown that the relationship between th e m oney supply and the adjusted m onetary base has weakened since the MCA. More im por- tant, they argue that the usual linear relationship between the m oney su p p ly and the m onetary base, as a m odel of the m oney supply process, fa ils to perform well since then. Since the adjusted monetary base is intended to be a summary measure of the policy actions of the Federal Reserve, an important question arises: are the differences in the STL and BOG measures of the adjusted monetary base important?22 Over a sufficiently long period of time, the answer to this question is an emphatic, “No!” As noted earli er, over sufficiently long periods of time, the average difference in the growth rates of the two series is negligible. Nevertheless, over shorter NOVEMBER/DECEMBER 1991 30 Table 3 Estimate of Changes in M1 on Changes in Alternative Measures of the Adjusted Monetary Base Monthly Results____________ St. Louis Period Constant AAMB 1959.02-1980.10 .080 (0.89) 1980.11-1984.02 Board R2 D.W . Constant AAMB 2.210* (13.86) .424 1.69 .169 (1.75) .730 (0.96) 2.025* (3.37) .210 2.11 1984.03-1991.04 .344 (0.53) 2.135* (6.11) .299 1984.03-1990.11 - 1 .1 5 6 (1.56) 3.244* (7.32) .396 R2 D.W . 2.237* (11.81) .348 1.53 .444 (0.51) 2.594* (3.23) .195 2.15 1.12 -.0 3 9 (0.05) 2.480* (5.51) .257 1.03 1,27 - 2 .2 3 9 * (2.64) 4.139* (7.58) .414 .81 * Indicates statistical significance at the 5 percent level. periods, the two measures could lead to different interpretations of monetary policy. The adjusted monetary base has been used fre quently in theoretical and empirical models of money stock control. Hence, one way to assess whether the difference in the two AMB measures is important is to see if either measure explains more of the movements in the money stock, M l.23 Some preliminary evidence on the relative per formance of these alternative measures in con trolling the money stock can be obtained by regressing changes in M l on changes in the alter native base measures. Estimates of these equa tions using monthly data are reported in table 3.24 The results are reported for three subperiods from February 1959 to April 1991 based on impor tant changes in the construction of the two ser ies. The results indicate that, in all cases, there there is a statistically significant relationship be23M2 is not included in the analysis because, since Decem ber 1990, there is no d irect relationship between th e nonM1 com ponents of M2 and policy actions. M2 can only be controlled directly through its M1 com ponent. Equations sim ilar to those reported here, using M2, produce results broadly sim ilar to those here, except w here noted below. 24These equations are sim ple and are not intended to be the specification of the m oney su p p ly process over the tim e period. M oreover, as G arfinkel and T hornton (1991) have shown, such a sim ple specification of the m oney supply function is inappropriate since February 1984. The strong FEDERAL RESERVE BANK OF ST. LOUIS tween changes in M l and changes in each of the base measures.25 The relationship between M l and either adjusted monetary base appears to have deteriorated since the effective implementa tion of the MCA in February 1984. This result ap pears to be due, however, to the sharp rise in depository institutions’ holdings of excess reserves following the December 1990 elimination of reserve requirements on non-transaction deposits and the sensitivity of ordinary-least-squares regression analysis to outliers. When the last five months of data are deleted, the adjusted R-squares are similar to those obtained over the first period—indeed, the adjusted R-square for the BOG’s measure rises by nearly 20 percent.26 The relationship between M l and the adjusted monetary base is somewhat tighter when the STL measure is used for all three periods; this is not the case when the last five months are deleted serial correlation of the residual since then is evidence of th is m isspecification using either base m easure. For these reasons, the results here are intended to be illustrative. 25The equations were also estim ated using the growth rates of M1 and th e two AM B m easures. In all instances, the ad justed R-squares were sm aller when grow th rates were used. Q ualitatively, the results are the sam e as those in th e te xt when q u arterly and annual data are used. 26As expected, after February 1984, there is no sta tistically significant relationship between M2 and e ith e r base m eas ure, at the m onthly or q uarterly frequency. 31 Table 4 Estimates of Changes in M1 on Changes in Alternative Measures of the Adjusted Monetary Base Quarterly and Annual Results St. Louis Period Constant Board AAMB R* D.W , Constant AAMB R2 D.W . QUARTERLY RESULTS .151 (0.75) 2.254* (16.77) .774 1.52 I/80-I/84 - 1 .2 5 0 (0.50) 3.086* (4.18) .507 2.32 11/84-1/91 .850 (0.23) 2.148* (3.02) .231 II/84-III/90 - 8 .0 9 6 * (2.14) 4.406* (5.34) .524 II/59-IV/79 2.266* (14.99) .732 1.41 - 1 .8 5 8 (0.58) 3.678* (3.45) .406 2.54 .37 1.892 (0.47) 2.014* (2.44) .156 .41 .42 - 7.089 (1.55) 4.360* (4.15) .394 .46 - .8 2 9 (0.26) 2.85* (8.75) .716 .428* (2.06) ANNUA L RESULTS 1959-90 -2 .3 2 1 (0.77) 2.794* (9.72) .757 .82 .77 * Indicates statistical significance at the 5 percent level. from the last period.27 Overall, the two meas ures appear to differ little in their relationship to monthly M l growth. Because the difference in the growth rates of the two AMB series declines at lower frequen cies, the equations also were estimated using quarterly data and annual data for the period from 1959 to 1990. These results are summa rized in table 4. The quarterly estimates are similar to the monthly ones; the overall perfor mance is better, however, using quarterly data during the first period and somewhat worse during the last. Again, much of the deteriora tion in the last period is due to including the quarters during and immediately after the elimi nation of reserve requirements on non-transaction deposits. For quarterly data, the St. Louis measure al ways explains somewhat more of the variation in M l growth. Estimates using annual data pro vide a similar result, with the STL series ex plaining about 4 percent more than the BOG series of the annual variation in M l. CONCLUDING REMARKS While no formal tests of the difference in the performance of the two base measures for money stock control have been made here, the differ ences might be statistically significant. Indeed, this has been the case in other applications. For example, Haslag and Hein (1990) have reported statistically significant differences in the explan atory power of the two monetary bases for nominal GNP. More importantly, Friedman (1988) and McCallum (1988a,b) report substantive dif ferences in the performance of the two meas ures for monetary policy analysis. The problem is that the two measures differ by their reserve adjustments, their treatment of vault cash and their methods of seasonal adjust ment. While differences in the reserve adjust ment procedures account for the bulk of the discrepancies, differences arising from the treat ment of vault cash and seasonal adjustments have become more important in explaining short-run variations between the two series in the 1980s. Because there is little objective rea- 27Also, the B O G ’s m easure produces a higher adjusted R-square than does the STL m easure in the last period w hen m onthly growth rates are used. NOVEMBER/DECEMBER 1991 32 son to prefer one method of technical adjust ment over the other, there is little basis for choosing one measure over the other in empiri cal studies, when the two measures produce substantially different results. In instances where the results are qualitative ly the same but quantitatively different, such as the results reported here or by Haslag and Hein, the researcher must be content to choose the measure that performs "best” for the problem at hand. If the problem is money stock control, the preliminary evidence presented here sug gests that the St. Louis measure holds an edge. G arfinkel, M ichelle R. and Daniel L. Thornton. “ The M ultiplier Approach to The Money S upply Process: A Precautionary Note,” this Review (July/August 1991), pp. 47-64. ________ .“ The Link Between M1 and the M onetary Base in the 1980s,” th is Review (S eptem ber/O ctober 1989), pp. 35-52. G ilbert, R. Alton. “A Revision in the M onetary Base,” this Review (August/Septem ber 1987), pp. 24-29. ________ . “ New Seasonal Factors for the Adjusted Monetary Base,” this Review (Decem ber 1985), pp. 29-33. ________ . “ C alculating the Adjusted M onetary Base under C ontem poraneous Reserve Requirem ents,” this Review (February 1984), pp. 27-32. ________ . “ Two M easures of Reserves: W hy Are They Dif ferent?” this Review (June/July 1983), pp. 16-25. ________ . “ Revision of the St. Louis Federal Reserve’s Adjusted M onetary Base,” this Review (D ecem ber 1980), pp. 3-10. REFERENCES Andersen, Leonall C. and Jerry L. Jordan. “ The M onetary Base-Explanation and Analytical Use,” th is Review (August 1968), pp. 7-11. Brunner, Karl. “A Schem a for the S upply T heory of Money,” International Econom ic Review (January 1961), pp. 79-109. Burger, Albert E. “Alternative Measures of the M onetary Base,” this Review (June 1979), pp. 3-8. Burger, Albert E., and Robert H. Rasche. “ Revision of the Monetary Base,” this Review (July 1977), pp. 13-28. Dickey, David A., Dennis W. Jansen and Daniel L. Thornton. “A Prim er on C ointegration w ith an A pplication to Money and Income,” this Review (M arch/April 1991), pp. 58-78. Engle, R obert T., and C. W. J. Granger. “ Co-Integration and Error C orrection: Representation, Estimation and Testing,” Econom etrica (March 1987), pp. 251-76. Farley, Dennis E., and Yueh-Yun C. O ’Brien. “ Seasonal Ad justm ent of the M oney Stock in the United States,” Journal o f O fficial Statistics (1987), pp. 223-33. Friedm an, B enjam in M . “C onducting M onetary Policy by C ontrolling C urrency Plus Noise: A Com m ent,” CarnegieRochester Conference Series on Public Policy (Autumn 1988), pp. 205-12. Haslag, Joseph H., and Scott E. Hein. “ U.S. M onetary Policy M easures: Are They Roughly Equivalent?” m imeo, Federal Reserve Bank of D allas (1990). Johansen, Soren. “ Statistical Analysis of C ointegration Vec tors,” Journal o f Econom ic Dynam ics and Control (June/Septem ber 1988), pp. 231-54. M cCallum , Bennett T. “ R obustness Properties of a Rule for M onetary Policy,” Carnegie-Rochester Conference Series on Public Policy (Autum n 1988a), pp. 173-204. ________ . “ Reply to Com m ents by Benjam in Friedm an,” Carnegie-Rochester Conference Series on Public Policy (Au tum n 1988b), pp. 213-14. M eulendyke, Ann-M arie. “ Possible Roles for the M onetary Base,” in Intermediate Targets a nd Indicators for M onetary Policy: A C ritical Survey (Federal Reserve Bank of New York, July 1990) pp. 20-66. Pierce, David A., M ichael R. G rupe, and W illiam P. Cleve land. “ Seasonal Adjustm ent of the W eekly M onetary Ag gregates: A M odel-Based Approach,” Journal o f Business a n d Econom ic Statistics (July 1984), pp. 260-70. Tatom, John A. “ Issues in M easuring an Adjusted M onetary Base,” this Review (Decem ber 1980), pp. 11-29. Zeller, Louis. “ W eekly Seasonal Adjustm ent Program for the IBM S/360 Computer,” unpublished m anuscript, Board of Governors of the Federal Reserve System (February 15, 1972). Appendix The Construction of the T w o Adjusted Base Meas ures Since February 1984 This section discusses how each series, not seasonally adjusted, is constructed in two steps: the raw data series (called "source base” by STL them. Similarly, table A2 shows the adjustments made to each series and their differences. The R aw Data and "not-break-adjusted monetary base” by the BOG) and the adjustment for reserve require ment changes. Table A1 shows the construction of the STL source base and the analogous BOG series and summarizes the differences between FEDERAL RESERVE BANK OF ST. LOUIS The St. L o u is Series. The STL source base measure is the sum of reserve balances at Fed eral Reserve Banks and currency in circula tion—i.e., currency held by the public. As shown in panel A of table A l, reserve balances are de 33 Table A1 Calculating the St. Louis Source Base and the Board’s Not-Seasonally-Adjusted, Not-Break-Adjusted Monetary Base (since February 1984) A. STL: $290,759 = RESERVE BALANCES = total reserve balances - required clearing balances + CU RR ENCY IN CIRCULATIO N = currency held by th e nonbank public + current total vault cash 34.090 (-) 35.714 1.624 256.669 223.000 33.669 B. BOG: $292,131 = TO TAL RESERVES = total reserve balances - required clearing balances + applied vault cash + CU RR E N C Y HELD BY THE NO N BA N K PUBLIC + SUR PLUS VAULT CASH (BOG) = current total vault cash - applied vault cash at non-bound banks - applied vault cash at bound QEDS - current vault cash at bound EDDS + REQUIRED CLEARING BALANCES + FLOAT-PRICING AS-OFs 62.931 (-) 35.714 1.624 28.841 223.542 3.550 (-) (-) (-) 33.619 3.331 .196 26.543 (-) 26.543 25.314 1.624 .484 C. S T L - B O G - $ 1 .3 7 2 1 = - A D JU STM EN T FOR LAGGED VA U LT CASH AT BOUND EDDS = current vault cash at bound EDDS - applied vault cash (lagged) bound EDDS -R E Q U IR E D CLEARING BALANCES -F L O A T -P R IC IN G AS-OFs - .8 7 9 (-) (-) (-) - 1 .2 2 9 1.624 .484 NOTE: Exam ple data are for January 1990. C olum ns m ight not add due to rounding errors. 'T h is d ifference does not equal the sum of colum n 2, section C, because the B O G ’s figure for currency held by the nonbank p ublic and vault cash used in ca lcu la tin g th e ir adjusted m onetary base differ from th e ir published series. The published data are m onthly averages of daily figures, w hile the m onthly data used to ca lculate th e ir m onetary base are obtained by prorating data averaged over reserve m aintenance periods. fined as total reserve balances net of required clearing balances of depository institutions.1 The B o a rd o f G o v ern o rs’ Series. As shown in panel B o f table A l, the BOG’s not-breakadjusted monetary base is essentially the same as the STL source base. Roughly speaking, the not-break-adjusted monetary base is the sum of total reserves and currency. Total reserves used in the calculation is defined as the sum of total 'R e q uire d clearing balances are deposit balances that depository institutions are required to m aintain at the Fed eral Reserve to ensure that the dollar volum e of their check clearings and other transfers of funds through the Federal Reserve System are covered. These balances are subtracted from total reserve balances since they do not satisfy statutory reserve requirem ents and, hence, do not reserve balances, net of required clearing bal ances, and applied vault cash.2 Required clearing balances plus currency held by the nonbank public and surplus vault cash are added to the measure of reserves. In constructing this mone tary base series, the BOG defines "surplus vault cash” as current vault cash net of lagged (applied) vault cash held by both non-bound and bound depository institutions that report quarterly constrain a depository in stitu tio n ’s expansion of deposit liabilities explicitly. 2As noted in the m ain text, applied vault cash refers to the vault cash held by depository institutions during the twoweek period ending 30 days before the end of the current m aintenance period. NOVEMBER/DECEMBER 1991 34 Table A2 Calculating the St. Louis Adjusted Monetary Base and the Board’s Break-Adjusted Monetary Base, Not Seasonally Adjusted (since February 1984) $299,961 A. STL in period t - s : 290.759 9.202 = SO URCE BASE ( t - s ) + RESERVE A D JU STM EN T M AG NITUDE ( t - s ) = rT(b) x transaction deposits (t - s) - required reserves (t - s) $277,636 B. BOG: = N O T-BR EAK-ADJUSTED M O NETARY BASE ( t - s ) - FLOAT-PRICING AS-OFs ( t - s ) - REQUIRED CLEARING BALANCES ( t - s ) - REQUIRED RESERVES AG A IN S T EUR OD OLLARS ( t - s ) + AD JU STM EN T FOR THE BREAK IN REQUIRED RESERVES ( t - s ) = rT (t) x transaction deposits (t - s) - rT (t - s) x transaction deposits (t - s) - A D JU STM EN T FOR BREAK IN APPLIED VAULT CASH + A D JU STM EN T FOR THE BREAK IN LAGGED VA U LT CASH AT BO UND EDDS AND BO UND QEDS C. S T L -B O G : = RAM (STL) - A D JU STM EN T FOR + A D JU STM EN T FOR - A D JU STM EN T FOR EDDS AND BOUND - A D JU STM EN T FOR (-) (-) (-) (+ ) 292.131 .484 1.624 1.347 - 1 2 .4 4 0 (-) - 1 .4 9 2 (+) $22,325' THE BREAK IN REQUIRED RESERVES THE BREAK IN APPLIED VA U LT CASH THE BREAK IN LAG GED VAULT CASH AT BOUND QEDS LAGG ED VA U LT CASH AT BO UND EDDS -.0 8 8 21.467 (-) (+ ) 9.202 - 1 2 .4 4 0 -1 .4 9 2 (-) (-) -.0 8 8 - 1 .2 2 9 NOTES: Exam ple data are for January 1990. The series are constructed for periods i = t,t —1,t —2 ..... t - s ..... w here t is the most recent period. rT(b) is the reserve requirem ent in the base period (12 percent). rT(i) is the average reserve requirem ent on transaction deposits in period i. As in table A1, the colum ns m ight not add due to rounding errors. ’ This does not equal the sum of colum n 2, section C. (QEDS) and net of current vault cash held by bound institutions that report weekly (EDDS).3 As shown in panel C of table A l, the key difference between the STL source base and the BOG's unadjusted base measure is the treatment of vault cash held by bound EDDS. In particu lar, the two measures are essentially identical except that the Board’s measure excludes the surplus vault cash of bound EDDS—i.e., current vault cash net of lagged applied vault cash at bound EDDS. The two base series also differ in that the BOG’s measure does not remove re quired clearing balances and includes floatpricing "as-ofs.” As noted in the main text and 3The m easure of surplus vault cash reported on the B oard's H.3 release is not the m easure used in th is com putation. R ather than using current vault cash of all depository institutions, th e surplus vault cash in the H.3 release includes lagged vault cash of only those institu FEDERAL RESERVE BANK OF ST. LOUIS below, however, required clearing balances and float-pricing as-ofs are removed when the Board calculates its break-adjusted series. Adjustments f o r R eserve R equ ire ment Changes The St. Lou is Series. Panel A of table A2 shows the computation of the STL reserve ad justment magnitude, RAM, which is simply ad ded to the source base. RAM is the difference between actual required reserves and the re serves that would have been required, given the actual level and composition of deposits held at depository institutions, under the reserve re- tions subject to reserve requirem ents. In addition, rather than subtracting cu rre n t vault cash o f bound EDDS, the surplus vault cash reported in the H.3 release subtracts applied vault cash of these institutions. 35 quirements of a chosen base period. The reserve requirement ratio on transaction deposits cur rently used for the base period is 12 percent, the marginal reserve requirement in effect since the full implementation of the MCA.4 This adjusted base measure indicates what the monetary base would have been given the cur rent level of currency held by the nonbank public and the current level and composition of deposits under the base period's system of re serve requirements. If the required reserve ratio were to be reduced in any period, the adjust ment would increase.5 In this case, the source base would not change initially, but the adjusted monetary base measure would increase as RAM increased, reflecting a release o f reserves into the system available to expand deposit liabilities.6 The B oard o f G o v e rn o rs’ Series. The BOG adjusts the series for breaks in reserve require ments historically. In particular, it treats the most current period as its base period so that the break-adjusted series and the unadjusted series are identical. Reserves are adjusted with 4The actual average reserve ratio is low er than 12 percent because, fo r the individual bank, the reserve ratio is only 3 percent for transaction deposits below the reserve tranch and 12 percent fo r deposits in excess of that tranch. Even before last ye a r’s elim ination o f reserve requirem ents on personal tim e and saving deposits, which was com pleted in tw o steps starting in the m iddle of D ecem ber 1990 and ending after the first m aintenance period in January 1991, th e STL-RAM assum ed that base period requirem ents on non-transaction deposits w ere zero. It should also be not ed that both RAM s assum e that the base period require m ents on Eurodollar deposits is zero since February 1984. 5For example, the recent e lim ination of reserve require m ents on non-transaction d eposits released about $13.6 billion during the two (two-week) m aintenance peri ods starting in the m iddle of December. 6Typically, however, the Fed offsets the release of reserves generated by a reduction in reserve requirem ents, at least partially, by rem oving reserves from the banking system. Such a defensive m easure would prevent the m oney sup ply from accelerating in response to the reserve require- a ratio method. For any period in which the system of required reserves differs from that in the current period, required reserves held against these deposits are multiplied by the ratio of cur rent required reserves to what reserves would have been required under the old system, given the current composition of deposits. As shown in panel B of table A2, this adjustment simplifies to a measure similar to the STL-RAM, where the base-period reserve requirements are re placed by the most recent reserve requirements. An adjustment for breaks in applied vault cash is subtracted and an adjustment for breaks in lagged vault cash at bound EDDS and QEDS is added.7 In addition, reserves held against Eu rodollar deposits are excluded from this series after the switch to contemporaneous reserve ac counting.8 As summarized in panel C of table A2, the main difference between these series (not sea sonally adjusted) revolves primarily around the treatment o f vault cash and the method of re serve adjustment.9 m ent change. Nevertheless, the change in reserve require m ents would a ffect the adjusted m onetary base provided that the resulting release of reserves were not perfectly offset by such a defensive measure. H 'he adjustm ent fo r the lag in applied vault cash at bound EDDS is not break-adjusted before the sw itch to contem poraneous reserve accounting. After the switch, the BOG also started to m ake the break-adjustm ent to lagged vault cash at QEDS. 8Because the STL-RAM assum es that the base period reserve requirem ents on E urodollar deposits are zero, the B oard’s exclusion of reserves held against these deposits plays no role in explaining the difference between the two AM B series. 9lt should also be noted that the B oard’s series is adjusted for the annual increase in the reserve tranch as if the change were being phased in gradually over the whole year. In contrast, the St. Louis m easure adjusts for the change when it occurs. See M eulendyke (1990), p. 59. NOVEMBER/DECEMBER 1991 36 Piyu Yue Piyu Yue, a research associate at the 1C2 Institute, University o f Texas as Austin, was a visiting scholar a t the Federal Reserve B ank o f St. Louis when this article was written. Lynn Dietrich provided research assistance. The au tho r wishes to thank Professor Leon Lasdon for the GRG2 Fortran code and Professor Douglas Fisher for providing the data. Estim ations that appear in this article were calculated at the University o f Texas at Austin a n d are the responsibility o f the author. A Microeconometric Approach to Estimating Money Demand: The Asymptotically Ideal Model T X HE DEMAND FOR MONEY plays a critical role in macroeconomics. In conventional money demand analysis, the demand for real money balances is typically expressed as a function of such variables as real income, the expected rate of inflation and the nominal interest rate.1 Em pirical investigations using these variables have not been particularly useful in predicting the demand for money or in formulating and evalu ating monetary policy.2 More recently, a number of researchers have attempted to estimate money demand in a man ner consistent with microeconomic foundations. ’ See Friedm an (1956) for one of the m ost com prehensive discussions of th e m oney dem and function. 2These fu n ction s have been subject to several unexplaina ble shifts and often im ply a larger liq u id ity effe ct than is typ ica lly experienced. Perhaps the most dram atic exam ple of this phenom enon occurred in the early 1980s w ith the yet unexplained break in the incom e velocity of M1. For this and other exam ples, see G oldfeld (1976), Friedm an (1984), Lucas (1988) and Rasche (1990). FEDERAL RESERVE BANK OF ST. LOUIS Even in these cases, however, the empirical results have been largely discouraging.3 This paper reviews the general micro-econometric approach to estimating the demand for money, culminating with an advanced microeconometric model, called the Asymptotically Ideal Model (AIM). AIM is applied to U.S. timeseries data and the results are compared briefly with those from previous studies. AIM results are consistent with microeconomic theory and provide insight into the behavior of money de mand in the 1970s and 1980s. fr e q u e n tly , th e estim ated own price e la sticitie s of dem and for m onetary assets are positive, im plying th a t th e ir de mand curves slope upward. For exam ple, see Serletis (1988), Fisher (1989) and Moore, Porter and Sm all (1990). 37 MICROECONOMIC MODELING As a result of developments in macroeconomic theory over the past two decades, "almost all macroeconomists agree that basing macroeco nomics on firm microeconomic principles should be higher on the research agenda than it has been in the past.”4 Problems arise, however, when aggregate, macroeconomic data are used to estimate microeconomic-based models of money demand. Some of these problems are il lustrated by a simple example that uses two ap proaches to microeconomic modeling: the demand function approach and the utility func tion approach.5 The Dem and Function Approach Consider an economy where the representa tive consumer allocates income between a com posite consumer good, A, and a monetary asset, M, that yields monetary services. The consumer's objective is to maximize the utility function (subject to a budget constraint), given the price of the composite commodity and the user cost of the monetary asset. Let P, u and E denote the price level (the price of A), the nominal user cost of holding one real unit of M and total ex penditures (or income), respectively.6 The con sumer’s decision problem is expressed by cision problem yields ordinary demand functions for A and M. In this case, the demand functions are: (1) A = rE/P = r/(P/E) = G,(u/E, P/E, r), and (2) M = (1 -r)E/u =(l-r)/(u/E) = G2(u/E, P/E, r). The demands for A and M are functions, Gj and G2, respectively, o f E, P, u and the unknown parameter r. Because the budget constraint is linear in P and u, the normalized price, P/E, and user cost, u/E, can replace P, u and E. In gener al, demand functions can be expressed by nor malized prices (including the user cost) and the unknown parameter. This parameter can be es timated by simultaneously fitting equations (1) and (2) using data on real quantities o f A and M and the normalized price and user cost. This approach is called the “demand function” approach because estimation begins after de mand functions are specified. For this approach to yield meaningful estimates, however, the specified system of demand functions must cor respond to the neoclassical utility function from which they were derived. Consequently, the conditions for estimating the system of demand functions are fairly restrictive. For instance, the Rotterdam model (a well-known demand system used in empirical studies) requires specific forms for demand functions and specific constraints on parameters during estimation.8 Max f(A, M) = ArM l r, subject to PA + uM = E.7 For simplicity, the utility function, f, is CobbDouglas, where r, an unknown parameter, characterizes the consumer’s taste or prefer ence. The optimal solution to the consumer’s de 4M ankiw (1990), page 1658. 5Som e econom ists argue that aggregate data cannot be ap plied to m icroeconom ic m odels w ithout considering the problem s of aggregation. A ggregation problem s are not discussed in this paper, although the aggregation error m ight be one source of the unsatisfactory perform ance of conventional m oney-dem and functions. 6The user cost of holding a unit of a real m onetary asset is com puted by the form ula, u = p*(t) [R (t)-i(t)]/[1 + R (t)], w here p *(t) is the “ tru e ” cost of living index defined as the geom etric average of the consum er price index and the consum ption goods deflator, R(t) is the benchm ark interest rate or the m axim um rate in the econom y at each period and i(t) is the interest rate on the m onetary asset. The form ula is derived from a widely applicable consum er decision m odel. Even if these conditions are satisfied, however, the Rotterdam model is still highly restrictive because the assumed underlying utility function (either Cobb-Douglas or CES) is a member of a narrow class o f utility functions with constant elasticities of substitution.9 s u m e r’s u tility function along with real consum ption. In the o th er approach, m oney is viewed as in trinsically worthless; consum ers hold it only to finance current and future con sum ption. As a result, real m oney balances do not enter the consum er’s u tility fu n ction per se. Instead, the liquidity cost o f holding real m oney balances is taken into account in the budget constraint. Feenstra (1986) shows that these tw o approaches are equivalent. 8For an application of the Rotterdam m odel to the moneydem and system, see Fayyad (1986). For the theory of the Rotterdam m odel, see Barnett (1981). 9lt is easy to e n counter d ifficu ltie s using the dem and fu n c tion approach. T he failure to specify fu n ction s correctly or im pose relevant restrictions can result in biased or ineffi cient param eter estim ates. 7D istinct views about m oney have resulted in tw o ap proaches to analyzing consum er dem and for m oney. In the first approach, m oney is viewed as a com m odity which provides a m onetary service flow to holders. Thus, real balances of the m onetary assets directly e nter the con NOVEMBER/DECEMBER 1991 38 The Utility Function Approach The utility function approach to demand esti mation also has been used in empirical studies. To understand this approach, reconsider the consumer's decision problem and the demand functions shown in equations 1 and 2. In the utility function approach, demand functions for A and M are substituted into the utility func tion, f(A, M), to obtain the indirect utility function, h(v1; v2, r) = f[A(v1; v2, r), M (v„ v2, r)], where v1 = P/E, v2 = u/E. Because the indirect utility function has properties that are the in verse of those for the utility function, it is more convenient to use the reciprocal o f the indirect utility function, F(v,, v2, r) = l/h(va, v2, r).10 By definition, demand functions can be ex pressed in terms of their expenditure shares, Sj = AP/E and s2 = Mu/E. That is, A = s jv 1 and M = s2lv2.ri In this way, demand functions can be ob tained without solving first-order conditions. Consequently, no matter how complicated the utility function might be, the derivation o f share equations and demand functions is straight forward. Of course, if the utility function is relatively simple and well-behaved (for example, when the Cobb-Douglas function is used), there is no need to use the utility function approach. However, if the utility function includes more than two goods or is sufficiently complicated, the Lagrange mul tiplier procedure cannot be used to derive de mand functions. THE SEMI-NONPARAMETRIC METHOD FOR ESTIMATING THE DEMAND SYSTEM The critical step in applying the utility func tion approach is the specification of the proper ,0The d u a lity theory states that if the reciprocal of the in direct u tility function, F, is nondecreasing and quasi concave w ith respect to norm alized prices, the respective u tility function, f, m ust be nondecreasing and quasi concave with respect to qu a ntity variables. In this sense, th e u tility fu n ction is equivalent to its reciprocal indirect u tility function. "E x p e n d itu re s shares can be obtained by using the modi- FEDERAL RESERVE BANK OF ST. LOUIS reciprocal indirect utility function, F. To simpli fy the terminology, the term “utility function” will indicate “the reciprocal of the indirect utili ty function” in the following discussion. Flexible Functional Form Modeling Cobb-Douglas and CES functions have been used extensively in theoretical and applied work because of their relative simplicity. Despite their apparent successes, however, such use has been criticized. For example, if there are more than two goods, the CES utility function can only generate demand systems when each pair of goods has the same constant elasticity of substi tution.12 Unless there is prior information to the contrary, however, the elasticities of substitu tion should be determined by the data rather than restricted by the specification of the utility function. This limitation has motivated research ers to look for utility functions that are more flexible and allow for data-determined elastici ties of substitution. Flexible functional form models have attracted considerable attention in economics literature since the early 1970s, when it was proposed that the translog and generalized Leontief func tions should replace neoclassical utility func tions. It was recognized that the values of the elasticities of substitution are determined by the value of the utility function and the values of its first- and second-order derivatives are evaluated at its extreme point. Consequently, if the values of the utility function and these derivatives can be estimated, so too can the elasticities of sub stitution. This idea forms the basis for the flexi ble functional form approach. A functional form is said to be flexible if its level and the first- and second-order derivatives at a point in its domain are allowed to reach the respective values of the "true” utility func tion at that point. The true utility function is as sumed consistent with the properties of the data, so that, in principle, elasticities of substitu tion consistent with the data can be estimated. fied R oy's identity from d u a lity theory. That is, s, = (Vj dF/dv,)/V'VF(V,r), i = 1 ,2 , w here the vector V ' = (vn, v 2) and the gra d ie n t vector V F(V,r) = (dF /dv1t dF/dv2)'. See Diewert (1974). 12See Uzawa (1962). 39 One flexible functional form is derived from a Taylor series expansion where all terms greater than second-order are eliminated, that is, (3) F = a0 + + Z^aijXiXi.13 This approximation is flexible because it has enough free coefficients, a0, ai( a,,, to allow for any desired value of the first- and second-order derivatives of function F. Two frequently used flexible functional forms, the translog and generalized Leontief functions, are given by simple substitution into equation 3. For the translog function, F = ln(f(x)) and x, = ln(q), where f denotes the utility function and q, represents the quantity of good i. The general ized Leontief function is attained by letting F = (fix))* and x, = (q )‘\ The coefficients in these functional forms can be estimated and, in turn, the demand system and the elasticities of substitution can be derived. Caveats F or Flexible Functional Form s Theoretically, the second-order Taylor approx imation can attain flexibility only at a single point or in an infinitesimally small region. Hence, estimates of the elasticities of substitu tion are valid only for the range of observations covered by the sample data. Therefore, the second-order Taylor series approximation should be viewed as "locally flexible." Such models are also subject to another, potentially more serious, problem. Experience has demonstrated that regularity conditions are frequently violated! Therefore, the restrictions that microeconomic theory imposes on con 13T his equation is w ritten in general form , w here a0 denotes all the constants (that is, the function evaluated at the point of interest and the partial derivatives evaluated at the same point). The use of this general form to estim ate these equations is one o f the p rocedure’s lim itations be cause the point about which th e expansion is made is esti mated by the data, rather than being specified by the researcher. Hence, there is no guarantee that this point w ill necessarily correspond with the m axim um value of the function itself. sumer behavior are not embedded in these flex ible functional forms. This point is illustrated later in the empirical section of this paper. In an attempt to solve these problems, micro economists have developed a variety of flexible functional forms that maintain their flexibility and have larger regularity regions.14 A family of such flexible functional form models has been proposed (for example, Barnett's (1981) minflex Laurent model).15 To gain global regularity, however, additional constraints are imposed on the parameters which result in a loss of local flexibility. This tradeoff between flexibility and regularity is characteristic of flexible functional form modeling. None of these models is both globally regular and globally flexible.16 Semi-nonparametric M ethod Gallant (1981) created the "semi-nonparametric method” to remove the local flexibility limita tion. His method specifies a series of models that approximate the underlying utility function at every point in the function’s domain. Hence, the models are globally flexible. The "semi-nonparametric method” is built upon a well-known result in mathematics: a Fourier series expansion can converge to any continu ous function.17 In contrast to the local conver gence of the flexible functional forms, the Fourier series can approximate a continuous function in the entire domain. Gallant proposed to use the Fourier series expansion to specify a series of utility functions that can converge to any neoclassical utility function. Because neoclas sical functions are a subset of continuous func tions, the property of the Fourier series expan sion will guarantee asymptotic convergence to an u n d erly in g neoclassical u tility function. Fourier series modeling consists of a series of expansions of models, with succeeding models nested in the preceding one. When the sample size increases, higher-order models can be speci15lnstead o f th e Taylor series expansion, B arnett used the Laurent expansion to enlarge the regularity region and m aintain enough param etric freedom to satisfy require m ents for fle xib ility. See Barnett (1983). 16See Diewert and W ales (1987). 17T he function m ust be integrable or, more generally, it must lie in a H ilbert space. See Telser and Graves (1972). 14The regularity region is the subset of the dom ain of the utility function in which all regularity co n ditions are satisfied. NOVEMBER/DECEMBER 1991 40 fied by simply adding more terms of the compo nent functions. For instance, the first-order model is defined by the utility function, F = a0 + I^ q , + + ZibiCoslQq,) + ZjdjSinfQq,). The jth-order model is defined by the utility function, F = a0 + S.afl, + 1,1,3,.qq, + X,Xjbiicos(jQqi) + lilidijSinfjQq). Asymptotic3lly, the model contains 3n infinite number of terms 3nd unknown parameters. Therefore, asymptotic inference based upon the Fourier series expansion models is free from functional-form specification error. This is its principal advantage. In empirical work, however, the number of terms must be finite. Consequently, the proper ties of lower-order Fourier models become deci sive. The harmonic component functions, such as sines and cosines which are frequently used in engineering and physics, are not suitable in economic applications because they do not satis fy the usual regularity conditions, such as monotonically increasing and strictly quasi concave. This means that lower-order Fourier series models can violate regularity conditions.18 Nevertheless, Gallant's approach permitted micro-econometric models to achieve both global regularity and global flexibility. The A IM Dem and System To solve the problems of Fourier series models, another infinite function series, called the Muntz-Szatz series, is adopted. A typical form of the series is expressed as: f = a0 + S.a1q|/2 + I,a,2q;A + 1,3? q* +... + 1,1,3,; q‘/;qj4 + I,I,a2q*q'f +... + S.Ijb,} qr qf + Ijljb.f q*q* +... The Muntz-Szstz series expansion converges to 3 continuous function, and any continuous function can be approximated by the Muntz18M oreover, the Fourier series m odels can easily overfit the noise of the data. Usually, the m easurem ent errors of eco nom ic variables can be decom posed into a pure white noise plus some high-frequency periodic functions. These latter functions m ight be m istaken for useful inform ation if th e ir frequencies are close to that of the sine and cosine fu n ction s in the Fourier series m odels. FEDERAL RESERVE BANK OF ST. LOUIS Szatz series.19 Consequently, this series C 3 n be used to approximate 3 neocl3ssic3l utility func tion asymptotically.20 The Muntz-Szstz series is 3 linear combin3tion of 3 set of specisl power functions. In contrast to the Fourier series, the component functions of the Muntz-Sz3tz series, q)2, q|4, ... , q; =qj*, ... , sre neoclassical functions. In other words, they are monotonically incressing and quasi-conc3ve with respect to variables q, and q,. The MuntzSzstz series is necessarily neoclassical, however, only if sll of the coefficients, 3j; 3,, , b,- , ... , sre non-neg3tive, becsuse only positive linear combinstions of the neoclsssicsl component functions sre necesssrily neoclsssicsl. As a result, the coefficient-restricted Muntz-Szatz series can ap proach a neoclassical function but may not approach any continuous function. Imposing these restrictions guarantees that the estimated function will not violate regularity conditions. The Muntz-Szatz series is used in place of the Fourier series in Gallant’s semi-nonparametric method. A series o f models can be defined by increasing degrees of the Muntz-Szatz approxi mations. Under the parameter constraint, these models are globally regular; the respective utili ty functions are neoclassical everywhere in their domain. When the sample size increases, higherorder models can be specified with more free parameters to best fit the data and derive the elasticities o f substitution that the data suggest. Hence, the Muntz-Szatz series gives rise to a model that is asymptotically globally flexible. Even a low-order approximation requires a fairly large number of parameters to be estimated, however. Hence, while the model is asymptoti cally globally flexible, finite samples will limit the researcher’s ability to fully utilize this property. The model has two additional features that make it particularly attractive for applied work. First, although there are a relatively large num ber of free parameters to be estimated, it is im possible to overfit the noise in the data. Because movements due to measurement errors are ir regular and cannot be expressed by the neoclas sical component functions, the model simply 19O nce again, the function m ust be integrable or, more generally, lie in a Hilbert space. See T elser and Graves (1972). 20See Barnett and Jonas (1983). 41 ignores them. Also, because the component functions are not periodic, the high-frequency, periodic movements in the data are likewise ignored. Consequently, models based on the MuntzSzatz series expansion are globally regular and are asymptotically globally flexible. This is why they are called Asymptotically Ideal Models (AIMs).21 ESTIMATION OF THE AIM MONEY DEMAND SYSTEM Four subaggregated goods are included in the empirical work presented here: a consumer good, A4, and three monetary assets, A 1; A2 and A3. A4 is an aggregate good of consumer dura bles, nondurables and services and its respective aggregate price is denoted by p*. A1 consists of currency, demand deposits of households and other checkable deposits; A2 is composed of sav ings deposits at commercial banks and thrifts, super NOW accounts and money market deposit accounts; and A3 is small time deposits at com mercial banks and thrifts.22 For each subset, an aggregate quantity is defined as a sum of per capita real balances of the component monetary assets.23 The opportunity cost of holding a unit of a real monetary asset is measured by its user cost, a quantity-share-weighted sum of the in dividual user costs that compose it. The user cost of A; is denoted by U;. The representative consumer solves an op timal allocation problem by selecting real con sumption, A4, and real balances of the monetary assets, A 1( A2 and A3, to maximize the utility function f(A 1( A2, A3, A4), subject to the budget constraint, given p*, ut, u2, u3 and the total ex penditure, E. Following the utility function ap proach and using the first-order AIM model, we specify the reciprocal indirect utility function for the four goods case as: 21See Barnett and Yue (1988) and Barnett, G eweke and Yue (1991). 22This division of m onetary assets is proposed in Fisher (1989), who has perform ed separability tests over a variety of deposit categories included in M2. He found that this d i vision (A ^ A2, A3, A 4) is w eakly separable in term s of the General Axiom of Revealed Preference (GARP) test. The weak separability condition offers theoretical assurance that A ,, A 2, A 3 and A 4 are m eaningful aggregate goods by the definition of econom ic aggregation theory. However, the partition, [(A,, A 2, A 3), A4] did not pass the GARP test for weak separability. This raises the question of the exis tence o f the m onetary aggregate, M2 (M 2 sA t + A 2 + A 3). If F(v,a) = ajV/2 + a2v2,/2 + a3v3‘/2 + a4v4'/2 + a5v 1'/4v2,/! + a6va,/2v3,/2 + a7v 1'4v4'4 + a8v2'!v3,/! + a9vz,/2v4'/2 + aioV3'"v4'2 + a11v1,/2v2,/2v3,/2 + a12v 1*v2,/iv4,/i + a13v 1,/2v3,/2v4,/2 + a14v2'iv3'!v4,J + alsv a,/!v2'/!v3‘/!v4,/;, where v 1; v2, v3 and v4 are the normalized prices, and a1; a2, ..., and a15 are the parameters of the indirect utility function. The share equation for each good is derived from equations 3 using the modified Roy’s iden tity. These are sa = S^S, s2 = S2/S, s3 = S3/S, and s4 = (1 — s, - s2 - s3), where 51 = a1v 1l/I + a5v1'sv2,i + a6v,'/2v3'/2 + a7v1’iv4Vi + aiav 1,/iv2,-'2v3'/- + aa2v1,/!v2,/!v4,/‘ + a13v 1'!v3''!v4l! + a15v 1'/!v2,/,v3’/,v4,/2, 52 = a2v2'/! + a5va,/2v2’'2 + a8v2'2v3'/2 + a9v2,/2v4'/2 + aiiv,l/2v2,/2v3‘/' + a12v 1,/!v2,/2v4,/2 + a14\2 v3 \4 + air,v, v2 v3 v.,' . 53 = a3v3’/2 + a6v 1'/!v3,/2 + a8v2'2v3'2 + a10v3,2v4"2 + a11v 1,/2v2,/2v3,/2 + a13v 1'/2v3'/2v4l/2 + a14v2,/2v3'2v4'2 + a15v 1'2v2'2v3'!v4'2 and S = a ^ / 2 + a2v2’/! + a3v3’2 + a4v4,/2 + 2a5v ,’/2v2'/2 + 2a6v 1'/2v3'/2 + 2a7v1'2v4,/! + 2a8v2l/2v3l/2 + 2a9v2'4v4'/2 + 2alu\3 v4 + 3a11v1'/2v2'/!v3‘/2 + 3a12v 1'2v2'2v4'2 + 3a13v 1'2v3'2v4l2 + 3a,,\2 v3 v, + 4aj5v j1/2v21/2v31/2v4,/!. Only the first three share equations are in dependent and can be written generally as: (4) Sj = S/S = g;(v,a) for i = 1, 2, 3. When the additional parameter normalization a, + a2 + a3 + a4 = 1 is imposed, one parameter, for example a4, can be eliminated by substitu t e s is the case, the dem and system of consum ption and the subaggregate M2 m onetary assets are the appropriate way to study dem and for m oney in term s of the econom ic aggregation theory. 23The reader is cautioned that these results are not directly com parable with those using conventional m oney dem and specifications because the latter em ploy conventional m onetary aggregate data that d iffe r in several respects from those used here. First, business dem and deposits are excluded from the data used here, so that M1 is not equal to A, and M2 is not equal to A, + A 2 + A 3. M oreover, un like the conventional m onetary aggregates, each of the sub-aggregates is seasonally adjusted separately. NOVEMBER/DECEMBER 1991 42 tion. Hence, in the case of four goods, the firstorder AIM system contains 14 free parameters.24 global optima was reduced, however, by an ex tensive search of the parameter space.26 The share equations are nonlinear with respect to the normalized prices and hence, to income and prices as well. By the definition of expendi ture shares, demand functions can be expressed as Aj = Sj/vj. The complicated nonlinearity of the share equations, however, makes it impossible to derive a closed-form expression for the de mand functions, such as the conventional linear or log-linear functions of income, prices and in terest rates. Fortunately, the estimated parameters and share equations can be used to compute the income and price elasticities for consumer goods and monetary assets. All parameters are subject to non-negativity constraints to guarantee that global regularity conditions are satisfied. Because inequality con straints limit the applicability of the existing the oretical sampling distribution theory, the usual methods for testing hypotheses cannot be used.27 Estimation o f the A IM Dem and System The AIM model is estimated by a maximum likelihood procedure under the assumption that each share equation in (4) has an additive error term, £it. That is, (5) Si = gi(v,a) + £it, i = 1,2,3. The disturbances are assumed to be indepen dent and identically distributed multivariate nor mal random variables with zero mean and covariance matrix X. The sample disturbance covariance matrix, X, is defined as £ = ( £,£’t)/N, Incom e Elasticities and Price Elasticities Because the share equations are so complex, AIM does not yield explicit functional forms for demand functions. This is the consequence for correctly embedding utility maximization into an econometrically estimable demand system that can be used to compute economically meaning ful income and price elasticities. The Allen Partial elasticities of substitution and income elasticities are defined and ex pressed by the following formulas:28 for i # j, _ d Aj Pj £ti = s( - g, (v,a), for i = 1, 2, 3. Maximizing the likelihood function for the sys tem is equivalent to minimizing the generalized variance, |X|.25 The estimation was accomplished using a non linear program (GRG2). To find a global optima, an extensive search over a large range of initial conditions was conducted. Because of the com plex nonlinearity o f the AIM demand system, the true maximum likelihood estimates are difficult to obtain. The possibility of missing the s| ( 1 0PjA- 3 s, — + -) 3E Pi EVjVj for i = j, oH=- 9 A? pj _ dp,A? where N is the sample size, and the sample dis turbance, £„ is computed by 1 3Sj 1 3 S; Sj •Vi 3 Pi v,Pi S; j 1 3sj s^ Ev2 Vi 3E Pi s2. Vi where pj are the prices (and user costs), Af denotes the in com e-com pensated dem and fu n c tions for the ith asset, s, denotes th e expen ditu re shares and E denotes total expenditures. T h e elements, oip constitute a symmetrical matrix called the Allen Partial matrix. The income elasticities are defined by >1i0 = 3 A,E 3 s; E 3E A. 3E sf + 1, and the uncompensated price elasticities are denoted by 24The higher-order AIM system contains m any m ore unknow n param eters; see Barnett and Yue (1988). 27See Dufour (1989), Kodde and Palm (1986), W olak (1989) and Barnett, Geweke and Yue (1991). 25See Barnett (1981). 28See Diewert (1974) and Barnett and Yue (1988). 26To estim ate param eters, a: \ = 1, 2, 3, 5, . . , 15, initial values of the unknow n param eters were assigned. A num ber of diffe re n t initial values of a,, 0.01, 0.03, 0.05, 0.08, 0 . 1, 0.2 ..... 1.0 , 2 .0 ,..., 10.0 were used. FEDERAL RESERVE BANK OF ST. LOUIS 43 _9APi nn = Sj*7iO< a PiA, where A, are the ordinary or uncompensated demand functions. The connection between com pensated and uncompensated demand functions is stated by the usual Slutsky equation. Gross substitutability and complementarity is provided by the off-diagonal terms of the un compensated price elasticity matrix. If rjit is posi tive, good i and good j are substitutes; in other words, when the price of good i rises, demand for good j increases to replace a cutback in de mand for good i. If it is negative, they are com plements—an increase in the price of good i (or j) causes the demand to fall for both goods. Similarly, the pure substitution effects are de fined by the Allen Partial matrix. If the utility function obeys regularity conditions, the own compensated price elasticities (s ^ ) and (aif), must be negative. Hence, the compensated price elasticity matrix represents potential movements along the consumer's indifference curves and can be used to examine whether the estimated underlying utility function satisfies regularity conditions. The computed elasticities of the AIM demand system are compared with other money demand systems in the next section. Because of the com plexity of share equations, a numerical method is used to compute the partial derivatives of the expenditure shares with respect to prices and income that occur in the elasticity formula. The computation of elasticities is calculated using the estimated share equations. Time series of the elasticities are produced by substituting time series of normalized prices and respective par tial derivatives into the elasticity formula. EMPIRICAL RESULTS OF THE AIM MONEY DEMAND SYSTEM In this section, the AIM demand system is esti mated and the income and substitution elasticities are compared with those for the translog and Fourier demand systems. In addition, charac teristics of monetary assets relative to consumer goods are analyzed. 29Douglas Fisher provided all of the data used to estim ate the AIM dem and system. He had previously used these data to estim ate the translog and the Fourier dem and sys tems. Hence, the em pirical results presented here can be com pared d irectly with his. See Fisher (1989). Table 1 Estimates of Coefficients of the AIM: 1970.1 to 1985.2 ai 32 33 .0000 .0000 .0122 34 .9878 35 .2225 a6 .2857 ^7 .2299 as .4466 39 .3568 3l0 .0515 311 .4446 312 .3961 313 .4056 314 .4535 3is .4489 Table 2 Income and Substitution Elasticities Allen Partial Matrix Income A, A, .502 (.005) - 1 6 .5 6 4 (1.880) a2 .512 (.008) 2.693 (.213) -15.450 (2.649) a3 .521 (.003) 5.553 (.721) 5.312 (.609) -31.375 (6.944) a4 1.045 (.007) .364 (.009) .378 (.008) .170 (.009) -.0 3 0 (.005) NOTE: Standard deviations in parentheses. Estimates o f Parameters and In com e and P rice Elasticities Table 1 displays the coefficient estimates from the AIM demand system derived by U.S. quar terly data from 1970.1 through 1985.2.29 These parameters represent the consumer's taste or preference and determine the utility function that underlies the estimated AIM demand sys tem. Because the taste parameters are assumed to be constant, the consumer's utility function and preference did not change over time. The estimates o f at and a2 were zero due to the non-negativity constraint.30 The estimated Allen Partial elasticities of sub stitution and income elasticities are reported in table 2. The numbers represent the averages 30To this extent, the AIM m odel is at odds with the data be cause the estim ated param eters w ould have been negative if unconstrained. NO VEMBER/DECEMBER 1991 44 Table 3 Income and Substitution Elasticities T ra n s lo g In co m e A, a2 _________ F o u rie r a3 A, 0.542 0.106 a2 3.669 - 0 .4 9 0 - 0 .9 6 7 a3 2.741 0.323 2.192 - 0 .2 6 2 A» 0.737 - 0 .1 5 0 0.805 0.428 a4 - 0 .6 7 2 and their standard deviations (in parentheses) over the sample period. Table 3 displays the es timated substitution and income elasticities from the translog and the Fourier series models previously reported by Fisher (1989, page 103). Table 4 presents the average uncompensated price elasticities and their standard deviations over the sample period for AIM. The cor responding elasticities for the other two models are not available. What’s W rong with the Translog and Fourier Dem and Systems? In co m e A, 0.879 -3 .8 5 1 2.714 1.258 - 8 .1 0 4 2.803 1.119 2.876 - 1 .5 9 0 0.898 0.171 0.412 - 0 .2 2 4 Problems in the Fourier series demand system cannot be seen in table 3 because the numbers reported there are the average values of these coefficients. According to Fisher, however, ex cept for >7 40, the income elasticities and Allen Partial elasticities of substitution changed signs frequently over the period.31 For example, in 1970, o12 was significantly negative, implying complementarity; in 1971-1972, it was signifi cantly positive, implying substitutability; and then in 1974-1975, it became negative again. Figure 1 displays a o12-comparison of the Fourier and AIM money demand systems. It is inexplica31See Fisher (1989), pp. 105-06. FEDERAL RESERVE BANK OF ST. LOUIS a3 A4 - 0 .0 4 2 Table 4 Uncompensated Price Elasticities A, A2 a3 a4 A, -.5 3 4 (.004) .075 (.010 ) .084 (. 011 ) - .1 2 6 (.009) a2 .069 (.005) - .5 3 7 (.007) .080 (. 010 ) -.1 2 3 (.007) a3 .157 (.005) .162 (.009) - .5 1 8 (.004) -.3 2 2 (.004) - .0 2 3 (.005) - .0 1 5 (.003) - .9 8 5 (.003) A„ In the translog demand system results (shown in table 3), the positive sign of indicates that the regularity condition is violated. This result suggests that the higher the opportunity cost of holding currency and demand deposits, the greater their demand. Given this violation of the "law o f demand,” the results from the translog demand system must be considered suspicious at best and, at worst, unreliable. a2 -.0 2 2 (.003) NOTE: Standard deviations in parentheses. ble that currency and demand deposits, A 1( and savings deposits and money market deposit ac counts, A2, should be complements during some periods and substitutes during others. Empirical Inference o f Characteristics o f M onetary Assets b y the A IM Dem and System The anomalies observed with the translog and the Fourier series demand systems do not occur in the AIM demand system. The own-price elasticities are negative and all estimated elastici ties maintain their signs over the entire sample period. Moreover, their smaller standard devia tions indicate that they are more stable; this can also be seen in figure 1. In the Allen Partial matrix, the diagonal elements are all negative 45 F ig u re 1 T h e A lle n P artial E la s tic ity o f A 1 a n d A 2 while the off-diagonal elements are positive. This implies that the three monetary aggregates and aggregate consumption are substitutes for each other in the presence of income compensa tion. Moreover, the pure substitution effect be tween each pair o f the three aggregated monetary assets is much greater than between the consumption good and monetary assets. The income elasticities in table 2 are all posi tive, with the income elasticity of the consump tion good about unity, and the income elasticities of the three monetary assets roughly equal to 0.5.32 These results suggest that con sumption goods and monetary assets are normal goods. Table 4 shows that the uncompensated crossprice elasticities of (A1( A2), (A1; A3) and (A2, A3) are positive, implying that these monetary assets are gross substitutes. The uncompensated cross price elasticities of (A1( A4), (A2, A4) and (A3, A4) are negative, indicating that consumption goods and monetary assets are gross complements. These results show that, if the user costs of savings deposits (or money market deposit ac counts, or small time deposits) rise, consumers will shift their funds to demand deposits (or to 32T his may not be ju stifie d by the unity incom e e lasticity in the reduced form of the aggregate m oney-dem and function. As pointed out in the text, no reduced form dem and fu n ction s are estim ated in our study and the incom e e lasticity is defined by the m icroeconom ic approach. Because their incom e elasticities are sign ifica n tly less than one, the m onetary assets in our study are not “ luxury goods,” as has been claim ed in some previous research. See Serletis (1988). checkable deposits or currency). An opposite shift of funds will take place if the user costs of currency, demand deposits and checkable deposits should rise. If these user-cost changes are sufficiently large, ignoring the cross-price effects among monetary assets will produce large errors in their demand functions. Monetary services and consumer goods are consumed jointly. If a consumer increases the consumption of commodities for some reason, the demand for monetary assets is also in creased. This is consistent with the idea that consumers hold monetary assets to finance cur rent and future consumption. Also, the negative rjj4 indicates that price inflation will reduce de mand for monetary assets. Not surprisingly, the own price elasticities of monetary assets are greater than their cross price elasticities. Similarly, it is not surprising to see that the cross-price effects o f a change in the price of consumption goods on the demand for monetary assets are greater than the crossprice effects of a change in the price of mone tary assets on consumption. Consequently, em pirical results from the AIM demand system provide a reasonable quantitative analysis of the characteristics of monetary assets—characteris tics that are broadly consistent with conventional views o f demand for money. A DYNAMIC ANALYSIS BY THE AIM DEMAND SYSTEM Constructing a dynamic model of demand for money has been difficult because the current state of economic knowledge about dynamic be havior is incomplete; the dynamics of money de mand are still very much a "black box” mystery.33 Unlike most multivariate time series models, the AIM model is static. It does not consider specific dynamic effects among monetary assets and consumption goods. The utility function is not intertemporal and its parameters are timeinvariant—the consumer's preference is not per mitted to change over time. 33This situation occurs in a recent debate in econom ic literature; see Hendry and Ericsson (1991) and Friedm an and Schwartz (1991). NOVEMBER/DECEMBER 1991 46 Figure 3 Uncompensated Own Price Elasticities F igure 2 Incom e Elasticities E ,, E2, E3 and E4 for A ,, A2, A 3 and A 4 - 0.54 - 0.55 1970 71 72 73 74 75 76 77 78 79 80 81 82 83 84 1985 Nevertheless, a simple dynamic analysis can be used to examine the AIM demand system. Time series of income and price elasticities can be computed using estimated share equations. Movements in these elasticities reflect both changes in user costs and the consumer's reac tions to such changes. Both of these are reflected in the shares, S;. In this way, the dynamics of the AIM demand system can be investigated even though demand for money is stable by assumption. This dynamic analysis is displayed in figures 2, 3 and 4. Figure 2 shows that the income elasticities of the three monetary assets are rela tively constant over the entire sample period. In contrast, the price elasticities (shown in figures 3 and 4), exhibit sizable fluctuations. Major shifts in price elasticities occurred during 1973.1-1974.4 and 1978.2-1982.2. During these periods, a num ber of studies have reported that demand for monetary aggregate M l was "erratic.”34 Price and user cost elasticities moved drasti cally during these periods. In the second period (1978.2-1982.2), the cross price elasticity, rj12, rose by 50 percent of its 1977 level, implying that demand for Aa (currency plus demand de posits, plus checkable deposits) became much more sensitive than it was previously to changes in the opportunity cost of holding A2 (savings deposits and money market demand accounts). Meanwhile, there was a sharp rise in the user 34See Goldfeld (1976) and Friedman (1984). FEDERAL RESERVE BANK OF ST. LOUIS Figure 4 U ncom pensated Cross Price Elasticities cost of A2. These factors would appear to ac count for the major shift of funds from A2 to A, during the period. The opposite price elasticity, r]21, also rose by 20 percent. Nevertheless, it was less than 80 percent of the value of rj12 and the rise in the user cost of A1 was more modest than that of A2. It was observed that the opportunity cost of A2 increased much faster than that of A,. Hence, the actual flow of funds from A, to A2 might not be significant. The cross price elasticity, r)13, dropped 30 per cent in 1979, implying that the demand for A, 47 was less sensitive to changes in the user cost of A3. Hence, the shift of funds from A3 to Aj should have been moderate despite a substantial increase in the user cost of A 3. Figure 5 G row th Rates o f A g greg ate A 1 A ctual vs. Sim ulation These results are roughly consistent with de velopments during the period. In November 1978, commercial banks w ere authorized to offer au tomatic transfer service (ATS) from savings ac counts to checking accounts. Other interest ceiling-free accounts were also introduced in the early 1980s. In January 1981, NOW ac counts were introduced nationwide. These finan cial innovations should have encouraged consumers to shift funds from savings accounts and money market deposit accounts in A2 into NOW accounts in A x. This may have increased the interest sensitivity of demand for A,.35 Simulating the Growth Rates o f Monetary Aggregates F ig u re 6 G ro w th R a te s o f A g g re g a te A 2 A c tu a l v s . S im u la tio n A further investigation of the behavior of mon etary aggregates can be made by a dynamic simulation. For example, suppose that demand for A; has been derived by utility maximization and expressed by the ordinary demand func tions of price and user costs and total expen diture (6) Aj = G, (uu u,, u3, u4, E). The total differentiation of (6) results in (7) dAj = X dGj/du, du, +dG/dE dE. i=i Dividing both sides of (7) by (6) and using defi nitions of the uncompensated price elasticities and the income elasticity gives (8) dAj/Aj = I Figure 7 G row th Rates o f A g greg ate A 3 Actual vs. Sim ulation Kjij (du/u;) + r),„ (dE/E). i- i Using time series of the elasticities and the growth rates o f price and user costs and total expenditure, the right-hand sides of the equa tions in (8) are computed. In this way, the growth rates of demand for A, can be "simulated." The actual and simulated growth rates of de mand for monetary aggregates and consumption are displayed in figures 5 through 8. The simu lations match the actual growth rates fairly well, especially for consumption. The simulation rates of monetary assets had large fluctuations 35See Thornton and Stone (1991) for a discussion of this possibility. NOVEMBER/DECEMBER 1991 48 Figure 8 Growth Rates of Aggregate A4 Actual vs. Simulation around the actual growth rates in the periods 1972.4-1975.1 and 1978.2-1982.2. Fluctuations in interest rates and inflation rates were substan tial during each period, causing corresponding fluctuations in growth rates of user costs.36 These changes are reflected directly in the sim ulation rates. Because the AIM model is static, sharp changes in user costs are necessarily reflected in cor respondingly sharp changes in the simulated growth rates of aggregates. Hence, it is not sur prising that the simulation errors are large dur ing periods when there are sharp changes in user costs. Nevertheless, figures 5 through 8 suggest that the AIM demand system has cap tured many of the characteristics of the U.S. m on etary system du rin g the sample period. Can the A IM Dem and System Ex plain the Case o f the Missing M oney? Although the simulation o f the growth rate of A! indicates that the AIM model produced rela tively large errors during the period of "missing money” (1973.4-1976.2), an analysis of the AIM results might provide a clue.37 During this peri od, there was a sharp decline in demand for 36Some term s are essentially zero and can be ignored. The follow ing grow th rate equations are accurate enough to produce the sim ulation: dA-|/Ai = f7ndui/ui + i7i2du2/u2 + f]i3du3/u3 + ^44du4/u4 + rj10dE/E dA 2/A 2 = rj22du2/u 2 + 7723d u 3/u 3 dA 3/A 3 = J733du3/u3 = r)42d u 2/u 2 + ^ 43du 3/u 3 + K)44du 4/u 4 + rjiodE/E. In the equation fo r A, th e re are m ore affecting elements; the own and cross-price effects of A2, A 3 and A 4 are im FEDERAL RESERVE BANK OF ST. LOUIS M l; conventional money demand equations con sistently overpredicted demand. A potential explanation can be obtained by considering figures 9 and 10. Figure 9 shows the partial derivative of demand for A1 with respect to its own user cost. Figure 10 shows the partial derivative of demand for A, with respect to the user costs of A2, A 3, A4 and E. Figure 9 shows a sharp rise in the rate of change in demand for A, with respect to a change in its user cost. Indeed, figure 11 shows that the user cost o f Aj increased relative to the price level. Hence, the demand for A 1 should have declined by a proportionately larger amount than the rise in its user cost. Conventional linear money demand equations with fixed regression coefficients could not accommodate this nonlinearity because, in such linear demand systems, the coefficients (derivatives) are assumed to be constant. However, figures 9 and 10 clear ly suggest that this is not the case. Moreover, ordinary least squares is relatively sensitive to "outliers.” Consequently, there will be substan tial changes in the estimated regression coeffi cients when the equations are estimated over periods when these derivatives change signifi cantly. Conventional money demand equations may be misspecified for another reason, as well. Usually they include a single short-term interest rate intended to reflect the opportunity cost of holding money. AIM analysis indicates that the demand for money does not depend on a single “representative” interest rate, but on its user cost and the user costs of “close” substitutes (recall that the demand for A1 was sensitive to changes in the user costs of A2). Hence, conven tional money demand equations may produce misleading results when interest rates change relative to the user costs of M l or relative to the user costs of close substitutes for M l. Therefore, the case of “missing money” and “unexplained” parameter shifts in conventional money demand functions may result from the fact that they are essentially linear approximaportant in sim ulating the growth rate of A ,. T he growth rates of dem and for the oth er tw o m onetary aggregates, however, are determ ined m ainly by th e ir own price effects and cross-price effect, rj23. T his suggests that ignoring the substitution effects of non-M1 com ponents of M2 m ight be one of the factors that d iscre d it reliability of the conven tional M1 dem and function. 37See G oldfeld (1976). 49 Figure 9 The First Derivative of A, w ith Respect to Own User Cost 1970 71 72 73 74 75 76 77 78 79 80 81 82 83 84 1985 F igu re 10 T h e First D e riv a tiv e s o f A t w ith R e s p e c t to U s e r C osts, Price and In c o m e d a 13 — ' * \ ^ d a 14 - 10 - 20 -3 0 S 'N s - n d a e 1970 ..... ................................ .- :: .... 71 72 73 74 75 76 77 78 79 80 81 82 83 F igure 11 U ser Cost o f A, vs. Price of A4 84 1985 tions to nonlinear demand functions. If so, they intuitively will provide much poorer approxima tions during periods when there are dramatic changes in user costs. Is The M on ey Demand Function Stable? ble.”38 Others have asserted that money demand is stable based on the observed stability of the consumption function.39 The AIM demand system integrates demand for both consumption and money and then esti mates them simultaneously. These estimates sug gest that, while the own price and cross price elasticities show considerable variation due to changes in the price level and user costs, they change little on average over the period (see figures 3 and 4).40 Moreover, the estimated in come elasticities for all three monetary aggre gates are nearly constant (see figure 2). Of course, these results are obtained from a model where the estimated parameters are time-invariant, that is, the preference function is constant. Because of this, it is necessarily true that de mand functions are "stable." Nevertheless, the relatively good performance of AIM provides some promise that, like consumption, the de mand for money will ultimately be shown to be a stable function of a relatively few economic variables—in this case, income and user costs. CONCLUSION The erratic behavior of conventional money demand functions and, more recently, the in come velocity of M l, have led many researchers to assert that the demand for money is “unsta Two distinctly different micro-econometric de mand system approaches to the demand for money were presented and discussed. An ad vanced AIM demand system was presented and estimated using U.S. time-series data. Unlike 38For a discussion of the velocity of M1 and an analysis of som e of the explanations, see Stone and Thornton (1987). ing. However, no form al tests of stationarity were per form ed in this study. 39For example, see Friedm an (1956) and Lucas (1988). 40ln the parlance of modern tim e-series analysis, these elasticities are said to be stationary, that is, mean revert NOVEMBER/DECEMBER 1991 50 other utility function-based approaches, AIM es timates are consistent with microeconomic the ory. Dynamic simulations of the growth rates of various monetary aggregates and consumption suggest that the estimated AIM model performed well; nevertheless, the largest simulation errors occurred in periods when there were relatively sharp swings in user costs or inflation. This is perhaps not too surprising given the static na ture of the AIM analysis. An analysis of changes in income and cross price elasticities are suggestive of portfolio shifts among monetary aggregates in the 1970s and 1980s consistent with the observed behavior of these aggregates. The results of AIM suggest that the reported failure of conventional linear (or log-linear) money demand equations may result from trying to fit fundamentally non linear functions with linear ones. The results shown here suggest that this problem will be particularly acute whenever there are sharp changes in user costs. Unfortunately, these are precisely the times when AIM performance was also poor. The key to solving this problem in AIM, however, is to find a way to make AIM ex plicitly dynamic. It may not be necessary to assume that consumer preferences are unstable. The sampling distribution theory for AIM has not been worked out at this time, so relevant hypothesis tests cannot be conducted yet. Also, because the time series on the relevant user costs o f monetary aggregates is limited, the available data cover a relatively short sample period. These factors, coupled with the fact that even low-order (first-order) AIM systems require a relatively large number of estimated parame ters, place severe limits on attempts to evaluate the performance o f AIM using out-of-sample forecasts. Despite these problems, the estimated AIM system appears to have captured many of the characteristics of monetary assets and offers some useful explanations to puzzling empirical issues. Hence, these results are encouraging to those who believe that microeconomic princi ples, such as utility maximization, can be ap plied usefully to macroeconomic problems. REFERENCES Barnett, W illiam A. Consumer D em and a nd Labor Supply (Am sterdam : N orth-Holland, 1981). ________ . “ New Indices of M oney S upply and the Flexible Laurent Dem and System,” Journal of Business and Eco nom ics Statistics (January 1983), pp. 7-23. FEDERAL RESERVE BANK OF ST. LOUIS Barnett, W illiam A., and Andrew B. Jonas. “ The MuntzSzatz Dem and System: An A pplication of a G lobally Well Behaved Series Expansion,” Econom ics Letters Vol. 11, No.4 (1983), pp. 337-42. Barnett, W illiam A., John Geweke, and Piyu Yue. “ Sem inonparam etric Bayesian Estim ation of the Asym ptotically Ideal Model: the AIM C onsum er Dem and System,” in W illiam Barnett, Jam es Powell, and G eorge Tauchen, eds., Nonparam etric a n d Sem inonparam etric M ethods in Economet rics and Statistics, Proceedings of th e Fifth International Sym posium in Econom ic T heory and Econom etrics (Cam bridge University Press, 1991). Barnett, W illiam A., and Piyu Yue. “ S em inonparam etric Esti m ation of the Asym ptotically Ideal M odel: th e AIM Demand System,” in G eorge Rhodes and Thom as Fomby, eds., N onparam etric and R obust Inference, A dvances in Econo metrics, Volum e 7 (1988), pp. 229-51. Diewert, W. E. “A pplications of D uality Theory,” in M ichael D. Intriligator and David A. Kendrick, eds., Frontiers o f Quantitative Economics, Vol. II (Am sterdam : North-Holland, 1974), pp. 106-71. Diewert, W. E., and T. J. Wales. “ Flexible Functional Forms and Global Curvature Conditions,” Econom etrica (January 1987), pp. 43-68. Dufour, Jean-M arie. “ Nonlinear Hypotheses, Inequality Res trictions, and Non-Nested Hypotheses: Exact Sim ultaneous Tests in Linear Regressions,” Econom etrica (March 1989) pp. 335-56. Fayyad, Salam K. “A M icroeconom ic System-W ide Approach to the Estim ation of the D em and for Money,” this Review (August/Septem ber 1986), pp. 22-33. Feenstra, Robert C. “ Functional Equivalence Between Li q u idity Costs and the U tility of Money,” Journal o f M onetary Econom ics (M arch 1986), pp. 271-91. Fisher, Douglas. M oney D em and a n d M onetary Policy (The University of M ichigan Press, 1989). Friedm an, Benjam in. “ Lessons From the 1979-82 M onetary Policy Experim ent,” Am erican Econom ic Review (May 1984), pp. 382-87. Friedm an, Milton. "T h e Q uantity T heory of M oney—A Restate ment,” in M ilton Friedm an, ed., Studies in the Q uantity The ory o f M oney (University of Chicago Press, 1956). Friedm an, M ilton, and Anna J. Schwartz. “ Alternative A p proaches to Analyzing E conom ic Data,” The Am erican Eco nom ic Review (March 1991), pp. 39-49. G allant, A. Ronald. “ On the Bias in Flexible Functional Forms and an Essentially U nbiased Form: The Fourier Flexible Form,” Journal o f Econom etrics (February 1981), pp. 211-45. ________ . “ The Fourier Flexible Form,” Am erican Journal of Agricultural Econom ics (May 1984), pp. 204-08. Gallant, A. Ronald, and Douglas W. Nychka. “ SemiN onparam etric M axim um Likelihood Estim ation,” Econom etrica (March 1987), pp. 363-90. G oldfeld, Stephen M. “ The Case of the M issing Money,” Brookings Papers on Econom ic A ctivity (3:1976), pp. 683-739. Hendry, David F., and Neil R. Ericsson. “An Econom etric Analysis of U.K. M oney D em and in M onetary Trends in the United States and the United Kingdom by M ilton Friedm an and A nna J. Schwartz,” Am erican Econom ic Review (March 1991), pp. 8-38. Judd, John P., and John L. S cadding. “ The Search for a Sta ble M oney D em and Function: A Survey of th e Post-1973 Literature,” Journal o f Econom ic Literature (Septem ber 1982), pp. 993-1023. 51 Kodde, David A., and Franz C. Palm. “ Wald C riteria for Joint ly Testing Equality and Inequality R estrictions,” Econom etrica (Septem ber 1986), pp. 1243-48. Serletis, Apostolos. “ Translog Flexible Functional Forms and S ubstitutability of M onetary Assets,” Journal o f Business a nd Econom ic Statistics (January 1988), pp. 59-67. Lasdon, Leon S., and A. D. W arren. "G R G 2 U ser’s G uide” University of Texas at Austin, 1989. Stone, Courtenay C., and Daniel L. Thornton. “ Solving the 1980s Velocity Puzzle: A Progress Report,” this Review (August/S eptem ber 1987), pp. 5-23. Lucas, R obert E., Jr. “ M oney Dem and in the United States: A Q uantitative Review,” Carnegie-Rochester Conference Series on Public Policy No. 29, (Autum n 1988), pp. 137-68. Mankiw, N. Gregory. “A Q u ick Refresher Course in M acro econom ics,” Journal o f Econom ic Literature (Decem ber 1990), pp. 1645-60. Moore, George R., Richard D. Porter, and David H. Sm all. “ M odeling the Disaggregated Dem ands for M2 and M1: The U.S. Experience in the 1980s,” in Peter Hooper and others, eds., Financial Sectors in Open Econom ies: Em piri c a l Analysis a nd Policy Issues (Board of G overnors of the Federal Reserve System, 1990), pp. 21-97. Pierce, Jam es L. “ Did Financial Innovation Hurt the Great M onetarist E xperim ent?” Am erican Econom ic Review (May 1984), pp. 392-96. Rasche, Robert H. “ Dem and Functions for M easures of U.S. Money and Debt,” in Peter Hooper and others, eds., Finan cia l Sectors in Open Econom ies: Em pirical Analysis and Policy Issues (Board of G overnors of the Federal Reserve System, 1990), pp. 113-61. Swofford, Jam es L., and G erald A. W hitney. “ Flexible Func tional Forms and the U tility A pproach to the Dem and for Money: a N onparam etric Analysis,” Journal o f Money, Credit and Banking (August 1986), pp. 383-89. Telser, Lester G., and Robert L. Graves. Functional Analysis in M athem atical Econom ics (The U niversity of Chicago Press, 1972). T hornton, Daniel L., and C ourtenay C. Stone. “ Financial Inno vation: Causes and Consequences,” in Kevin Dowd and Mervyn K. Lewis, eds., Current Issues in M onetary Analysis a nd Policy (M acM illan Publishers, 1991). Uzawa, H irofum i. “ Production Functions with Constant Elasticities of Substitution,” Review o f Econom ic Studies (October 1962), pp. 291-99. W olak, Frank A. “ Local and G lobal Testing of Linear and Nonlinear Inequality C onstraints in N onlinear Econom etric Models,” Econom etric Theory (April 1989), pp. 1-35. NOVEMBER/DECEMBER 1991 52 Mark D. Flood M ark D. Flood is an econom ist at the Federal Reserve Bank of St. Louis. David H. Kelly provided research assistance. Microstructure Theory and the Foreign Exchange Market A GROWING BODY OF theoretical literature, known as the study of securities market microstructure, deals with the behavior of participants in securities markets and with the effects of in formation and institutional rules on the economic performance of those markets. These institu tional factors may arise from technology, tradi tion or regulation. Microstructure and its impact are important, because of the vast amounts of wealth which pass through securities markets — including the foreign exchange market — every day. Microstructure is of interest to students of the foreign exchange market: microstructural analy ses of other markets have yielded insight into traders’ behavior and the effect of various insti tutional arrangements. Conversely, the foreign 'S im ila r arrangem ents exist for other se cu ritie s— for exam ple, the federal funds m arket and the secondary m arket fo r T reasury securities— but these too have been relatively neglected in the literature. 2The shaded insert on the opposite page provides a context in which the m icrostructural approach can be com pared with m ore traditional approaches to m arket efficiency. Follow ing some early articles by Dem setz (1968), Tinic (1972) and T inic and W est (1972), G arm an (1976) per form ed the crucial task of defining m arket m icrostructure as an independent area o f th e literature, th u s focusing the debate. Since then, m arket m icrostructure has burgeoned, led by Cohen, Maier, Schw artz and W hitcom b (1978a, 1978b, 1981, 1983), Am ihud and M endelson (1980, 1986, 1988), Stoll (1978, 1985, 1989) and Ho and Stoll (1980, 1981). See also Beja and Hakansson (1977), Cohen, Hawawini, Maier, Schw artz and W hitcom b (1980), Cohen, Maier, Ness, O kuda, Schwartz and W hitcom b (1977), A m i hud, Ho, and Schwartz (1985), Schreiber and Schwartz (1986), Schwartz (1988) and Cohen and Schw artz (1989). FEDERAL RESERVE BANK OF ST. LOUIS exchange market is also of special interest to students o f microstructure, because it combines two very different arrangements for matching buyers and sellers — bank dealers trade with one another both directly and through foreign exchange brokers.1 Standard models of exchange-rate determina tion concentrate on relatively long-run aspects, such as purchasing power parity. While micro structure theory cannot address these issues directly, it can illuminate a more narrowly fo cused array of institutional concerns, such as price information, the matching of buyers and sellers, and optimal dealer pricing policies. De spite the substantial literature on microstructure, little attention has been paid to the particular microstructure of the foreign exchange market.2 Cohen, Maier, Schw artz and W hitcom b (1979, 1986) and Stoll (1985) have surveyed th e m icrostructure literature. In addition to the early note by Allen (1977), very recently there have appeared some m icrostructural studies of the foreign exchange m arket: Bossaerts and Hillion (1991), Lyons (1991), Rai (1991) and Flood (1991). There is also an em pirical literature m easuring the determ inants of the bid-ask spread in the foreign exchange m arket. See Black (1989), W ei (1991) and Glassm an (1987) as well as the references therein. Because th e focus of this article is on m icrostructure theory, such em pirical studies receive little attention here. Finally, although a consideration of the results of laborato ry experim ents would expand th e scope of th is paper to unw ieldy dim ensions, th e ir role in establishing the sensitiv ity of m arket behavior to in stitutional fa cto rs m ust at least be acknow ledged; see Plott (1982, 1991) fo r an in troduction. 53 Price Efficiency in a Heterogeneous Marketplace Implicit in most microstructural models is a presumption that participants in any given market are heterogeneous, that is, that they differ in certain key determinants of econom ic behavior: information, beliefs, preferences and wealth. Although this assumption con sumes little attention in the microstructure literature — it is taken for granted — it is valuable to discuss it in the more familiar theoretical context of market efficiency. The standard definition o f price efficiency is: fm(p,|lmt_i) = f *(pt1 1 ,_ In other words, the joint distribution over future prices, fm(pt), as sessed by the monolithic market (or a repre sentative agent in that market) and made conditional on the current information, Iml_„ available to the market is equal to the “true" joint distribution, f*(p t), made conditional on all current information, I,_,. Roughly speak ing, the market sorts things out as accurately as possible.1 This approach breaks down in a microstructural analysis. First, the simplifying as sumption of homogeneous participants is abandoned. Although it is widely recognized “that investors do not show the homogeneity of beliefs which characterize our theories," the benefits o f realism (i.e., the heterogeneity assumption) are often outweighed by other criteria (e.g., testability, tractability, etc.).2 Em phasizing testability, Ross offers a standard rejoinder, namely that “since a single ex post distribution of returns is observed by all, over time one would not expect to observe systematic and persistent differences.” This is a rational expectations argument, which de pends crucially on the stationarity o f the returns distribution and which ignores the ef fect o f differences in opinions and beliefs, which go beyond differences in information. 1See Fama (1976), ch a p te r 5, for the de fin itive presentation. In general then, at the level of detail involved in microstructural studies, the homogeneity assumption is not an excusable flaw; in a homogeneous market why — let alone how — would anyone trade? More fundamentally, the notion of a “true” price must be questioned. In the context of the literature on price efficiency, the intro duction o f a “true” distribution as a theoreti cal conceit leads to joint testing problems, as the “true” distribution is ipso facto unobserva ble. More fundamentally, positing a “true” distribution confuses the chain of causality; it presumes that future prices are drawn from some exogenous probability distribution and that investor behavior is concerned with ac curately estimating that distribution. In fact, investor behavior in the market place determines the distribution o f future prices, not the other way around. This fact in no way depends on the ultimate basis or mo tivation for investor behavior. In an explicit model of price discovery, the assertion o f an ex ante exogenous equilibrium price is mean ingless. As Schreiber and Schwartz put it, “the fact that security analysts assess the value o f a stock for their own portfolios does not imply that they undertake a treasure hunt to find some golden number which one might call an intrinsic value.”3 In sum, the standard theory of efficient markets is illsuited to the modeling of price discovery. In comparing observed prices to an imputed “true” distribution, studies o f market efficien cy ignore more immediate concerns — for ex ample, how well the institutional structure transmits information, whether arbitrage op portunities occur, and how well the market allocates assets among investors. These con cerns are the focus o f microstructural analysis. 3See S chreiber and S chw artz (1985), p. 22. 2See Ross (1978), pp. 889-90. See Varian (1989) for a more thorough review of the theoretical issues involved in the heterogeneity assum ption. NOVEMBER/DECEMBER 1991 54 This paper examines the extant literature on market microstructure to determine how it might be applied to the foreign exchange market. The paper begins with a brief description of the foreign exchange market. Aspects of the literature concerned with institutional details are addressed second, noting how such details can affect the performance of the market. Next, the literature dealing with behavioral details, es pecially the communication and interpretation of price information, is considered. Finally, the interaction of institutional and behavioral fac tors, notably the bid-ask spread, is discussed. Figure 1 Spot Market Volume by Transactor (4/89) INSTITUTIONAL BASICS OF THE FOREIGN EXCHANGE MARKET The foreign exchange market is the interna tional market in which buyers and sellers of currencies "meet.”3 It is largely decentralized: the participants (classified as market-makers, brokers and customers) are physically separated from one another; they communicate via tele phone, telex and computer network. Trading volume is large, estimated at §128.9 billion for the U.S. market in April 1989. Most of this trad ing was between bank market-makers.4 The market is dominated by the market-makers at commercial and investment banks, who trade currencies with each other both directly and through foreign exchange brokers (see figure l).5 Market-makers, as the name suggests, "make a market” in one or more currencies by providing bid and ask prices upon demand. A broker ar ranges trades by keeping a "book” of marketmaker’s limit orders — that is, orders to buy (al ternatively, to sell) a specified quantity of for eign currency at a specified price — from which he quotes the best bid and ask orders upon re quest. The best bid and ask quotes on a broker’s book are together called the broker's "inside spread.” The other participants in the market are the customers of the market-making banks, w ho generally use the market to com plete transactions in international trade, and central banks, who may enter the market to move ex3For m ore thorough descriptions of the w orkings of the for eign exchange m arket, see Burnham (1991), C hrystal (1984), Kubarych (1983) and Riehl and Rodriguez (1983). 4See Federal Reserve Bank of New York (1989a) and Bank fo r International S ettlem ents (BIS) (1990). E xtending this fig u re over 251 trad in g days per year, this im plies a trad ing volum e of roughly $32 trillio n for all of 1989. Volum e FEDERAL RESERVE BANK OF ST. LOUIS - Interbank Direct (55.0%) - Interbank Brokered (39.9%) change rates or simply to complete their own international transactions. Market-makers may trade for their own account — that is, they may maintain a long or short position in a foreign currency — and require significant capitalization for that purpose. Brokers do not contact cus tomers and do not deal on their own account; instead, they profit by charging a fee for the service o f bringing market-makers together. The mechanics of trading differ substantially between brokered transactions and direct deals. In the direct market, banks contact each other. The bank receiving a call acts as a market-maker for the currency in question, providing a twoway quote (bid and ask) for the bank placing the call. A direct deal might go as follows: M o n go b a n k : '‘Mongobank with a dollar-mark please?” (Mongobank requests a spot market quote for U.S. dollars (USD) against German marks (DEM).) has roughly doubled every three years fo r the past decade. 5Federal Reserve Bank of New York (1989a) lists 162 m arket-m aking institutions (148 are com m ercial banks) and 14 brokers; an earlier study, Federal Reserve Bank of New Y ork (1980), lists 90 m arket-m aking banks and 11 brokers. 55 Loans 'n Things: “20-30” (Loans n’ Things will buy dollars at 2.1020 DEM/USD and sell dollars at 2.1030 DEM/USD —the 2.10 part of the quote is understood.) M ongobank: “Two mine.” (Mongobank buys $2,000,000 for DEM 4,206,000 at 2.1030 DEM/USD, for payment two business days later. The quantity traded is usually one of a handful of "customary amounts.”) Loans ’n Things: “My marks to Loans 'n Things Frankfurt." (Loans n' Things requests that payment of marks be made to their account at their Frankfurt branch. Payment will likely be made via SWIFT.)6 M ongobank: “My dollars to Mongobank New York. ” (Mongobank requests that payment of dol lars be made to them in New York. Payment will most likely be made via CHIPS.)7 Spot transactions are made for “value date” (payment date) two business days later to allow settlement arrangements to be made with cor respondents or branches in other time zones. This period is extended when a holiday inter venes in one of the countries involved. Payment occurs in a currency's home country. The other method o f interbank trading is brokered transactions. Brokers collect limit orders from bank market-makers. A limit order is an offer to buy (alternatively to sell) a speci fied quantity at a specified price. Limit orders remain with the broker until withdrawn by the market-maker. The advantages of brokered trading include the rapid dissemination o f orders to other market-makers, anonymity in quoting, and the freedom not to quote to other market-makers on a reciprocal basis, which can be required in the direct market. Anonymity allows the quoting bank to conceal its identity and thus its inten tions; it also requires that the broker know who is an acceptable counterparty for whom. Limit 6The Society for W orldw ide Interbank Financial Telecom m u nication (SWIFT) is an electronic m essage network. In this case, it conveys a standardized paym ent ord e r to a Ger man branch or correspondent bank, which, in turn, effects th e paym ent as a local interbank transfer in Frankfurt. orders are also provided in part as a courtesy to the brokers as part of an ongoing business relationship that makes the market more liquid. Because his limit order is often a market-maker’s first indication of general price shift, Brooks likens the posting of an order with a broker "to sticking out the chin so as to be acquainted with the moment that the fight starts.”8 Schwartz points out that posting a limit order extends a free option to other traders.9 A market-maker who calls a broker for a quote gets the broker’s inside spread, along with the quantities of the limit orders. A typical call to a broker might proceed as follows: M ongobank: “What is sterling, please?” (Mongobank requests the spot quote for U.S. dollars against British pounds (GBP).) Fonm eister: “I deal 40-42, one by two.” (Fonmeister Brokerage has quotes to buy £1,000,000 at 1.7440 USD/GBP, and to sell £2,000,000 at 1.7442 USD/GBP) M ongobank: “I sell one at 40, to whom?” (Mongobank hits the bid for the quantity stated. Mongobank could have requested a different amount, which would have re quired additional confirmation from the bid ding bank.) Fonmeister: [A pause while the deal is reported to and confirmed by Loans ’n Things] "Loans 'n Things London.” (Fonmeister confirms the deal and reports the counterparty to Mongobank. Payment ar rangements will be made and confirmed separately by the respective back offices. The broker's back office will also confirm the trade with the banks.) Value dates and payment arrangements are the same as in the direct dealing case. In addition to the payment to the counterparty bank, the banks involved share the brokerage fee. These fees are negotiable in the United States. They are also quite low: roughly $20 per million dollars trans acted.10 8See Brooks (1985), p. 25. 9See Schw artz (1988), p. 239. 10See Burnham (1991), p. 141, note 16, and Kubarych (1983), p. 14. H ’he C learing House for Interbank Paym ents System (CHIPS) is a private interbank paym ents system in New Y ork City. NOVEMBER/DECEMBER 1991 56 The final category o f participants in the fo r eign exchange market is the corporate cus tomers of the market-making banks. Customers deal only with the market-makers. They never go through brokers, who cannot adequately monitor their creditworthiness. Typically, a customer transacts with a bank with which it already has a well-established relationship, so that corporate creditworthiness is not a concern for the bank's foreign exchange desk, and trustworthiness is not an issue for the customer. The mechanics of cus tomer trading are similar to those of direct deal ing between market-makers. A customer requests a quote, and the bank makes a two-way market; the customer then decides to buy, sell or pass. The chief difference between this and an inter bank relationship is that the customer is not ex pected ever to reciprocate by making a market. Participants in the foreign exchange market also deal for future value dates. Such dealing com poses the forward markets. Active forward mar kets exist for a few heavily traded currencies and for several time intervals corresponding to active ly dealt maturities in the money market. Markets can also be requested and made for other ma turities, however. Since the foreign exchange market is unregulated, standard contract speci fications are matters o f tradition and con venience, and they can be modified by the transacting agents. Forward transactions generally occur in two different ways: outright and swap. An outright forward transaction is what the name implies, a contract for an exchange of currencies at some future value date. “Outrights” generally occur only between market-making banks and their commercial clients. The interbank market for out rights is very small, because outright trading im plies an exchange rate risk until maturity of the contract. When outrights are concluded for a commercial client, they are usually hedged im mediately by swapping the forward position to spot. This removes the exchange rate risk and leaves only interest rate risk. A swap is simply a combination of two simul taneous trades: an outright forward contract and an opposing spot deal. For example, a bank might "swap in” six-month yen by simultaneously buying spot yen and selling six-month forward " H e d g in g an o utright purchase of currency with an oppos ing swap deal still leaves an open spot purchase of the currency. This can be easily covered in the spot m arket. FEDERAL RESERVE BANK OF ST. LOUIS Figure 2 Market-Maker Volum e by Type (4/89) yen. Such a swap might be used to hedge an out right purchase of six-month yen from a bank cus tomer.11 In effect, the swapping bank is borrowing yen for the six months of the outright deal. The foreign exchange market-maker swaps in yen — rather than simply borrow yen on a time deposit — because banks maintain separate foreign exchange and money market accounts for administrative reasons. Swapping is generally the preferred means of forward dealing (see figures 2 and 3). In practice, the vast majority of foreign ex change transactions involve the U.S. dollar and some other currency. The magnitude of U.S. for eign trade and investment flows implies that, for almost any other currency, the bilateral dollar ex change markets will have the largest volume. Consequently, the dollar markets are the most li quid. The possibility of triangular arbitrage en forces the law of one price for the cross rates. The upshot is that liquidity considerations out weigh transaction costs. A German wanting 57 Figure 3 Broker Volume by Type (4/89) Options (3.5%) erature is that institutional differences can af fect the efficiency of pricing and allocation. As described above, the foreign exchange market combines two disparate auction struc tures for the same commodity: the interbank direct market and the brokered market. Defying a naive application of institutional Darwinism, whereby only the fitter of the two systems would survive, these trading methods appear to coexist comfortably.12 The direct market can be classified as a decentralized, continuous, openbid, double-auction market. The brokered mar ket is a quasi-centralized, continuous, limit-book, single-auction market. The meanings of these classifications are explained below. Centralization Spot (56.8%) -Outright Forward (0.3%) pounds, for example, will typically convert marks to dollars and then dollars to pounds, rather than trading marks for pounds directly. Though this is especially true in the American market, it holds for foreign markets as well. CLASSIFYING MARKETS The microstructure literature is by nature market-specific, and much of it concerns U. S. equity markets. This specificity has the advan tage of realism, but it makes the immediate ap plicability of some microstructural models to the foreign exchange market questionable. The first task is to define some basic microstructural con cepts, identifying where the foreign exchange market fits into the context they provide. Such a taxonomy is important, because one of the fundamental lessons of the microstructure lit 12A sim ilar situation obtains on the New Y ork Stock Ex change, where specialists act as either brokers or marketmakers, depending on the level of a ctivity in the market. 13See W olinsky (1990), p. 1. He goes on to analyze theoreti ca lly the difference in the price discovery process between centralized and decentralized m arkets. Schw artz (1988), pp. 426-35, refers to centralization as “ spatial consoli d a tio n.” In a centralized market, “trades are carried out at publicly announced prices and all traders have access to the same trading opportunities.” In a decentralized market, in contrast, “prices are quoted and transactions are concluded in private meetings among agents.”13 A New York Stock Exchange’s (NYSE) specialist system is a centralized market; the interbank direct market for foreign exchange is a decentralized one. The distinction between centralized and de centralized markets might seem to provide a neat dichotomy of possible market structures. The multiplicity of brokers in the foreign ex change market violates this simple taxonomy, however. Each foreign exchange broker accum ulates a subset of market-makers’ limit orders. This network of “brokerage nodes” is as dif ferent from a fully centralized system as it is from a fully decentralized one. This arrange m ent is labeled h ere as "qu asi-centralized.” Most microstructural studies have confined themselves to centralized markets, especially the NYSE’s specialist system and the National Associ ation of Securities Dealers Automated Quotation (NASDAQ) System on the over-the-counter (OTC) market.14 Although there are a number o f im portant decentralized markets, including the in terbank direct foreign exchange market, rela14For m odels o f specialist system s, see Dem setz (1968), Tinic (1972), G arm an (1976), Bradfield (1979), Am ihud and M endelson (1980), Conroy and W inkler (1981), Glosten and M ilgrom (1985) and S irri (1989). For studies of the OTC m arket, see Tinic and W est (1972), Benston and Hagerm an (1974), Ho and M acris (1985) and Stoll (1989). NOVEMBER/DECEMBER 1991 58 tively few studies have focused on the impact o f decentralization. There is some evidence that differences in the degree of centralization between various markets cause differences in market perfor mance. Garbade, in studying the largely decen tralized Treasury securities market, concludes that because brokerage tends to centralize trading and price inform ation, it "uses time more efficiently,” “eliminates the most important arbitrages,” and benefits dealers by ensuring that orders are executed according to price priority.15 The efficiency gains of centralized price infor mation may imply economies o f scale and, thus, a natural monopoly for brokers in securities markets. This is entirely consistent with the text book presentation of the relatively greater opera tional efficiency of centralized markets.16 Thus, the fact that a number of brokers service the foreign exchange market seems to represent a discrepancy between theory and reality. Brokers do communicate among themselves, however, to eliminate the possibility of arbitrage between limit order books. While this helps explain the multiplicity of brokers, it does not fully resolve the issue o f decentralization in the interbank direct market. Temporal Consolidation The distinction between a continuous market and a call market involves what Schwartz refers to as the degree of "temporal consolidation.”17 In a call market, trading occurs at pre-appointed tim es (the "calls”), w ith arriving transaction ord ers detained until the next call for execution. In continuous markets, like the foreign exchange market, trading occurs at its own pace, and transaction orders are processed as they arrive. A range of intermediate arrangements falls be tween these two extremes. 15See G arbade (1978), p. 497. 16The textbook argum ent counts trips to m arket. Briefly, if there are N traders, then a total of N trips to a central m arketplace are required for each to haggle with everyone else; to pair them bilaterally requires a total of N(N-1)/2 trips. If trips are costly, then centralization is more ef ficient. 17See Schw artz (1988), pp. 435-47. G arm an (1976), pp. 25758, also describes continuous and call m arkets; he refers to these as asynchronous and synchronous markets, respectively. 18See Hahn (1984), Negishi (1962), Beja and Hakansson (1977), as well as the references therein. 19A continuous m arket cannot be viewed as a continuum of in fin ite sim a lly lived call m arkets. C learing supply and de- FEDERAL RESERVE BANK OF ST. LOUIS Most microeconomic models assume call mar kets. In a Walrasian tatonnement model, for ex ample, an auctioneer calls out a series of prices and receives buy and sell orders at each price. When a price is found for which the quantities supplied and demanded are equal, all transactions are consummated at that price. Interestingly enough, Walras based this price discovery model on the mechanics of the Paris Bourse. Temporal consolidation can affect the perfor mance of a market. Theoretical work indicates how continuous trading can alter allocations, the process of price discovery and even the ulti mate equilibrium price.18 The basic thrust of these arguments is that, with continuous trading, earlier transactions satisfy some consumers and producers, causing shifts in supply and demand that affect prices for later transactions. As a result, the Pareto-efficiency characteristic of Walrasian equilibria does not necessarily obtain in continuous markets.19 On the other hand, the periodic batching of orders that occurs in a call market also has dis advantages. The difference in time between ord er placement and execution can impose real costs on investors. A recurring argument in the literature is the willingness of investors to pay more — a liquidity premium — for the ability to trade immediately. Similarly, periodic calls delay any information conveyed by prices until the time of the call, introducing price uncertainty in the period between the calls. In sum, a trade-off exists between the allocational efficiency of the nearly Walrasian call market system and the informational efficiency and immediacy of the continuous market sys tem.20 It is not clear whether the microstruc ture of the foreign exchange market represents a globally optimal balance of these relative admand in each such call m arket would require an infinite trading volum e over the course of a day. Cohen and Schw artz (1989) recom m end an e lectronic order-routing system for the stock exchanges, to fa cilita te the placem ent and revision o f orders. T his would encourage additional trading volum e, m aking m ore frequent calls feasible. 20See Stoll (1985), p. 72, and especially S chw artz (1988), pp. 442-53, for a m ore thorough exposition o f the pros and cons of tem poral consolidation. Interm ediate arrangem ents are also possible. For exam ple, Schw artz argues that m any of the problem s caused by infrequent batching in a call m arket m ig ht be overcom e by expanding access to the m arket with co m p u te r technology, w hereby the in creased num ber of traders w ould allow for m ore frequent calls. 59 vantages. A persistent deviation from optimality might be explained, for example, by arguing that the allocational benefits of a call market system are a public good. Communication o f Prices The terms "open-bid” and "limit-book” refer to ways in which price information is communi cated. In an open-bid market — the open outcry system on the futures exchanges, for example — offers to buy or sell at a specified price are announced to all agents in the market. At the opposite extreme, in a sealed-bid market, orders are known only to the entity placing the order and perhaps to a disinterested auctioneer. Direct trading in foreign exchange approxi mates the standard open-bid structure. The salient difference between the foreign exchange market and the standard arrangement is the bilateral pairing of participants in the foreign exchange market. In principle, any participant can contact a market-maker at any time for a price quote. The bilateral nature of such con tacts and the time consumed by each contact together imply, however, that all participants cannot be simultaneously informed of the cur rent quotes of a market-maker. This practical constraint on the dissemination of price infor mation is significant: it introduces the possibility of genuine arbitrage, that is, of finding two market-makers whose current bid-ask spreads do not overlap. The limit order book, which is used by both foreign exchange brokers and stock exchange specialists, is another intermediate form of price communication. Although it would be possible in principle for foreign exchange brokerage books to be fu lly open fo r pu blic inspection, in practice only certain orders — namely, the best bid and ask on each book — are revealed to market-makers, while the others remain con cealed. As in the direct market, market-makers must contact brokers bilaterally to get these "in side spreads.” Knowledge of the concealed limit orders would be of speculative value to marketmakers, because an imbalanced book suggests that large future price movements are more likely in one direction than the other. More generally, price communication is inti mately related to the role of market-makers as 21This term is due to Dem setz (1968), p. 35. Tinic (1972), p. 79, calls in “ liquidity se rvices.” providers of "predictable immediacy.”21 Market participants are willing to pay a liquidity pre mium, usually embedded in a market-maker’s spread, for the reduction in search costs im plied by constant access to a counterparty. The costs of "finding” the other side of a transaction can be further broken down into the liquidity concession, the cost o f communicating the in formation and the cost of waiting for potential counterparties to respond.22 Other things equal, an efficient system o f price communication is one that minimizes such transaction costs. While the communication of price information is a central function of securities markets, the fact that the systems of price communication in the foreign exchange market are not fully central ized suggests that these systems do not represent a cost-minimizing arrangement. Structure o f Prices The terms "double-auction” and "single-auction” refer to the nature o f the prices quoted. In a double-auction market, certain participants pro vide prices on both sides of the market, that is, both bid and ask prices. Participants providing double-auction quotes upon demand are known as market-makers, and they must have sufficient capitalization to back up their quotes. In a single auction market, prices are specified either to buy or to sell, but not both. In the foreign ex change market, market-makers provide double auction prices, while brokers try to aggregate single-auction quotes into tw o-w ay (inside) spreads. A broker's book may occasionally be empty on one or both sides. Rather than make a market in such cases, the broker provides, respectively, a single-auction quote or n on e at all. Thus, whether double or single-auction prices are quoted depends largely on w hether the agent quoting prices is providing market-making services or simply attempting to acquire (or sell) the commodity. This issue is related to the degree o f centralization in the market. The absence of market-makers in a single-auction market, together with the presence of search costs, results in a tendency toward centraliza tion of price information, thus facilitating the search for a counterparty. Inversely, decentrali zation of price information leads to a tendency 22See Logue (1975), p. 118. NOVEMBER/DECEMBER 1991 60 toward double-auction prices, again to facilitate the search for a counterparty.23 MODELING TRADERS’ REHAVIOR The microstructure literature extends well be yond a simple description of market institutions. Modeling the behavior of market participants is central to almost all discussions o f microstructure. Although numerous approaches to such modeling have been taken, two common concerns are of special interest. These are the treatment of price information by market par ticipants, and determination of the bid-ask spread. The latter raises the interrelated issues of inven tory and quantity transacted. Price Expectations Modeling the interpretation of price informa tion is a crucial step in constructing microstructural models of price discovery.24 Many diverse approaches have been taken in such modeling. An almost universal simplification is to model securities markets in partial equilibrium, so that prices are not determined endogeneously in the traditional general equilibrium sense. This allows the modeler to focus on the microstructure's finer details. Another common simplification is to assume that agents ignore the impact o f their own behavior on the market.25 Rather than explicitly model such forces as general equilibrium or recursive beliefs, models posit probability distributions that produce the prices of orders in the market. Modelers have included randomness at one or both of two lev els, depending on their focus. First, order prices can be generated by objective distribu tions, that is, by stochastic processes exogenous 23Note that the converse does not appear to hold. T hat is, centralization does not tend to elim inate double-auction quoting. For exam ple, the NASDAQ system on the OTC stock m arket centralizes price inform ation w hile still sup porting num erous m arket-m akers for every stock. 24Notably, the term “ p rice ” is generally too inexact in a m icrostructural context. One m ust often distinguish at a m inim um between quoted prices, transaction prices and equilibrium prices. There are also reservation prices, m arket-clearing prices and closing prices (see Schwartz (1988), chapter 9, for the d istinction between equilibrium and clearing prices). If unspecified here, the intended defi nition should be clear from the context. 25The alternative, w hich dates at least to Keynes' “ beauty c o n test,” is recursive beliefs, in which an agent considers the feedback of her own actions on the beliefs of others, and thence how the behavior on the other agents m ight af fect her own beliefs, etc. See Keynes (1936), p. 156. The lim iting case— an infinite recursion of beliefs— presum es FEDERAL RESERVE BANK OF ST. LOUIS to the market. For example, there may be a stochastic process that generates the “ tru e” equilibrium price. Second, probability models of participants’ subjective beliefs about prices can be used. Conroy and Winkler, for example, at tribute subjective normal price distributions to market-makers, who use Bayesian updating to learn about the prices of incoming limit orders.26 Objective processes can coexist with subjective beliefs about those processes. Harsanyi suggests a consistency requirement for the subjective price distributions of multiple agents; these dis tributions are each equated with a conditional distribution of a single distribution known to all.27 Models can be further classified according to how they relate supply and demand. In particu lar, there are both models with single price processes and with dual price processes. In dual price models, purchase orders (whether market or limit orders) are generated by one process, while sale orders are generated by another.28 The salient point here is that purchase and sale orders come from independent distributions. This independence is especially clear in Conroy and Winkler, where the distributional assump tions are explicit; there, independence implies that any sequence of buy orders, regardless of their prices and quantities, has no effect on the subjective probability of a sell order at any price. Statistical independence implicitly restricts the ways in which orders can be generated. Pur chase and sale orders are somehow motivated independently, although the cause o f this separation is not always specified. Statistical in dependence is not a necessary component of a dual price process, however. Cohen, Maier, Schwartz and Whitcomb (1981), for example, as sume that actual market bid and ask prices are extrem e inform ational and com putational resources on the part of agents, and m odels based on it are usually intrac table. Interm ediate approaches allow ing a finite degree of recursion m ust som ehow ju stify the truncation of recursive beliefs, just as the standard m odel of atom istic agents al lows no beliefs about beliefs and is ju stifie d by an as sum ption on the relative size of individual agents. 26See shaded insert on opposite page. 27See Harsanyi (1982), especially chapter 9, and the refer ences therein. His consistency requirem ent identifies a unique e q uilibrium for the game. 28A m arket order is an order to trade at the best price avail able; a lim it ord e r specifies a price. These m odels represent a strain of the literature that was pioneered by Dem setz using straightforw ard supply and dem and sched ules (see shaded insert on page 63). S im ila r approaches were later taken by G arm an (1976), Am ihud and Mendelson (1980) and Conroy and W inkler (1981), am ong others. 61 Bayesian Learning of Price Inform ation Conroy and Winkler (1981) developed a Bayesian model of market-maker price expec tations, which is outlined here. Consider an expected-profit-maximizing, monopolistic market-maker who faces streams o f buy and sell limit orders from investors.1 All orders are for a single round lot. Assume that the market-maker believes that reservation prices of buy orders, pd, are generated by a normal distribution, Fd(pd:^d,od); reservation prices of sell orders, ps, are generated by a second, in dependent, normal distribution, Fs(ps:/is,os). That is, the market-maker has two indepen dent, normal, subjective price distributions (with corresponding densities fd and fs). Fur ther assume that the market-maker currently holds his desired inventory level. How should he set his spread? The inventory condition implies that the chosen bid and ask rates, B and A, must satisfy the constraint, FS(B) = l - F d(A), so that the expected change in inventory is zero.2 Given this constraint, the expected profit per period is: E(tt) = (A -B ) FS(B) = (A -B ) ( l - F d(A)). Maximizing this over B and A yields: B* = M - os<D(B')/0(B') and A* = M + odO(A')/<£(A'), where M - (od^s+ osfid)/(od+ o5), B' = (B-pts)/os, A ' = (A -/^d)/od, and <£(•) and 0(0 are the stan dard normal density and distribution func tions. It can be shown that this optimal spread shrinks (i.e., A * - B * decreases), ceteris paribus, as the subjective variances, os and od decrease. The important aspect of this study is that it provides an explicit mathematical model for a market-maker’s interpretation o f price infor mation. The market-maker is assumed to behave in a Bayesian fashion, using the observed prices on incoming limit orders to 'C o n ro y and W in kle r (1981) also consider a risk-averse m arket-m aker and the inform ation conveyed by a m ar ket order, w hich does not specify a price. They do not incorporate the im pact of inventory on p ricing, nor do they generalize beyond the unrealistic assum ption of norm ally distributed prices. Profit Per Pair of Trades and Expected Number of Trades Per Unit Time f d(p„) p Profit per pair of trades Expected num ber of sell trades *1 N— Expected num ber of buy trades refine the parameters o f his subjective distributions. For example, assume that the market-maker views purchase prices as com ing from a normal distribution gd(pd:^d,od), but is unsure about the mean of this distribution. Represent this uncertainty by a normal prior density h'(^d:m',v') over the possible values for the mean, ptd. Given this, the marginal subjective density over the prices of incoming limit orders, fd(pd) = Jgd(pd)’h'(^d), is normal with mean m' and variance (od2+ v '2). Follow ing a sample of n buy orders with mean price m, the market-maker is able to refine his subjective distribution of the mean. The posterior parameters o f h"(^d:m ",v") are m " = (m'/v'2+ nm/od2)/(l/v'2 + n/od2) and v " = l/(l/v'2 + n/od2). The upshot of this refined es timate is that the variance of the marginal subjective price density, fd(p), is now smaller, and the market-maker's optimal spread, (B*, A*), is narrower. the interval from B to A, and the inventory constraint is satisfied w hen the two shaded ta ils have equal area. T heir optim ization problem is sim ilar in sp irit to that of Allen (1977), although the latter does not consider learning. 2T his is depicted in the figure above, w here price is on the horizontal axis, and the relative frequency of orders is on the vertical axis. The m arket-m aker’s spread is NOVEMBER/DECEMBER 1991 62 independent Poisson processes and give inves tors joint subjective distributions over those prices. For the latter distributions, probabilistic independence of bid and ask prices is not ex plicitly required. Black (1989) models quantities (independent o f prices) of market orders. Quan tities supplied and demanded are drawn from different distributions, but the distributions are constrained to have the same mean. Garbade (1978), on the other hand, assumes a single, unknown and fixed equilibrium price, around which market-makers set their spreads. Incom ing buy and sell orders arrive via random processes whose mean arrival rates depend on the difference between the quoted bid (or ask) price and the exogenous equilibrium price and, thus, are not independent. The most common alternative to separate pur chase and sale processes is to model prices as some function of a single scalar process. This approach is in the spirit of the efficient markets literature, which posits a unique value for a security conditional on the available informa tion. Ross (1987) points out that this approach can be regarded conceptually as a special case of the dual price process, with supply and de mand infinitely elastic at a common price. Many authors reveal their theoretical roots by using terminology drawn from the literature on effi cient markets. Thus, for example, Barnea des cribes a stock’s "intrinsic value,” which follows a random walk.29 Similarly, Copeland and Galai posit a "'true' underlying asset value ... known (ex ante) to all market participants.”30 In con trast, Garbade’s (1978) exogenous equilibrium price is unknown. It is possible to extend the single price ap proach beyond the efficient markets tradition by modeling the value of a security subjectively rather than as an objective fact. Glosten and Milgrom (1985), for example, begin with an ex ogenous random value representing the consen sus value of a stock given all public information. Investors do not act on this exogenous value directly; instead, they act on their expectation of it, conditional on their information set. Ho and Stoll personalize price expectations in a similar fashion:31 29See Barnea (1974), pp. 512-14. 30See Copeland and Galai (1983), p. 1458. 31See Ho and Stoll (1981), p. 48. For a sim ilar exam ple, see Stoll (1978), especially p. 1136. FEDERAL RESERVE BANK OF ST. LOUIS We take the dealer's opinion of the "true’’ price of the stock to be exogenously determined by his in formation set and ask how the dealer prices rela tive to his "true” price... This subjectivization of the pricing process is significant, because it allows for heterogeneous expectations and thus for more realistic model ing of price discovery. Research into the microstructure o f the for eign exchange market should presume such het erogeneity among market-makers. There are numerous market-makers for foreign exchange: The Federal Reserve Bank of New York (FRB-NY) (1989a) lists 162 dealing institutions in the U.S. interbank market. There would be little point in such superfluity if all market-makers were iden tical. Furthermore, it is well known that “taking a view/’ that is, speculating on future prices, is routine for many participants.32 To omit this heterogeneity from a model is to ignore an im portant characteristic of the market. The large proportion o f market-makers in the foreign exchange market has another important modeling implication. It implies that a single price process is more appropriate as a theoreti cal representation of agents’ expectations. Market-makers consistently face other market-makers, who can hold positive or negative inventories of foreign currency with equal ease. A quote that is “o ff the market” on the high side will be hit (i.e., traded upon) just as surely as a quote that is o ff on the low side. This is also true of cus tomers, who normally enter the market with a predilection to either buy or sell. As Burnham notes:33 The customer knows that if the first marketmaker is too far off the market price, he can unexpected ly take the other side of the quote and resell the position to a second marketmaker. The point is that the market-maker must expect to be penalized for underestimating as well as overestimating his counterparty’s valuation of the currency. From the perspective of the mar ket-maker, who quotes a spread and observes a response, the forces determining short-run ef fective demand and supply are not merely re lated, but indistinguishable. 32See, for exam ple, Kubarych (1983), p. 29, or Burnham (1991), p. 139. 33See Burnham (1991), p. 136. 63 Dealer Services and the Bid-Ask Spread Traditional wisdom refers to the bid-ask spread as the “jobber’s turn/’ suggesting that it provides compensation to the dealer for the provision of services.1 Demsetz (1968) formalized this rationale for the spread, defining the particular service provided as “predictable im mediacy” and offering a simple model to describe the spread. Market Buy and Sell Curves with and without the Provision of Immediacy Price Consider a continuous market with aggregate supply (sell) and de mand (buy) schedules, S and B, for a security (see figure at right). In an idealized world, investors would come together simultaneously, and the market would clear at price P* and quantity Q*. In this market place, however, such coordination o f trading is impossible. By assump tion, the market is continuous, and there is no mechanism (e.g., a limit order book) for holding orders over time. Thus, S and B do not represent standard static supply and demand schedules, but rather time rates of supply and demand. At any given instant, there may be no orders on either side. Instead, we introduce a monopolistic mar ket-maker who allows the trading to occur by standing in as a counterparty to all trades. In the process, he provides a service to investors that Demsetz labels “predictable immediacy.” The market-maker knows the aggregate sup ply and demand propensities. The supply and demand curves that he presents to the public, S' and B', however, are both shifted leftward. Investor purchases clear for price Pa, at the intersection o f the market demand schedule, B, with the market-maker’s supply schedule, S'. Similarly, investor sales clear at the inter section o f S and B', for price Pb. The differ ences: Pa- P * and P * - P bare liquidity premia. In the figure, the quantities, Q r, purchased and sold by the market-maker happen to be equal, so that no market-maker inventory is accumulated. His profit thus equals QT(Pa- P b); this is th e "jo b b er's tu rn .” 1See, for exam ple, Keynes (1936), p. 158, or Stigler (1964), p. 129. A market-maker's constant contact with other well-capitalized market-makers implies that this is not a theoretical fine point. In the words of one market-maker:34 Ninety percent of what we do is based on percep tion. It doesn’t matter if that perception is right or wrong or real. It only matters that other people in the market believe it. I may know it’s crazy. I may think it’s wrong. But I lose my shirt bv ignoring it. In other words, as a direct implication of their readiness to buy or sell, market-makers must strive first to achieve a price consensus. The im- 34Jam es Hohorst, as quoted by M ossberg (1988), p. 29R. Mr. Hohorst directed foreign exchange trading in North A m erica for M anufacturers Hanover. NOVEMBER/DECEMBER 1991 64 perative o f arbitrage avoidance must be re garded as the first priority in individual marketmaker pricing, to which all other factors (e.g., purchasing power parity) must be subordinated. Market-m akers’ Bid-Ask Spreads The bid-ask spread has attracted considerable interest in the literature on market microstructure. The complexity of modeling the spread is largely because it requires incorporating a sub stantial amount of institutional detail. At a facile theoretical level, a market-maker's spread appears to be a direct violation of the law of one price, since it assigns tw o prices to the same com modity. Several explanations have been offered to resolve this seeming inconsistency. They can be roughly categorized as involving the cost of dealer services, the cost of adverse selection and the cost o f holding inventory.35 The dealer services argument can be traced back at least as far as Stigler (1964), who argues that stock exchange specialists charge a "job ber’s turn” as compensation for the costs of act ing as a specialist. The analysis o f dealer services was formalized bv Demsetz (1968), who identi fied "predictable immediacy” as the particular service for which investors are willing to pay. This identification hints at the complex question of what liquidity is and where it comes from. In a busy market, liquidity is a public good: a con tinuous stream of buyers and sellers generates predictable immediacy as a by-product of their trading. The determinants o f the level of compensation are themselves a topic of debate. Stigler argues that, because centralization of exchange limits fixed costs and aggregates separate transaction orders into less risky actuarial order flows, it implies economies of scale and thus a natural monopoly for market-making.36 Smidt (1971) counters that barriers to entry among NYSE 35This is essentially the sam e taxonom y as provided by Barnea and Logue (1975), although they use the term s “ liquidity th e ory,” “ adversary th e o ry,” and “ dynam ic price/inventory adjustm ent th e o ry,” respectively. 36See S tigler (1964), p. 129. 37See Sm idt (1971), p. 64. 38For exam ple, in the context of the OTC stock m arket, Benston and Hagerm an (1974), p. 362, conjecture that, “ deal ers may face positively sloped m arginal cost curves which shift down as industry o utput in creases.” The idea is that m arket-m aking per se is not a natural m onopoly, even FEDERAL RESERVE BANK OF ST. LOUIS specialists allow them to exact monopoly rents from other investors. In his view, the natural monopoly argument, while used as an apology for barriers to entry, remains unsupported em pirically: “There is no empirical evidence to sup port the proposition that [market-making] is, in fact, a natural monopoly.”37 Indeed, if marketmaking is a natural monopoly, barriers to entry should be unnecessary. The foreign exchange market has no apparent barriers to entry other than the need for suffi cient capitalization. It also has no apparent bar riers to exit. The market supports a large and increasing number o f competing market-makers. Unless it can be shown that there is some sub tle restriction in the foreign exchange market that prevents consolidation of the market-making function, one must conclude that market-making per se is not a natural monopoly.38 The multi tude of market-makers also implies that they cannot earn monopoly rents by embedding a premium fo r predictable immediacy in the spread, although the spread may still cover the costs of processing orders. Other research suggests that a market-maker’s job is m ore complex than the m ere sale o f counterparty services. A second explanation for the bid-ask spread — adverse selection — can be traced to Bagehot (1971). He starts with “ liquidity-motivated transactors” who pay the market-maker the price o f the spread in ex change fo r the service o f predictable immedi acy. The market-maker also confronts traders who have inside information, however, and who can therefore speculate profitably at the expense of the market-maker.39 The market-maker must charge everyone a wider spread to compensate for losses to the information-motivated traders. Because of the relatively abstract nature of currencies as commodities, it is difficult to con struct examples o f “ inside” inform ation on foreign exchange rates. One exception is money supply announcements, which, if known before though the industry as a whole experiences econom ies of scale. Ham ilton (1976) also addresses th e natural m onopo ly question; Reinganum (1990) provides evidence on li qu idity prem ia for NYSE vs. NASDAQ stocks. 39This situation is called adverse selection, because, in a m arket with com peting m arket-m akers, the one who gets the in sider’s business is a loser rather than a winner. Bagehot also posits a th ird class of investors, who only th in k they have inside inform ation; th e y speculate, but lose on average, and are in d istinguishable to the m arket-m aker from the liquidity-m otivated traders. 65 publicly distributed, might provide a basis for profitable speculation. Another form of informa tion that can be construed as inside information is knowledge of an arbitrage opportunity. Con sider a hypothetical market in which there are numerous decentralized market-makers who do not quote spreads, but single prices at which they are willing both to buy and sell. Unless there were a perfect consensus among the mar ket-makers on the value of the foreign currency, all of them would be vulnerable to arbitrage. A decentralized market makes a perfect consensus difficult to achieve. Without centralizing price information, it is impossible to know if no arbi trage opportunities exist. A bid-ask spread, in contrast, allows a market-maker to include an error tolerance in her prices, thus facilitating a price consensus: it is easier to get bid-ask spreads to overlap than to get scalar prices to coincide. The spread also provides the m ar ket-maker with some degree o f protection from adverse selection in the form of arbitrage. The bid-ask spread is also affected by invento ry considerations. This idea dates back at least as far as Barnea and Logue (19 75).40 The notion of a desired inventory level for the marketmaker underlies all o f these models. In the simplest case, the desired level is set at zero, and a constant spread is shifted up and down on a price scale to equalize the probability of receiving a purchase order with that of receiv ing a sale order. The result is that the expected change in inventory is always equal to zero, and (with all trades for one round lot) the inventory level follows a simple random walk. An undesirable implication of random-walk models of inventory is the inevitable bankruptcy of the market-maker. Finite capitalization levels for market-makers impose upper and lower bounds on allowable inventories. Because inven tory follows a random walk, with probability one it will reach either its upper or lower bound in a finite number of trades.41 The dynamic op timization models of Bradfield (1979), Amihud and Mendelson (1980) and Ho and Stoll (1981) resolve this problem. They conclude that a market-maker, optimizing his bid and ask prices 40Barnea and Logue a ttribute it to Sm idt (1971), although S m id t’s paper does not explicitly develop the connection between the m arket-m aker’s inventory and his spread. For mal m odels of the relationship between inventories and spreads can be found in Stoll (1978), Am ihud and M endel son (1980), Ho and Stoll (1981) and Sirri (1989), am ong others. 41See, for exam ple, Ross (1983), pp. 106-07. over time in the face of a stochastic order flow, will shift both bid and ask rates downward (up ward) and increase the width of the spread when a positive (negative) inventory has accumulated.42 We should expect two of these three ration ales for the spread to apply to market-makers’ bid-ask spreads in the foreign exchange market. Because there are numerous market-makers, competition should eliminate their ability to earn monopoly rents by charging a premium for pre dictable immediacy per se. The adverse selec tion argument does apply in the foreign ex change market, however, since the spread allows market-makers some protection against arbit rage opportunities. Arbitrage opportunities can be construed as a form of inside information in a market where price information is not cen tralized. In accordance with the dynamic optimi zation models, a market-maker’s inventory level should affect the spread, widening and shifting it as inventories accumulate. B rok ers’ Spreads So far, the discussion of the bid-ask spread has focused on models in which bid and ask prices are set by individual market-makers. The dual role o f the stock exchange specialist sug gests that this is only part of the story. Spreads are produced in two fundamentally different ways. It is only when limit orders are sparse that a NYSE specialist must step in as a marketmaker to provide an “orderly market.”43 When limit order volume is sufficient, the specialist acts as a broker, accounting for incoming limit orders on the limit order book, and pairing market orders against them. Cohen, Maier, Schwartz and Whitcomb (1979) note that inade quate attention has been given to the fact that not all prices are market-maker spreads. The market often makes itself without specialist as sistance, through the aggregation o f limit ord ers on the book. The foreign exchange market differs from the NYSE in that the market-making and brokerage roles are separated: market-makers do not act as brokers, and brokers do not make markets. 42See shaded insert on page 66. 43The NYSE defines this role in rule 104: “ the specialist should m aintain a continuous m arket w ith price continuity and close bid and asked prices, and m inim ize the e ffect of tem porary d isparity between public supply and d em and.” See Leffler and Farwell (1963), pp. 211-12. NOVEMBER/DECEMBER 1991 66 Dynamic Price-Inventory Adjustment Models Amihud and Mendelson (1980, 1982) pro vide a model of market-maker spread-setting that takes inventory into account. Assume that a market-maker faces order flows of buy and sell market orders that arrive according to independent Poisson processes. The buy and sell arrival rates (i.e., process intensities), d and s, respectively, depend on the ask and bid prices, P., and Pb, that the market-maker quotes: d = D(P.,) and s = S(P1}). Denote the inventory level by k € {-Z ,...,A }, where X and A are the largest allowable short and long positions, respectively. Let dkand skde note the order arrival rates when prices are set as functions of the inventory level: dk = D(Pa(k)) and sk = S(Ph(k)). ' <dk+ sk) = dk • Pa(k) - sk • Ph(k). The market-maker's objective is to maximize the expected profit per unit time, given by: „ = £ 0 kQ(k). k= - X Where 0 is the probability of being at inven tory level k. The solution to this optimization problem gives the values for Pa(k) and Pb(k), which are depicted in the figure below. The market-maker controls inventory by adjusting prices up (down) to make an investor sale (purchase) more likely when inventory is low (high). The spread must widen as the invento ry nears its bounds, in order to avoid the problem o f a random walk for inventory; at the extremes, d x = sA = 0. The expected sojourn at k (i.e., the time un til an order arrives) is given by the Poisson processes known as l/(dk+ sk). The probability that the next order will be a buy order is dk/(dk+ sk), and the probability that it will be a sell order is sk/(dk+ sk). Thus, the expected cash flow per unit time at position k is given bv: Optimal Prices and Bid-Ask Spread as a Function of Inventory Position Price k' Therefore, it is even more appropriate to model brokered spreads as determined in a fundamen tally different way from market-maker spreads. The separation of roles also has other implica tions for modeling foreign exchange brokerage. A brokered spread is the combination of the best bid and best ask, received by the broker as FEDERAL RESERVE BANK OF ST. LOUIS k -1 k In v e n to ry Po sitio n separate limit orders. This arrangement might be modeled as a pair of extreme order statistics from independent distributions of purchase and sale limit orders. The distribution of these statis tics would have to he conditional on limit order volume and on the fact that the best ask must always exceed the best bid, since crossing ord 67 ers transact immediately and are removed from the hook.44 Perhaps because of its complexity, such a derivation has not been attempted. Cohen, Maier, Schwartz and Whitcomb (1979) model limit orders as generated by ' vawl” distri butions. These distributions satisfy three heu ristics for the incentives of investors placing limit orders.45 The heuristics are motivated by a notion of the centralized exchange as a market for immediacy; placers of limit orders produce immediacy, and placers of market orders con sume it. This relationship between limit and market orders is formalized in Cohen, Maier, Schwartz and Whitcomb (1981), where each half of the brokered spread is assumed to be gener ated by a compound Poisson process. A mini mum brokered spread results: if the limit order’s bid (ask) price is sufficiently close to the special ist’s ask (bid), the benefit to the investor of being able to specify the price of a limit order is over whelmed by the cost of foregone immediacy. Because models of the informational content of brokered spreads are few, the literature offers little guidance in modeling brokered quotes in the foreign exchange market. The yawl distribution is the only explicit distribu tional form for brokered spreads in the litera ture. Unfortunately, its heuristic basis cannot be transferred directly to the foreign exchange market, because market-makers there differ from stock market investors. Indeed, this may be an instance in which the foreign exchange market informs microstructure theory rather than the other way around. The extant ap proaches to brokerage treat it as a service facilitating predictable immediacy. This aspect of brokerage is redundant in the foreign exchange market, because of the multitude of marketmakers, each providing immediacy. This red u n dancy suggests instead that foreign exchange brokerage serves some other function. One motive for trading through a foreign ex change broker is to maintain anonymity — the name of the bank placing a limit order is not revealed unless a deal is consummated and then only to the counterparty.46 Anonymity is valu able, because revealing a need to buy or sell a 44An order statistic is defined as follows: the sam ple realiza tions of a fin ite num ber of independent random variables are ranked in increasing order, and the kth order statistic is the kth num ber in that list. For the foreign exchange m arket, the m odeling is still m ore com plex, since brokers com pare books am ongst them selves in the sense that in com ing orders can cross against any book. currency puts a market-maker at a bargaining disadvantage. In addition, anonymity can help pair market-makers who ordinarily would not contact each other directly. These issues have not been explored at a theoretical level. Until an adequate microstructural model of the strategic benefits of anonymity is developed, the theoreti cal understanding of foreign exchange broker age will be limited. CONCLUSIONS Students of the foreign exchange market can draw several lessons from the literature on market microstructure. The most fundamental of these is that the institutional details of ex change in a market can affect all aspects — price, allocational, informational and operational — of the market’s efficiency. A multitude of market-makers who can provide liquidity, or predictable immediacy, arises in response to the decentralization of the market. As a result, search costs are reduced relative to a world without market-makers, because finding one of many market-makers amounts to finding a counterparty. Brokerage also reduces search costs by achieving a degree of centralization in price information. An unanswered question is why the specific combination of trading structures characteristic of the foreign exchange market — a decentral ized, open-book, direct arrangement and a quasi-centralized, limit-book, brokered arrange ment — should coexist. Apparently, each struc ture has relative advantages, but a full analysis of these advantages is lacking. Is there a single microstructure that would combine the relative advantages of the direct and brokered arrange ments? Put another way, why does the microstructure of the foreign exchange market differ from that of the stock exchanges, the futures pits and the OTC stock market? Answering these questions will require a fuller specification of the objectives of a trading system and a better understanding of the impact of microstructural arrangements on those goals. These issues provide a motive for deeper in vestigation of the behavior o f the foreign ex- 45The yawl distribution, named for its resem blance to a sail boat, is a probability d istribution contrived for m odeling the generation o f buy (or sell) lim it orders. See Cohen, Maier, Schwartz and W hitcom b (1979, 1983, 1986) for details. 46See Kubarych (1983), p. 16, Burnham (1991), p. 141, and Federal Reserve Bank of New York (1989b), p. 41-3. NOVEMBER/DECEMBER 1991 68 change market and its participants. Marketmakers are the crucial element: they provide all transaction prices in the market and are in volved in at least one side of every deal. The microstructure literature has developed numer ous models of the interpretation and setting of prices by traders. The diversity of expectations models used in the literature illustrates the im portance of tailoring such models to the specific environment confronted by market participants. Given that a foreign exchange market-maker’s double-auction quote can be hit on either side (bid or ask) with equal ease, he must try to maneuver his spread to bracket the market’s consensus valuation of the foreign currency. In other words, suppliers and demanders of cur rency are indistinguishable to the market-maker ex ante. The inability to separate the forces de termining effective demand from those de termining effective supply in the very short run imply that a single-price expectations process (rather than a dual-price process) is appropriate in modeling market-makers in the foreign ex change market. A market-maker’s bid-ask spread serves several purposes. Competition among market-makers in the foreign exchange market implies that they should be unable to charge a monopoly premi um for the service of predictable immediacy. In stead, the spread obviates the need for perfect price consensus by giving the market-maker some protection from arbitrageurs with superior price information. While arbitrage avoidance must be considered a primary goal in setting a market-maker’s bid and ask quotes, the spread provides flexibility elsewhere. Just as a rbitrage avoidance is concerned with accurately estimat ing current prices, speculation is concerned with estimating future prices. By changing in size and shifting up or down, the spread can control stochastically the market-maker’s foreign currency inventory in the face of random order flows. Systematic empirical study of the effect of inventories on market-makers’ spreads is still needed, however. The brokered spread is less well understood than the market-maker’s spread, and certain areas are ripe for further research. Theoretical models of brokered spreads are few. The exist ing rationales for brokerage maintain that it provides liquidity services. In the foreign ex change market, however, numerous marketmakers make the liquidity services provided by brokerage superfluous. Descriptions of the for eign exchange market suggest instead that FEDERAL RESERVE BANK OF ST. LOUIS anonymity is an important motive for trading in the brokered market. Yet the strategic value of anonymity in foreign exchange quoting is not well understood at a theoretical level. In addi tion, there is not a clear understanding of the differences in price information between a market-maker’s spread and a broker’s spread; this too remains a topic for future research. From a broader perspective, a better under standing of institutional choice and change as regards securities market microstructure is necessary. Most microstructural research has been devoted to analyzing the impact of microstructural factors on important economic vari ables, such as price and allocation. Relativ ely little attention has been paid to the effect of economic factors on the choice of an institution al microstructure. REFERENCES Allen, W iliam A. “A note on uncertainty, transaction costs and interest parity,” Journal o f M onetary Econom ics (July 1977), pp. 367-73. A m ihud, Yakov, T hom as S. Y. Ho, and Robert A. Schwartz, eds. M arket M aking a nd the Changing Structure o f the Securities Industry, (Lexington Books, 1985). Am ihud, Yakov, and Haim M endelson. “ D ealership Market: M arket-M aking with Inventory,” Journal o f F inancial Eco nom ics (March 1980), pp. 31-53. ________ . “Asset Price B ehavior in a Dealership Market,” Financial Analysts Journal (May/June 1982), pp. 50-59. ________ . “Asset Pricing and the Bid-Ask Spread,” Journal o f Financial Econom ics (Decem ber 1986), pp. 223-49. ________. “ Liquidity and Asset Prices: Financial M anage m ent Im plications,” Financial M anagem ent (Spring 1988), pp. 5-15. Bagehot, Walter. “ The O nly G am e in Town,” Financial A n alysts Journal (M arch/April 1971), pp. 12-14, 22. ________ . Bank for International Settlements. “ Survey ot Foreign Exchange Market Activity,” (February 1990). Barnea, Amir. “ Perform ance Evaluation of New York Stock Exchange Specialists,” Journal o f Financial a n d Quantitative Analysis (Septem ber 1974), pp. 511-35. Barnea, Amir, and Dennis E. Logue. “ The Effect of Risk on the Market M aker’s Spread,” Financial Analysts Journal (N ovem ber/D ecem ber 1975), pp. 45-49. Beja, Avraham, and Nils H. Hakansson. “ D ynam ic Market Processes and the Rewards to Up-to-date Inform ation,” Journal o f Finance (May 1977), pp. 291-304. Benston, George J., and Robert L. Hagerm an. “ Determ inants of Bid-Asked Spreads in the O ver-the-Counter Market,” Journal of Financial Econom ics (1974), pp. 353-64. Black, Stanley W. “ Transaction Costs and Vehicle C urren cies,” IMF VJorking Paper WP/89/96, International M onetary Fund (Novem ber 21, 1989). 69 Bossaerts, Peter, and Pierre H illion. “ M arket M icrostructure Effects of G overnm ent Intervention in the Foreign Ex change Market,” Review o f Financial Studies (forthcom ing). B ra d fie ld , Jam es. “ A F orm al D yn a m ic M od e l of M arket M aking,” Journal o f Financial and Quantitative Analysis (June 1979), pp. 275-91. Brooks, J. A. “ The dealing room and the dealer,” in Rudi Weisweiller, ed., M anaging a Foreign Exchange Departm ent (W oodhead-Faulkner, Cam bridge, 1985), pp. 18-33. Burnham , Jam es B. “ Current Structure and Recent Develop ments in Foreign Exchange Markets,” in Sarkis J. Khoury, ed., Recent Developments in International Banking and Finance, Volume IV (Elsevier Science Publishers, Am ster dam , 1991), pp. 123-53. Chrystal, K. Alec. “A G uide to Foreign Exchange Markets,” this Review (March 1984), pp. 5-18. Cohen, Kalman J., G abriel A. Hawawini, Steven F. Maier, Robert A. Schwartz, and David K. W hitcom b. “ Im plications of M icrostructure Theory for Em pirical Research on Stock Price Behavior,” Journal o f Finance (May 1980), pp. 249-57. Cohen, Kalman J., Steven F. Maier, W alter L. Ness, Jr., Hitoshi O kuda, Robert A. Schwartz, and David K. W h it comb. “ The im pact of designated m arket makers on securi ty prices,” Journal o f Banking a nd Finance (Novem ber 1977), pp. 219-47. Cohen, Kalman J., Steven F. Maier, Robert A. Schwartz, and David K. W hitcom b. “ The Returns G eneration Process, Returns Variance, and the Effect of T hinness in Securities Markets,” Journal o f Finance (March 1978a), pp. 149-67. ________ . “ Lim it Orders, M arket Structure, and the Returns G eneration Process,” Journal o f Finance (June 1978b), pp. 723-36. ________ . “ Market Makers and the Market Spread: A Review of Recent Literature,” Journal of Financial a nd Quantitative Analysis (N ovem ber 1979), pp. 813-35. ________. “ Transaction Costs, O rder Placem ent Strategy, and Existence of the Bid-Ask Spread,” Journal of Political Econ omy (April 1981), pp. 287-305. ________. “A sim ulation m odel of stock exchange trading,” Simulation (1983), pp. 181-91. ________. The M icrostructure o f Securities Markets (PrenticeHall, Englewood Cliffs, 1986). Cohen, Kalman J., and Robert A. Schwartz. “An Electronic Call Market: Its Design and Desirability,” in H enry C. Lucas, Jr. and R obert A. Schwartz, eds., The Challenge of Inform a tion Technology for the Securities Markets: Liquidity, Volatili ty, a nd G lobal Trading (Dow Jones-lrw in, Homewood, 1989) pp. 20-58. Conroy, Robert M., and Robert L. Winkler. “ Inform ational Differences between Lim it and Market Orders for a Market Maker,” Journal o f Financial and Quantitative Analysis (Decem ber 1981), pp. 703-24. Copeland, Thom as E. “A M odel of Asset Trading under the Assum ption of Sequential Inform ation Arrival,” Journal of Finance (Septem ber 1976), pp. 1149-68. Copeland, Thom as E., and Dan Galai. “ Inform ation Effects on the Bid-Ask Spread," Journal of Finance (Decem ber 1983), pp. 1457-69. Demsetz, Harold. “ The Cost of Transacting,” Quarterly Journal of Econom ics (February 1968), pp. 33-53. Fama, Eugene F. Foundations of Finance (Basic Books, 1976). Federal Reserve Bank of New York. “ Sum m ary of Results of U.S. Foreign Exchange Market Turnover Survey Conducted in April 1989,” (Septem ber 13, 1989a). ________ “ The Foreign Exchange C om m ittee: Annual Report 1988,” (June 1989b). ________ . “ G ross Foreign Exchange Transactions by Banks in U. S. M arket Exceeds $23 Billion Daily: Broker Turnover Separately Am ounts to $8.5 Billion Daily, New York Fed Survey Shows,” (June 23, 1980). Flood, M ark D. “ Market S tructure and Inefficiency in the For eign Exchange Market,” W orking Paper 91-001A, Federal Reserve Bank of St. Louis (July 1991). Garbade, Kenneth D. “ The Effect of Interdealer Brokerage on the Transactional Characteristics of Dealer Markets,” Journal o f Business (July 1978), pp. 477-98. Garm an, M ark B. “ Market M icrostructure,” Journal o f Finan cia l Econom ics (June 1976), pp. 257-75. Glassm an, Debra. “ Exchange Rate Risk and Transaction Costs: Evidence from Bid — Ask Spreads,” Jornal o f Inter national M oney and Finance (Decem ber 1987), pp. 479-490. Glosten, Lawrence R., and Paul R. M ilgrom . “ Bid, Ask and Transaction Prices in a Specialist M arket with Heterogene ously Inform ed Traders,” Journal of Financial Economics (March 1985), pp. 71-100. Hahn, Frank H. Equilibrium a nd M acroeconom ics, (Basil Blackwell, O xford, 1984). Ham ilton, Jam es L. “ Com petition, Scale Econom ies, and Transaction Cost in the Stock Market,” Journal o f Financial and Quantitative Analysis (Decem ber 1976), pp. 779-802. Harsanyi, John C. Papers in Game Theory, (D. Reidel, Dor drecht, 1982). Ho, Thom as S. Y., and Richard G. Macris. “ Dealer Market Structure and Perform ance,” in Yakov Am ihud, T hom as S. Y. Ho, and Robert A. Schwartz, eds., M arket M aking and the Changing Structure o f the Securities Industry (Lexington Books, Lexington, 1985), pp. 41-66. Ho, Thom as S. Y., and Hans R. Stoll. “ On Dealer Markets under Com petition,” Journal o f Finance (May 1980), pp. 259-68. ________ . “ O ptim al Dealer Pricing under Transactions and Return Uncertainty," Journal of Financial Econom ics (March 1981), pp. 47-73. Keynes, John M. The General Theory of Employment, Interest a nd Money (M acM illan, London, 1936). Kubarych, Roger M. Foreign Exchange Markets in the United States (Federal Reserve Bank of New York, 1983). Leffler, G eorge L., and Loring C. Farwell. The Stock Market, (Ronald Press, New York, 1963). Logue, Dennis E. “ M arket-M aking and the Assessm ent of Market Efficiency,” Journal of Finance (March 1975), pp. 115-23. Lyons, Richard K. “ Private Beliefs and Inform ation Externali ties in the Foreign Exchange Market,” W orking Paper, C olum bia U niversity (Novem ber 1991). Mossberg, W alter S. “ M aking Book on the Buck,” Wall Street Journal, Septem ber 23, 1988. NOVEMBER/DECEMBER 1991 70 Negishi, Takashi. “ The Stability of a C om petitive Economy: A Survey Article,” Econom etrica (O ctober 1962), pp. 635-69. Sirri, Erik R. “ Bid/Ask Spread, Price, and Volume in a S pec ialist Market,” W orking Paper, University of California, Los Angeles (February 1989). Plott, C harles R. “ Industrial O rganization T heory And Exper im ental Econom ics,” Journal o f Econom ic Literature (De cem ber 1982), pp. 1485-527. Smidt, Seymour. “ W hich Road to an Efficient Stock Market,” Financial Analysts Journal (Septem ber/O ctober 1971), pp. 18-20, 64-69. ________ . “ W ill Econom ics Becom e an Experim ental S ci ence?” Southern Econom ic Journal (April 1991), pp. 901-19. Rai, Anoop K. “A Market M icro-Structure Investigation of Dealer Q uotes and Risk Prem ium in the Foreign Exchange Markets,” W orking Paper, (1991). Reinganurh, Marc R. “ M arket M icrostructure and Asset Pricing: An em pirical investigation of NYSE and NASDAQ securities,” Journal of Financial Econom ics (Novem ber/ Decem ber 1990), pp. 127-47. Riehl, Heinz, and Rita M. Rodriguez. Foreign Exchange and Money Markets: M anaging Foreign and Dom estic Currency Operations (McGraw-Hill, 1983). Ross, Sheldon M. Stochastic Processes, (John W iley & Sons, 1983). Ross, Stephen A. “ The Current Status of the Capital Asset Pricing M odel (CAPM),” Journal o f Finance (June 1978), pp. 885-901. ________ . “ The Interrelations of Finance and Econom ics: Theoretical Perspectives,” Am erican Econom ic Review (May 1987), pp. 29-34. Schreiber, Paul S., and Robert A. Schwartz. “ Efficient Price Discovery in a Securities Market: The O bjective of a Trad ing System,” in Yakov A m ihud, Thom as S. Y. Ho, and Robert A. Schwartz, eds., M arket M aking a nd the Changing Structure o f the Securities Industry (Lexington Books, Lex ington, 1985), pp. 19-39. ________. “ Price Discovery in Securities Markets,” Journal of Portfolio M anagem ent (Sum m er 1986), pp. 43-48. Schwartz, Robert A. Equity Markets: Structure, Trading, and Performance (Harper & Row, New York, 1988). FEDERAL RESERVE BANK OF ST. LOUIS Stigler, George J. “ Public Regulation of the Securities Mar kets,” Journal o f Business (April 1964), pp. 117-42. Stoll, Hans R. “ The Supply of Dealer Services in Securities Markets,” Journal o f Finance (Septem ber 1978), pp. 1133-51. ________. “Alternative Views of Market M aking,” in Yakov Am ihud, Thom as S. Y. Ho, and Robert A. Schwartz, eds., M arket M aking a nd the Changing Structure o f the Securities Industry (Lexington Books, Lexington, 1985), pp. 67-91. ________. “ Inferring the Com ponents of the Bid-Ask Spread: Theory and Em pirical Tests,” Journal o f Finance (March 1989), pp. 115-34. Tinic, Seha M. “ The Econom ics of Liquidity Services,” Quar terly Journal of Economics (February 1972), pp. 79-93. Tinic, Seha M., and Richard R. West. “ Com petition and the Pricing of Dealer Service in the O ver-the-Counter Stock Market,” Journal of Financial and Quantitative Analysis (June 1972), pp. 1707-27. Varian, Hal R. “ Differences of O pinion in Financial Markets,” in Courtenay C. Stone, ed., Financial Risk: Theory, Evi dence and Im plications (Kluwer Academ ic Publishers, Boston, 1989), pp. 3-37. Wei, Shang-Jin. “A nticipations of Foreign Exchange Volatility and Bid — Ask Spreads,” International F inance Discussion Paper No. 409, Board of G overnors of The Federal Reserve System (August 1991). Wolinsky, Asher. “ Inform ation Revelation in a M arket with Pairwise M eetings,” Econom etrica (January 1990), pp. 1-23. 72 FEDERAL RESERVE BANK OF ST. LOUIS REVIEW INDEX 1991 JANUARY/FEBRUARY JULY/AUGUST Ft. Alton Gilbert, “ Do Bank Holding Companies Act as ‘Sources of Strength’ for Their Bank Sub sidiaries?” Cletus C. Coughlin, “ U.S. Trade-Remedy Laws: Do They Facilitate or Hinder Trade?” Cletus C. Coughlin, “ A C onsum er’s Guide to Regional Economic M ultipliers” Keith M. Carlson, The Future of Social Security: An Update” James B. Bullard, Learning, Rational Expecta tions and Policy: A Summary of Recent Research” MARCH/APRIL John A. Tatom, “ Should Government Spending on Capital Goods Be Raised?” Alison Butler, “ A Case for a Bilateral Trade Deficit” Mark D. Flood, “ An Introduction to Complete Markets” David A. Dickey, Dennis W. Jansen and Daniel L. Thornton, “ A Primer on Cointegration with an Application to Money and Incom e” Jeffrey D. Karrenbrock, “ The Behavior of Retail Gasoline Prices: Symmetric Or Not?” M ichael T. Belongia, “ Monetary Policy and the Farm/Nonfarm Price Ratio: A Comparison of Ef fects in Alternative M odels” Michelle R. Garfinkel and Daniel L. Thornton, “ The M ultiplier Approach to the Money Supply Process: A Precautionary Note” Cletus C. Coughlin and Thomas B. Mandelbaum, “ Measuring State Exports: Is There a Better W ay?” SEPTEMBER/OCTOBER Keith M. Carlson, “ The U.S. Balance Sheet: What Is It and W hat Does It Tell Us?” Piyu Yue and Robert Fluri, “ Divisia Monetary Ser vices Indexes for Switzerland: Are They Useful for Monetary Targeting?” Steven Russell, “ The U.S. Currency System: A Historical Perspective” NOVEMBER/DECEMBER MAY/JUNE John A. Tatom, “ Public Capital and Private Sec tor Perform ance” R. Alton Gilbert, “ Supervision of Undercapitalized Banks: Is There a Case for Change?” James B. Bullard, “ The FOMC in 1990: Onset of Recession” Allan H. Meltzer, “ U.S. Policy in the Bretton Woods Era” FEDERAL RESERVE BANK OF ST. LOUIS John A. Tatom, “ The 1990 Oil Price Hike in Per spective” Mark D. Flood, “ M icrostructure Theory and the Foreign Exchange M arket” Piyu Yue, “ A M icroeconom ic Approach to Es tim ating Money Demand: The Asym ptotically Ideal M odel” Michelle R. Garfinkel and Daniel L. Thornton, “ Al ternative Measures of the Monetary Base: W hat Are the Differences and Are They Im portant?” Federal Reserve Bank of St. Louis Post Office Box 442 St. Louis, Missouri 63166 The Review is published six times p er year b y the Research and Public Information Department o f the Federal R eserve Bank o f St. Louis. Single-copy subscriptions are available to the public f r e e o f charge. Mail requests f o r subscriptions, back issues, o r address changes to: Research and Public Information Department, Federal R eserve Rank o f St. Louis, P.O. Rox 442, St. Louis, Missouri 63166. The views expressed are those o f the individual authors and do not necessarily reflect official positions o f the Federal R eserve Rank o f St. Louis o r the Federal R eserve System. Articles herein may be reprinted provided the source is credited. Please provide the Rank’s Research and Public Information Department with a copy o f reprinted material.