The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
R Vol. V 73, No. 4 July/A ugust 1991 3 U.S. T rad e-R em ed y Laws: Do T h ey Facilitate o r H ind er F ree T rad e? 19 T h e B ehavior of Retail G asoline Prices: S y m m etric o r Not? 30 M o n e ta ry Policy an d th e F arm /N o n fa rm Price Ratio: A C o m p ariso n of Effects in A ltern ativ e M odels 47 T h e M ultiplier A p p ro a ch to th e M oney Supply Process: A P re c a u tio n a ry Note 65 M e asu rin g State E xports: Is T h e re a B etter W ay? THE FEDERAL RESERVE RANK of ST.IjOIIS 1 Federal Reserve Bank of St. Louis R eview July/August 1991 In This Issue . . . Trade-remedy laws are intended to lessen the hardship on U.S. firms resulting from the actions and policies of foreign firms and govern ments. Theoretically, these laws produce a “fair” and "free” trading en vironment. In practice, however, the concept of fair trade is often used by special-interest groups to pursue their own agenda at the expense of the national interest. In the first article in this Review, "U.S. Trade-Remedy Laws: Do They Facilitate or Hinder Trade?” Cletus C. Coughlin analyzes the primary trade-remedy laws and concludes that their increasing use is hindering free trade. These trade-remedy laws, says the author, are producing a protectionist trading environment, which lowers economic well-being in the United States, rather than a fair and free environment, which would benefit the nation. *** In the second article of this Review, "The Behavior of Retail Gasoline Prices: Symmetric Or Not?” Jeffrey D. Karrenbrock examines the behavior of retail gasoline prices. As the author first points out, oil pro ducers and refiners are not the only entities on the supply side who in fluence gasoline prices. Because retailers also play a role in determining the retail price of gasoline, they could be equally responsible for any price anomalies that occur in the industry—including the popular belief that gasoline prices are increased more and reduced less in response to rises and falls in the underlying price of crude oil. Karrenbrock’s analysis finds little evidence to support this "pricegouging” hypothesis, however. He finds that retail gasoline prices res pond symmetrically to wholesale price increases and decreases in both the timing and the amount of price pass-through. Karrenbrock does note that the retail price adjustment lags are shorter for a wholesale price increase than they are for a decrease. *** During the early 1980s, while the Federal Reserve was pursuing an antiinflationary monetary policy, the U.S. farm sector also experienced one of its worst downturns since the 1920s. Many observers linked the two events, arguing that restrictive monetary policy hurt farm income because it caused farm product prices to decline more quickly than farm input prices; the result was a “cost-price squeeze” that, according to this argument, caused farm income to fall. In the third article of this issue, "Monetary Policy and the Farm/Non farm Price Ratio: A Comparison of Effects in Alternative Models," Michael T. Belongia reviews several models that have been used to link monetary policy to the relative price of farm products. The author then JULY/AUGUST 1991 2 attempts to synthesize the conflicting empirical evidence that has been brought to bear on each. By using consistent measures of prices and monetary actions and estimating each model over the same time period, he finds that monetary actions have weak and short-lived effects on the farm/nonfarm product price ratio. * * * In the fourth article in this issue, “The Multiplier Approach to the Money Supply Process: A Precautionary Note," Michelle R. Garfinkel and Daniel L. Thornton show that the money multiplier is not independent of monetary policy actions as is commonly assumed. They note that, for the multiplier to be independent of policy actions, movements in the ratio of currency to checkable deposits (the most important determinant of the multiplier) must be due to individuals simply adjusting their holdings of currency and checkable deposits. Most of the movement in this ratio, however, is due to policy-induced changes in checkable deposits, say the authors; moreover, the influence of policy actions on the multiplier has become more important since the implementation of the Monetary Control Act of 1980. Since then, the relationship between checkable deposits and reserves has become particularly close. *** Reliable information about the amount of production a state ultimately exports is essential for anyone interested in state economic develop ment. Unfortunately, no ideal measure of state export activity exists. In the final article in this issue, “Measuring State Exports: Is There a Better Way?” Cletus C. Coughlin and Thomas B. Mandelbaum contrast the two currently available state export series. The most prominent defi ciency of both is that they are based on the value of export shipments from firms within a state rather than on the value of economic activity related to exports within a state. To address this deficiency, Coughlin and Mandelbaum develop a third measure of state m anufactured ex ports. Comparisons between this and an existing measure of export ac tivity illuminate the shortcomings of the two available series and the ad vantages of the new series. *** FEDERAL RESERVE BANK OF ST. LOUIS 3 Cletus C. Coughlin Cletus C. Coughlin is a research officer at the Federal Reserve Bank of St. Louis. Thomas A. Pollmann provided research assistance. The author acknowledges the valuable suggestions of Seth Kaplan. U.S. Trade-Remedy Laws: Do They Facilitate or Hinder Free Trade? "Our fair trade laws are the bedrock on which free trade stands.” — Malcolm Baldrige A N INCREASINGLY contentious issue in in ternational trade pertains to so-called "trade- remedy laws.” These laws are intended to remedy hardships for U.S. firms resulting from the actions and policies of foreign firms and governments. Allegedly, these laws produce a “fair” and "free” trading environment. The possibility exists, however, that the concept of fair trade is simply a pretext used by interest groups to pursue their own interests at the ex pense of the national interest. This can result in a protectionist trading environment, which lowers economic well-being in the United States, rather than a fair and free one. This paper provides an introduction to U.S. trade-remedy laws. As background to under standing the justification and effects of these laws, the concepts of fair trade, free trade and protectionism are described. Next, an overview of the primary laws is provided. This is follow ed by evidence on the increasing use of traderemedy laws. Finally, evidence on the adminis tration and effects of these laws is examined to assess competing claims that these laws facilitate or hinder free trade. FAIR TRADE, FREE TRADE AND PROTECTIONISM To understand the controversy involving trade-remedy laws, one must become familiar with the basic concepts underlying the dispute. The most elusive concept is that of fair trade. On the surface, it is hard to argue against fair trade; however, there are different interpreta tions of this term and, thus, its application in concrete situations varies across individuals. Two interpretations of fair trade are related directly to differing impressions of reciprocity, which is a concept of fairness used in interna tional trade negotiations.1 Before negotiations to reduce trade barriers, two countries will gener ally have different levels and types of trade bar iSee Bhagwati and Irwin (1987) for a discussion of fair trade and its relation to U.S. trade policy. JULY/AUGUST 1991 4 riers. "First-difference” reciprocity means that a fair outcome is characterized by reductions in trade barriers such that the value received by each country stemming from the other country’s reduction in trade barriers is equal. Consequent ly, after the completion of negotiations, the two countries may still retain different patterns of trade barriers. On the other hand, "full” reciprocity requires that two countries allow identical access to their respective markets, which implies identical trade restrictions. Full reciprocity means that reciprocity of access must be met for individual sectors. This is known as a level playing field. Negotiations under the auspices of the General Agreement on Tariffs and Trade (GATT) use first-difference reciprocity as a procedural de vice. Nonetheless, the implicit goal of GATT is to generate a free trade environment, which im plies full reciprocity of market access. In such an environment, certain actions, such as govern ment attempts to directly influence the pattern of trade, are viewed as inappropriate and, thus, can be counteracted.2 Even though actions taken to open foreign markets and counteract inappropriate behavior by foreign firms and governments can be justi fied in the name of fair and free trade, these actions might not achieve their stated purpose. If they do not, then the result is higher levels of barriers with adverse consequences. Trade restrictions tend to reduce the competi tion faced by domestic producers; this protec tion is at the expense of domestic consumers. Empirical evidence shows clearly that the losses suffered by consumers exceed the gains reaped by domestic producers and government.3 Not only are there inefficiencies associated with ex cessive domestic production and restricted con sumption, but there are costs associated with both the enforcement of protectionist legislation and attempts to influence trade policy. Empiri cal research also shows that the adverse effects of protectionist policies persist because such 2Bhagwati (1988) characterizes GATT as a “ contractarian” institution that regulates inappropriate actions. Political pressures make it difficult to maintain a free trade stance unilaterally, so GATT attempts to prevent those actions that induce others to move away from free trade. 3See Coughlin et al. (1988) and Richardson (1989) for re cent surveys. 4Administered protection and procedural protectionism are two other terms for contingent protection. FEDERAL RESERVE BANK OF ST. LOUIS policies generate relatively lower growth rates than free trade policies. THE BASICS OF TRADE-REMEDY LAWS The United States employs various traderemedy laws to provide relief from imports for U.S. industries. These laws are frequently char acterized as "contingent protection” because the import relief is provided only under certain conditions.4 Table 1 lists the principal traderemedy laws and summarizes their key features. The Escape Clause The escape clause, contained in Section 201 of the Trade Act of 1974, allows temporary import barriers when rising imports can be shown to injure a domestic industry seriously.5 The legis lation requires that the increase in imports con stitutes "a substantial cause” of serious injury. While a substantial cause is not defined precise ly, a working definition is that the cause is im portant and no less important than any other cause of serious injury. Two primary justifications exist for escape clauses. The first justification relies on the im portance of an “economic adjustment” goal. Rapidly increasing imports can harm selected groups, especially import-competing domestic firms and their workers. Such firms must adjust to rising imports by enhancing productivity or by laying-off employees. Proponents of the es cape clause argue that the costs of this adjust ment can be reduced if the firm is provided temporary relief from imports. This argument, however, has some problems. Foremost is that there are numerous circum stances in which firms are forced to make ad justments. Changes in consumer demand, en ergy price shocks and governmental changes in spending, taxation and regulation necessitate ad justments. If rising imports justify governmental intervention, then it can be argued that these 5Prior to 1974, escape clause legislation required that the rising imports be due to a prior reduction of a trade bar rier. The elimination of this necessary relationship by the Trade Act of 1974 appears to make U.S. law inconsistent with GATT. See Jackson (1990) for a comparison of the legal nuances of U.S. law with Article XIX of GATT. 5 Table 1 Principal U.S. Trade Law Provisions1 Statute Focus Criteria for action Response Responsibility Section 201: Fair Trade (escape clause) Increasing imports Increasing imports are substantial cause of injury Duties, quotas, tariffrate quotas, orderly marketing arrangements, adjustment assistance President (ITC recommendation) Section 301: Unfair Trade Foreign practices violating a trade agreement or injurious to U.S. trade Unjustifiable, unreason able, or discriminatory practices, burdensome to U.S. commerce All appropriate and feasible action U.S. trade representative subject to direction by the president Section 701: Subsidized Imports Manufacturing, production or export subsidies Material injury or threat of material injury2 Duties ITC-lnjury determination ITA-Subsidy determination Section 731: Dumped Imports Imports sold below cost of production or below foreign market price Material injury or threat of material injury Duties ITC-lnjury determination ITA-Dumping determination 'O rigin of current provisions: Tariff Act of 1930 (Smoot-Hawley), as amended; Trade Act of 1974, as amended; Trade Agreements Act of 1979, as amended; Trade and Tariff Act of 1984; Omnibus Trade and Competitiveness Act of 1988. 2The material injury test is extended only to countries that fulfill certain conditions. SOURCE: Council of Economic Advisers (1988, p. 152), modified by author. other causes of adjustment costs should be miti gated as well. While there are cases other than rising imports that do lead to governmental in tervention, where should the line be drawn? Another facet of the adjustment argument focuses on the fact that rising imports provide benefits to many consumers and impose costs on relatively few firms and workers. An equity argument can therefore be made for shifting some of the burden of adjustment from the few who are harmed to the many who benefit through the tax on consumers imposed by im port restrictions. The second primary justification for escape clauses relates to this argument. A relatively small yet potentially well-organized group harmed 6Lande and VanGrasstek (1986) note that the escape clause allows member countries to impose trade restric tions to mitigate the perceived adverse effects of rising im ports, while remaining cognizant of their obligations to by rising imports could be a formidable force for import restrictions. From a national perspec tive, it is much better to provide temporary and limited protection for such a group not only to mitigate the burdens of adjustment, but also to reduce the political pressures for more perma nent import restrictions. Unfortunately, these temporary measures often become long-lived.6 Petitions for relief can be filed by any one of the following groups—individual firms, labor unions, trade associations or selected govern ment bodies (such as the United States Trade Representative, the House Ways and Means Committee and the Senate Finance Committee). The International Trade Commission (ITC) is a six member, appointed body that assesses injury fellow GATT members. This flexibility allows temporary departures from trade obligations without full-fledged repudiations. JULY/AUGUST 1991 6 after the filing of petitions. Significant declines in sales, production, profits, wages or employ ment are evidence of serious injury. Negative ITC decisions require a majority of the commissioners to reject the petition and ter minate the process. Affirmative ITC decisions require either a tie or the majority of the com missioners to accept the petition; they are for warded to the president. A recommendation as to the appropriate trade restriction and/or ad justment assistance to prevent or ameliorate the injury is included. The president, however, is not bound by the ITC’s injury finding or its sug gested relief. Nevertheless, the Trade Act of 1974 instructs the president to provide relief (which can take the various forms identified in table 1), unless such relief is deemed not to be in the national interest. U nfair T rad e L egislation The escape clause allows a nation to restrain imports regardless of w hether the imports have been assisted by “unfair” practices. Examined below are the three most prominent pieces of U.S. legislation that address the issue of offset ting the effects of unfair actions: 1) Section 701, which deals with governments subsidizing ex ports; 2) Section 731, which deals with dump ing, that is, with foreign firms selling their goods at lower prices in the United States than in their home markets; and 3) Section 301, which deals with violations of trade commitments and a wide range of other actions. S ection 701: C ountervailing D uty L egislation The legal purpose of countervailing duty legis lation is to offset government-provided benefits that assist the exports of foreign firms. These benefits include export subsidies, such as direct government payments, tax relief and subsidized loans to a nation’s exporters and low-interest loans to foreign buyers. By inducing additional foreign export activity as U.S. consumers substi tute these goods for similar domestically pro duced goods, these subsidies can injure importcompeting U.S. industries. Assuming certain provisions are met, U.S. trade law allows sub sidized exports to be counteracted with tariffs term ed countervailing duties. 7Lande and VanGrasstek (1986) point out that an injury test is used in countervailing duty cases only when the foreign country meets certain conditions. When an injury test is not required, the ITA has complete control. FEDERAL RESERVE BANK OF ST. LOUIS Even if one acknowledges that export subsi dies harm a domestic industry, it is possible to question the wisdom of countervailing duties. Many economists argue that if a foreign govern ment subsidizes exports, the importing coun try should accept the gift of cheaper goods. Resources no longer needed by the importcompeting U.S. industries can be employed pro ductively in other sectors of the economy. Countervailing duty legislation, however, fo cuses on the harm to these import-competing industries and ignores the benefits reaped by consumers and other producers. Bhagwati (1988), while acknowledging the economic validity of the preceding argument, argues that a free trade regime might depend on unfair trade legislation. Countries pursuing a free trade policy find it difficult to resist the demands for protection when the decline of an industry is due to the market-determined advan tages of foreign producers. If the decline is due to the use of export subsidies, demands for pro tection are heightened because issues of fairness are stressed. A free trade regime that does not counteract artificial advantages might find itself unable to defend and perpetuate its free trade stance. The administration of countervailing duty laws is the joint responsibility of the ITC and the International Trade Administration (ITA) of the Department of Commerce. The ITA deter mines the existence and magnitude of any sub sidy, negotiates agreements to offset any sub sidy, imposes duties, reviews the effectiveness of the remedy and determines when the rem edy is terminated. Concurrently, the ITC applies an injury test to determine whether subsidized exports have caused or will threaten material injury to a domestic industry or have retarded the establishment of a domestic industry.7 Material injury, as defined by the Tariff Act of 1930, is "harm which is not inconsequential, immaterial, or unimportant.” While this defini tion is far from clear, the law does require that the ITC incorporate volume, price and impact considerations in its determination of harm. In examining the volume of imports, the ITC is charged with determining w hether an increase in that volume, either absolute or relative to either U.S. consumption or production, is signif- 7 Figure 1 Statutory Timetable for Countervailing Duty Investigations (in days)_________ Total Days 205 235 270 300 SOURCE: United States International Trade Commission (1987). icant. With respect to the effect of imports on prices, the ITC looks for significant price under cutting by the imports relative to domestically produced goods and attempts to assess the re sulting price consequences. Finally, the impact on the domestic industry is assessed by examin ing changes in production, employment, market share, profits and wages. Countervailing duty cases begin when a peti tion is filed with the IT A and the ITC by either an interested party or the ITA itself. (A complete timetable for countervailing duty investigations is provided in figure 1.) If the ITA concludes that an investigation is warranted, then the ITC must reach a preliminary determination as to whether a "reasonable indication” of material in jury exists. A negative ITC determination term i nates the proceedings, while a positive deter mination leads to additional investigation. A preliminary affirmative ITA decision leads to the announcement of a preliminary estimate of the export subsidy and an order that import ers make a cash deposit or post a bond equal to the estimate of the subsidy for each entry. If the preliminary ITA decision is negative, no de posit or bond is posted; however, the ITA inves tigation continues until it reaches a final deci sion. If the final ITA decision is negative, then the case is terminated; otherwise, the ITA must determine the final subsidy margin. An affirmative final determination by the ITA leads to an ITC hearing in which all interested JULY/AUGUST 1991 8 parties participate. If the ITC finds no material injury, then the case ends. On the other hand, a finding of material injury by the ITC leads to an ITA order of countervailing duties against the imported merchandise. Such an order continues until revoked. Section 731: A nti-dum ping L egislation The legal purpose of anti-dumping legislation is to prevent two unfair practices: 1) price dis crimination in which foreign firms sell in the United States at prices lower than they charge in their home markets, and 2) export sales in the United States at prices below the average total cost of production.8 Both practices tend to lower the price of the good in the U.S. market causing U.S. consumers to purchase less of similar do mestically produced goods. This decrease in de mand harms the domestic industry by reducing profitability, sales, employment and other mea sures relative to a market without dumping. Although domestic consumers do benefit from the lower price, their interests are ignored by this legislation. Focusing on the first practice, if the IT A determines that the product in question is being sold in the United States at a price less than its foreign market value, the case is referred to the ITC. The ITC then investigates whether, as a result of the dumping, a domestic industry is in jured, likely to be injured or prevented from be ing established. The ITC’s assessment of material injury in dumping cases is the same as in countervailing duty cases. The ITA’s determination of the dumping margin, however, differs from the determination of the subsidy margin. The dump ing margin is simply the difference between the home market sales price and the export sales 8Boltuck (1991) identifies international price discrimination, promotional pricing, predatory pricing and hidden export subsidies as instances of price dumping. Note that for price dumping to be profitable, barriers must exist that prevent the imported good from being resold in the ex porter’s market at the higher home market price. Interna tional price discrimination is profitable when an exporter possesses more market power at home than in the United States (that is, demand in the firm ’s home market is less elastic than in the United States) and charges a higher price in its home market than in the United States. Promo tional pricing arises when an exporter induces consumers in a foreign market to try a product by introducing it at a low price. Predatory pricing is a rarely used strategy in which an exporter attempts to eliminate competitors by reducing the export price below its rival’s costs and below its own production costs. Once the competitors have ex FEDERAL RESERVE BANK OF ST. LOUIS price. While the concept of a dumping margin is simple, applying the concept to the real world is complicated.9 Details on calculating the dump ing margin, which can affect the consequences of this legislation, are highlighted later. The administrative procedure for anti dumping cases is virtually identical to that of countervail ing duty cases. The primary difference is that while an injury test is automatic in anti-dumping cases, it may not be required in certain counter vailing duty cases. In addition, the timetable for dumping investigations (provided in figure 2) is longer. Section 301 a n d th e A u th o rity to R etaliate Section 301 grants the United States Trade Representative (USTR) authority (subject to any directions from the president) to take all “appro priate and feasible action” to remove foreign trade barriers that hinder U.S. exports and to fight foreign subsidies that hinder U.S. exports to third-country markets.10 This legislation is primarily a Congressional response to dissatis faction with GATT’s ineffectiveness in resolving trade disputes.11 Formally, Section 301 allows the USTR to respond against any act, policy or practice of a foreign country that is determined to be: 1) inconsistent with the provisions of, or otherwise denies benefits to the United States under, any trade agreement; or 2) unjustifiable, unreasonable, or discriminatory and burdens or restricts U.S. commerce. Unreasonable is broadly defined in the Trade and Tariff Act of 1984 so that offending foreign restrictions are not limited to violations of trade agreements. The term includes, and goes be yond, any act, policy or practice that denies fair and equitable opportunities to begin and oper ate a business. Unjustifiable, as well as discrimited the market, the exporter raises the price. Finally, hid den export subsidies are classified as dumping either because there is no direct subsidization or the ITA is unable to demonstrate the existence of a subsidy. 9See Jackson (1990) for details on this complexity. 10The President (rather than the USTR) had this authority prior to the Omnibus Trade and Competitiveness Act of 1988. "A lth ou gh GATT is frequently involved in dispute-settlement proceedings, it has no authority to impose sanctions or en force its decisions. A GATT ruling favorable to the United States simply justifies unilateral U.S. action when the other party does not abide by the decision. In addition, Section 301 allows for the settlement of trade disputes with coun tries not belonging to GATT or when the issue is not covered by GATT. 9 Figure 2 Statutory Timetable for Anti-dumping Investigations (in days)_______________ Total Days 280 340 310 370 330 390 360 420 SOURCE: United States International Trade Commission (1987). JULY/AUGUST 1991 10 Figure 3 A Representative Case for the Section 301 Process Lesser of 12 months (18 months for trade agreement cases) OR 30 days after pro register notice hearings for public hearings hearings register notice to stop retaliation SOURCE: Grinols (1989). inatory, includes any act, policy or practice that denies either national or most-favored-nation treatment. In the context of the United States and a specific foreign country, national treat ment focuses on whether U.S. firms operating in that country are treated as favorably as the firms of the foreign country are treated in the United States. Most-favored-nation treatment refers to the best treatm ent accorded to firms from any other country operating in a specific foreign country. Even though all foreign firms (including U.S. firms) may be treated identically in a specific country, Section 301 could still be invoked if the treatm ent given were not as fav orable as the treatm ent given the foreign firms in the United States. Similar to other trade remedy proceedings, U.S. firms may formally petition to initiate Sec tion 301 proceedings or the USTR may initiate the case. A typical case in which the petition is filed with the USTR is illustrated in figure 3. The USTR’s role in Section 301 cases varies from the roles of the ITA and ITC in other trade-remedy cases. The USTR acts as both judge and advocate, while, relatively speaking, the ITA and ITC primarily judge on the basis of the objective merits as defined by the relevant 12See G rinds (1989). FEDERAL RESERVE BANK OF ST. LOUIS statutes. The USTR's task is much more subjec tive because it must also devise and pursue a negotiated settlement with a foreign government. If a negotiated settlement is not reached, the USTR may: 1) suspend, withdraw, or prevent the application of, or refrain from proclaiming, benefits of trade agreement concessions to carry out a trade agreement with the foreign party in volved; and 2) impose duties or other import restrictions on the products of, and fees or restrictions on the services of, such foreign par ty. This retaliation can be applied to all coun tries or to selected countries. Furthermore, the retaliation can be applied to goods and services other than those identified in the petition. In addition to dissatisfaction with the GATT dispute-settlement process, Congress has been unhappy with the operation of Section 301. Changes included in the Omnibus Trade and Competitiveness Act of 1988 generally reduce the president’s input into the process and en courage more frequent use of Section 301.12 A particular provision of the legislation known as Super 301 reflected Congress’ desire to “get tough” with our foreign rivals and pry open 11 Table 2 U.S. Trade-Remedy Petitions: Number and Percentage of Total Escape clause 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 4 2 6 1 5 6 3 3 2 2 0 1 6.5% 5.4 11.3 0.5 6.2 4.5 2.7 2.9 7.7 3.2 0 1.8 Anti-dumping Countervailing duty 16 24 15 63 47 73 65 70 14 40 23 43 37 11 22 145 22 52 38 26 5 13 9 8 25.8% 64.9 28.3 29.2 58.0 54.9 58.6 66.7 53.8 64.5 56.1 76.8 59.7% 29.7 41.5 67.1 27.2 39.1 34.2 24.8 19.2 21.0 22.0 14.3 Section 301 5 0 10 7 7 2 5 6 5 7 9 4 8.1% 0 18.9 3.2 8.6 1.5 4.5 5.7 19.2 11.3 22.0 7.1 Total petitions 62 37 53 216 81 133 111 105 26 62 41 56 SOURCE: Escape clause, anti-dumping and countervailing duty data for 1979-1988 are from Messerlin (1990). The remaining years of data are from the Office of the United States Trade Representative (1991). The data for Section 301, which includes peti tions as well as cases instigated by the President of the United States and the USTR, is based on a listing from the Office of the United States Trade Representative. their markets.13 Super 301 required the USTR to name (by May 30 of each year) those nations with the most restrictive barriers to U.S. ex ports and to identify the specific practices that most hinder U.S. exports. The listed countries faced retaliatory measures if no agreement on removing the trade barriers was reached within 12 to 18 months. THE FREQUENCY OF TRADEREMEDY PETITIONS During the 1980s, import-competing firms throughout the world increasingly used anti dumping actions rather than countervailing duty and escape clause actions.14 As shown in table 2, this change has occurred in the United States as well; however, this change is not as pro nounced in the United States as it is elsewhere. From 1983 onward, anti-dumping actions in the United States have exceeded 50 percent of the total number of trade-remedy petitions. In fact, the use of the escape clause mechanism has be 13See Bhagwati and Patrick (1990), especially Chapter 1, for an overview of the reasons for and the issues generated by Super 301. 14See Messerlin (1990) for details. come negligible. Since 1984, escape clause cases have generally totaled less than 5 percent of the petitions. The total number of actions show a substan tial decline since 1987. However, to argue that this decline indicates a sharp reduction in the demand for protection would be erroneous. Be tween 1982 and 1986, the total number of ac tions exceeded 100 cases in every year but one; however, approximately 200 cases involving steel products were initiated prior to the volun tary steel agreement of October 1984.15 In addi tion, the duties imposed under anti-dumping and countervailing duty actions can persist for some time.16 Similarly, other non-tariff barriers that were negotiated and imposed as a result of the pressure represented by anti-dumping and coun tervailing duty actions also persist for some time. Thus, many industries had their concerns resolved (at least temporarily) which resulted in a reduced number of new cases in the late 1980s. 16According to the Office of the United States Trade Representative (1991), 71 countervailing duty orders were in effect at year-end 1990. These 71 orders exceed the total number of countervailing duty petitions (61) between 1986 and 1990. ,5See Ahearn et al. (1990). JULY/AUGUST 1991 12 The trend since 1987 of declining actions, however, is not exhibited by Section 301 peti tions. On the other hand, there is no clear up ward trend either. Recent legislative changes in the Omnibus Trade and Competitiveness Act of 1988 suggest that this mechanism will assume increasing importance in the future. TRADE-REMEDY LAWS: THEIR EFFECTS AND THE ROLE OF ADMINISTRATIVE BIASES To assess the consequences of trade-remedy legislation, information on the frequency of these petitions is supplemented with details of their administration and outcome. The evidence pre sented below highlights biases in the administra tion of these laws and the proliferation of trade barriers resulting from these cases. As a result, it is highly unlikely that the legislation is facili tating free and fair trade. The Negligible E ffects o f Escape Clause Legislation The small number of escape clause petitions in recent years suggests that this trade-remedy legislation is having virtually no effect on the pattern of trade. The underlying criteria for a successful petition, plus the possibility of for eign retaliation unless other U.S. trade barriers are reduced, have deterred the use of escape clause petitions and induced industries to seek protection using other trade-remedy avenues.17 Despite the “requirement” that anti-dumping and countervailing duty actions can be invoked only to counteract the specific unfair trade practices of dumping and export subsidies, industries have increasingly resorted to these traderemedy laws rather than use the escape clause route. This apparent anomaly is explained when the administrative details of these less-than-fair value procedures are scrutinized. 17Article XIX of GATT allows trading partners affected adversely by an escape clause action to retaliate by withdrawing “ substantially equivalent concessions” affect ing the goods of the country invoking the escape clause. An alternative is to provide compensation to these GATT members by lowering trade barriers on their exports. See Hamilton and Whalley (1990) and Lande and VanGrasstek (1986) for additional details. 18See Finger and Murray (1990). 19Recent research has suggested another way that the ad ministration of this legislation could be biased. Moore (1990) found that ITC anti-dumping decisions were biased toward affirmative decisions when the complaining industry was FEDERAL RESERVE BANK OF ST. LOUIS Effects o f Countervailing Duty and Anti-dumping Legislation and Adm inistrative Biases Countervailing duty and anti-dumping actions are associated with rising trade barriers. For ex ample, of 774 countervailing duty and anti-dump ing cases completed in the United States be tween 1980 and 1988, 70 percent were resolved to restrict trade in some way.18 W hether the resulting duties, as well as the other negotiated restrictions on trade, should be viewed as pro tectionist, however, requires additional information. Countervailing and anti-dumping duties can be viewed as responses to actions taken by foreign governments and firms. For example, if an in vestigation determines that dumping is occurr ing, then the U.S. response to impose a duty (a dumping bond) equal to the dumping margin is automatic. In this case, the effect of the duty is to offset the injury to the domestic industry. If each instance of dumping is counteracted, then the net effect on trade would be zero in the sense that the level of trade would return to the same level prior to dumping. Many, how ever, do not feel that the actual workings of the legislation are quite so benign. A dm in istra tive B iases Bias has been found to enter the administra tion of less-than-fair value statutes through their interpretation by administrative agencies.19 The administering agencies have discretion that is sufficiently broad to allow for bias. For exam ple, Boltuck et al. (1991) conclude that the pro cedures used by the ITA in measuring subsidy rates and dumping margins are biased toward finding dumping and export subsidies.20 A foreign firm is found to be dumping when the price of their good in the U.S. market is located in the state of a senator on the Senate Finance Committee. 20See Kaplan (1991) for a demonstration of the protectionist bias in injury and causation determinations before the ITC. 13 below: 1) the price of the good in its home m ar ket; 2) the price of the good in a third market if no home market exists; or 3) its production cost. Such comparisons appear to be simple, but the description below suggests otherwise. Similar ly, the measurement of subsidy rates appears straightforward with the focus on devising ac counting rules to allocate government subsidies across the volume of exports. Closer inspection, however, indicates much complexity as well as the potential for bias in these calculations. The calculation of price dumping margins is subject to error at four different stages: 1) in identifying the home (domestic) market; 2) in adjustments to make domestic and export prod ucts comparable; 3) in adjustments to calculate the ex factory price (that is, the price as it leaves the factory) of domestic and export prod ucts; and 4) in comparing the ex factory prices of domestic and export products. Errors at the first and last of these stages are illustrated. The identification of the foreign firm’s home market tends to produce the highest possible fair value of the foreign good and, thus, the largest possible dumping margin. To illustrate, assume a firm occasionally charges a price be low its production cost in its home market. Such sales are made in the "normal course of busi ness” for any firm in a competitive industry that faces a demand for its product that varies randomly.21 Below-cost sales, that occur when demand falls below its average level, are bal anced by above-cost sales, that occur when de mand is above its average level, allowing the firm to earn a competitive rate of return. The 21Fred Smith of the Competitive Enterprise Institute, as quoted by Bovard (1987), notes, “ If the same antidumping laws applied to U.S. companies, every after-Christmas sale in the country would be banned.” 22To illustrate this particular bias, assume the foreign firm ’s sales occur at the same ex factory price in both the United States and its home market. Let the price be $10 in the first half of the investigation period and $5 in the second half. The firm sells five units both in the United States and at home in the first half of the period and 15 units in the United States and 10 at home in the second half. The average home price is $6.67 because one-third of the home sales occurred at $10 and two-thirds at $5. This average price is compared to the individual sales prices in the United States. Thus, each U.S. sale in the first half of the period would show a negative dumping margin (that is, dumping in the foreign firm ’s home market rather than in the United States) of $3.33 because the sales price of $10 exceeds the average home price of $6.67; each sale in the second half of the period would show a positive dumping margin of $1.67 because the sales price of $5 is less than ITA, however, excludes all below-cost sales from its calculation which raises the fair value of the product and creates an upward bias in the dumping margin. The comparison between the ex factory price of individual sales in the foreign firm’s export market (that is, the United States) to the average ex factory price in its domestic market can pro duce bias as well. For example, if prices were declining and exports increased relative to home sales during the period used to assess the exis tence of dumping by the foreign firm, then a positive dumping margin would be found. The low-priced export sales at the end of the period and the high-priced home m arket sales at the beginning of the period would generate a posi tive margin. Note that dumping would be found in this case even if home and export prices were identical for every sale made on the same day.22 In theory, the preceding bias should also oper ate in the other direction to generate negative margins. In practice, however, this does not oc cur because the ITC excludes the possibility of negative dumping, not only on average, but on each individual price comparison. Thus, all ex port sales below fair market value carry a posi tive dumping margin, while all export sales above fair market value carry a margin of zero23 Consequently, a positive dumping margin will be found even when all sales are made at the same price on the same day and the weights on each sale are identical.24 Thus, this procedure punishes foreign firms for not price discriminat ing because the only way to avoid dumping duties is to charge substantially more in the United States than in other markets. $6.67. One-fourth of the U.S. sales would have a negative dumping margin of $3.33, while three-fourths of the sales would have a positive dumping margin of $1.67. If this were the only bias in the ITA’s calculations, a positive dumping margin of $.42 would be calculated even though no actual dumping occurred. Additional biases, however, exacerbate the error further. 23ln the numerical example in the preceding footnote, the one-fourth of the U.S. sales with the negative dumping margin of $3.33 would be treated as having a zero dump ing margin. This increases the calculated dumping margin to $1.25 even though no actual dumping occurred. 24Boltuck et al. (1991) show that the bias in the margin equals approximately one standard deviation of the prices from the mean. Thus, if a company charges an identical price in the United States and other markets and its prices generally vary within 10 percent of the average price charged, the ITA will calculate a dumping margin of 10 percent. In other words, if the firm ’s average price is $6.00, the bias in the dumping margin is about $.60. JULY/AUGUST 1991 14 Similar bias exists in the ITA’s calculation of subsidies in countervailing duty cases, especially when the subsidy is not in the form of a direct per unit export subsidy. Biased accounting methods provide one route for finding inequity. For example, assume a foreign firm produces more than one product and that one of the pro ducts is allegedly subsidized. If the firm is found to be subsidized, the ITA allocates the en tire subsidy to the specific product that is the focus of the investigation irrespective of the degree, if any, to which the product is subsidized.25 Because of these sources of bias, virtually every investigation that proceeds to a formal determination finds in favor of positive dumping margins and export subsidies.26 Thus, less-thanfair-value cases, similar to escape clause cases, hinge on injury tests. Since the criteria for the injury test are less stringent for less-than-fairvalue than they are for escape clause cases, the infrequent usage of the escape clause is not sur prising.27 D uties a n d U ncertainty The reasons for the protectionist conse quences of less-than-fair-value legislation are not limited to biased administration. By law, any dumping margin or export subsidy greater than 0.5 percent justifies an affirmative determina tion. While a small duty is unlikely to have a large effect on competitiveness, the existence of any anti-dumping and countervailing duties creates costs of uncertainty that may have large effects for the exporter. This possibility is related to the fact that the importer (not the ex porter) assumes the risk of incurring higher duties than those originally paid. The bond initially posted on imports which are subject to these duties is only an estimate of the final duty. At the end of each year, the ITA allows interested parties the right to request a review of outstanding anti-dumping and counter 25See Boltuck et al. (1991) for additional examples. 26See Finger and Murray (1990). 27The fact that the president cannot set aside affirmative less-than-fair-value decisions but can set aside affirmative escape clause decisions, and that less-than-fair-value cases can be targeted against firms from specific coun tries while escape clause cases cannot, also explains the relative use of less-than-fair-value cases. 28This problem is not so pronounced in the case of dump ing. An exporter that continued to dump would be sharing FEDERAL RESERVE BANK OF ST. LOUIS vailing duty orders. Such a review will typically require three to four years before the final du ty rate is set. If there is no request for a review, then the estimated deposit rate is equal to the final duty rate. The uncertainty over deposit rates means that by importing under anti-dumping and counter vailing duty orders, importers are creating openended contingent liabilities for themselves. Find ings of underpayment require additional pay ments including interest to the government for the imports. To assess the risk of underpay ment, an importer must have substantial infor mation. For example, importers of goods subject to countervailing duties must be knowledgeable about the various industrial programs in the ex porting country and be able to assess the bias that exists in the calculation of duties.28 H arassm ent Another line of argument suggesting the pro tectionist consequences of less-than-fair-value legislation is known as the harassment thesis. Gregory (1979) noted a barrage of administra tive complaints or court suits (35) filed by U.S. electronic appliance and component manufac turers against Japanese competitors. He con cluded that even when these actions are not directly successful, they impose lengthy and costly delays on Japanese firms that indirectly produce a protectionist result. This harassment is not confined to Japanese firms nor are the consequences limited to in creasing the cost of penetrating the U.S. m ar ket. Firms from virtually all developed countries as well as many developing countries have been subjected to less-than-fair-value petitions.29 While these petitions directly increase the cost of pen etrating the U.S. market, the threat of a petition also increases the risk of exporting to the United States for actual and potential exporters. The ultimate result is that foreign firms will reduce their efforts to export to the United States. revenue with the U.S. Treasury that it could have retained. Nonetheless, an exporter more concerned with short-term profits from price discrimination than with a long-term rela tionship with the importer might still find it profitable to price discriminate. 29Finger and Murray (1990) note that before a firm files a less-than-fair-value petition, it frequently makes inquiries with law firms and holds informal discussions with the ITA, neither of which are kept secret. In effect, the period of harassment can begin before an official complaint is lodged. 15 N on -tariff B arriers As suggested by arguments discussing uncer tainty as well as harassment, less-than-fair-value legislation can have protectionist consequences apart from the actual duties resulting from specific cases. Another route to protectionism is that this legislation can result in the more fre quent use of non-tariff barriers. Finger (1981) pointed out that less-than-fair-value mechanisms can be used by a domestic industry to generate public support for protection. Until all the stan dard means of seeking protection have been ex hausted, it is unlikely that there will be strong political support for protection. Mechanisms such as less-than-fair-value cases must be util ized prior to gaining access to more political forms of protection. Many non-tariff barriers originated as less-than-fair-value cases. For ex ample, Finger and Murray (1990) found that nearly half of the less-than-fair-value cases in the 1980s were superseded by some form of negotiated export restraint. The Use o f Less-Than-Fair-Value L egislation in O ther C ountries A final line of argument suggesting that lessthan-fair-value cases lead to protectionism is that these cases induce other less-than-fair-value cases. As indicated by Messerlin (1990), this avenue of protection was not used by develop ing countries through 1985. The increased fre quency of these cases by developed countries could have spurred their increased use by de veloping countries in the late 1980s. In 1988 more than 20 percent of anti-dumping actions originated in developing countries. Some have argued that use of this mechanism by develop ing countries is even more capricious than use by developed countries because importers may be subject to anti-dumping duties without either due process or even formal notification.30 Thus, the use of less-than-fair-value legislation in the United States could backfire by generating addi tional inequities for U.S. producers—in this case, exporters—and in subjecting international trade to more barriers.31 S ection 301 a n d S u p er 301: C o n tro versia l C o n sequ en ces 30This point is made by Powell (1990) based on an interview with Robert McNeill, the executive vice chairman of the Emergency Committee for American Trade. petitor that is 48 percent British-owned. This new type of trade complaint will likely cause some supporters of this legislation to reconsider their positions as it becomes clear that foreign-owned firms can benefit at the expense of U.S. consumers. 31A related point is that the filing of anti-dumping petitions in the United States is not limited to domestically owned firms. As reported by Bradsher (1991), the American manufacturing subsidiary of a Japanese company recently asked the ITC to impose duties on the imports of a com- Although Super 301 has generated much con troversy since its passage, it has produced only a small number of offenders and practices. In 1989, for example, Brazil was cited for quantita tive restrictions involving her balance of pay ments; Japan was cited both for technical bar riers to trade hindering forest products and government procurement practices involving supercomputers and satellites; and India was cited for barriers limiting trade in foreign in surance services and for trade-related invest ment measures that imposed export performance requirements on foreign investors. It is note worthy that these priority practices were not necessarily those with the greatest export poten tial and that they were similar to those gener ally handled under Section 301. Nonetheless, these Super 301 actions generated protests from our major trading partners.32 Barfield (1990) criticizes the Super 301 process because it is ultimately controlled by the same political judgments the United States criticizes other countries for using in their trade policy decisions. For example, the naming of India and Brazil was in retaliation for their role as leaders of a group of developing countries that opposed U.S. goals in the Uruguay Round. The politicization charge can be levied against the Section 301 process in general. Powell (1990) argues that voluntary export restraints on steel in the 1980s were highly politicized. In the course of the 1984 presidential campaign, Re publican political leaders bowed to steel in dustry pressure for protection. Finding no other avenue available, the USTR threatened to file Section 301 cases unless numerous countries agreed to limit steel exports. Other ways also exist to manipulate Section 301 cases. For example, in November 1987, the USTR invited public comment to identify po tential Brazilian imports as targets for retalia tory tariffs in a computer piracy case. Represen tatives from various industries producing goods 32See Bhagwati and Patrick (1990), especially Part 3, for the reactions to Section 301 by various U.S. trading partners. JULY/AUGUST 1991 16 unrelated to computers, such as leather shoes and dinner dishes, made appeals for retaliatory tariffs of 100 percent to the USTR. Barfield (1990) also criticizes Super 301 be cause it violates the fundamental premises of GATT. GATT relies on negotiated reciprocal re ductions of trade barriers on a multilateral basis across many industries. Actions in which coun tries unilaterally define unfair practices and force bilateral negotiations under a retaliation threat are antithetical to GATT. Since GATT is the foundation for an orderly world trading sys tem, it is quite difficult to accept any argument suggesting that use of this legislation by the United States can facilitate free trade. It is more likely that other countries will develop their own versions of 301 legislation and that they will be used to counteract the United States. In such an environment, trade barriers will pro bably rise rather than decline. A more fundamental criticism of Section 301 (in general) and Super 301 (specifically) is that trade retaliation and retaliatory threats are inef fective in opening foreign markets. After study ing a large number of cases, Powell (1990) con cluded that this "crowbar” approach generally fails and that markets are opened because of domestic conditions rather than external ones. From 1975 through March 1990, only 13 of 79 Section 301 cases that were filed led to market openings.33 In many cases, countries have re sponded to retaliation by further closing their markets. Numerous reasons are offered to explain the ineffectiveness and shortcomings of this ap proach to open foreign markets. First, nation alism in the target country is inspired by retali ation; a coercive attempt by a foreign govern ment tends to unite the target country against the threat. Second, the target country reorients its economy toward alternative suppliers and markets. Firms and consumers in targeted coun tries can replace their transactions with the United States by selling to and purchasing from other countries. Third, the government's role in the target country generally expands. This in tervention to manage the changes induced by the retaliation involves trade-distorting policies, many of which are difficult to eliminate once they have been instituted. Fourth, the tougher the sanctions, the larger the costs incurred by the retaliating country in terms of higher con sumer and input prices.34 The preceding assessment, however, is not shared by everyone. Ahearn et al. (1990) note a congressional perception in recent years that Section 301 and Super 301 are working. The 1989 Super 301 complaints against Japan and Brazil were resolved in 1990. In addition, South Korea and Taiwan, both frequently mentioned as potential targets of Super 301, made advance concessions to avoid being named as priority countries.35 Nonetheless, even in situations where this legislation is generating results, it is far from clear that barriers to trade are actually being reduced. For example, recent U.S.-Japan discussions (known as the Structural Im pediments Initiative) were largely a Section 301 negotiation. It can be questioned whether the U.S.-Japan agreement to reduce structural im pediments to trade, such as Japanese “conces sions” to review tax policies that favor agricul ture over new construction and American "concessions” to reduce its budget deficit, will promote a freer trading environment or even reduce the U.S. bilateral trade deficit with Japan.36 33Powell (1990) found that some openings for U.S. exports to South Korea have led to new restrictions on imports from other countries, especially those from Japan. Thus, the South Korean market in not more open overall. While the actions by South Korea have served the interests of certain U.S. exporters, the actions reflect the fact that political clout rather than economic efficiency is determin ing the pattern of trade. 1982, President Reagan ordered higher steel tariffs and more restrictive import quotas as a result of a petition charging European steel subsidies. While U.S. steel pro ducers undoubtedly benefited, American manufacturers that required competitively-priced steel as an input were harmed. 34Numerous examples are available to suggest the harm. In 1988, President Reagan imposed 100 percent tariffs on Brazilian paper products, pharmaceuticals and consumer electronics as a result of a Section 301 petition alleging in adequate Brazilian protection of pharmaceutical patents. In FEDERAL RESERVE BANK OF ST. LOUIS CONCLUSION The alleged purpose of nearly all trade-remedy laws is to ensure that international competition is fair. Certain commercial practices, such as dumping and export subsidies, are viewed as unfair and, thus, should be counteracted. The elimination of these unfair practices will pro duce an economic environment in which the 35The 1989 Super 301 complaints involving two practices in India remain to be resolved. A review is to be conducted after the conclusion of the Uruguay Round. 36See Butler (1991) for details on these negotiations as well as for a general overview of U.S.-Japan trade. 17 success of firms and resource suppliers depend on their own performance in a competitive market rather than on their access to gov ernment subsidies or the use of questionable practices. Reality, however, bears little resemblance to the alleged purpose. Less-than-fair-value trade laws, especially anti-dumping laws, are the most frequently used trade-remedy laws and provide potential relief from all imports, whether they are traded fairly or unfairly. In fact, the opera tions of these laws are biased toward findings of dumping and export subsidies. Therefore, they have become standard devices to protect specific domestic producer interests at the ex pense of domestic consumer and other pro ducer interests. In addition, U.S. less-than-fairvalue trade laws explicitly instruct ad ministrators to protect specific domestic pro ducers, while ignoring the interests of domestic consumers and other domestic producers. The costs imposed by the increasing use of less-than-fair-value trade laws are not restricted to the consequences of the actual import duties. The threat of such cases leads to “voluntary" agreements to limit trade, agreements that harm potential importers. Furthermore, the uncer tainty associated with the actual duties collected functions as a type of non-tariff barrier. Finally, there is evidence that the use of trade-remedy laws tends to encourage protectionism in other countries. Despite some instances where Section 301 and Super 301 cases might have had positive ef fects in liberalizing foreign markets, there is substantial evidence suggesting that the crowbar approach generally fails. Domestic conditions rather than external pressures provide the pri mary motivation for the liberalization of m ar kets. Nonetheless, there is a high probability that this trade-remedy approach will be used more frequently in the future. If so, then other countries are likely to develop and use their own versions of 301 legislation to counteract the United States' actions. Similar to the world’s experience with less-than-fair-value laws, protec tionism is a likely consequence. Overall, the evidence is that trade-remedy laws hinder rather than facilitate free trade. U.S. fair trade laws can be more accurately characterized as the bedrock for protectionism rather than the bedrock for free trade. As such, trade-remedy laws need to be remedied by elim inating the bias toward protection of domestic producers. REFERENCES Ahearn, Raymond, Allan Mendelowitz, and Alfred Reifman. Congress and U.S. Trade Policy in the Nineties, No. 663, (National Bureau of Economic Research, October 5, 1990). Barfield, Claude E. “ The Grand Inquisitor on Trade,” Washington Post (May 1, 1990). Bhagwati, Jagdish N. Protectionism (MIT Press, 1988). Bhagwati, Jagdish N., and Douglas A. Irwin. “ The Return of the Reciprocitarians— U.S. Trade Policy Today,” World Economy (June 1987), pp. 109-30. Bhagwati, Jagdish N., and Hugh T. Patrick, eds. Aggressive Unilateralism: America’s 301 Trade Policy and the World Trading System (University of Michigan Press, 1990). Boltuck, Richard. ’’A ssessing the Effects on the Domestic Industry of Price Dumping,” in P.K.M. Tharakan, ed., Policy Implications of Antidumping Measures (North-Holland, 1991), pp. 99-141. Boltuck, Richard, Joseph F. Francois, and Seth Kaplan. “ The Economic Implications of the Current Administration of the U.S. Unfair Trade Laws,” in Down in the Dumps: Ad ministration of the Unfair Trade Laws, Robert Litan, ed., (Brookings, forthcoming). Bovard, James. “ U.S. Fair Trade Laws Are Anything But,” Wall Street Journal, June 3, 1987. Bradsher, Keith. ’’New Kind of U.S. Trade Complaint: By Foreign Companies,” New York Times, April 19, 1991. Butler, Alison. “ Bilateral Trade Balances and Economic Theory: The Case for a U.S.-Japan Trade Deficit,” this Review (March/April 1991), pp. 16-31. Council of Economic Advisers. Economic Report of the President (GPO, 1988). Coughlin, Cletus C., K. Alec Chrystal, and Geoffrey E. Wood. “ Protectionist Trade Policies: A Survey of Theory, Evidence and Rationale,” this Review (January/February 1988), pp. 12-29. Finger, J. Michael. “ The Industry-Country Incidence of ‘Less than Fair Value’ Cases in U.S. Import Trade,” Quarterly Review of Economics and Business (Summer 1981), pp. 260-79. Finger, J. Michael and Tracy Murray. Policing Unfair Imports: The U.S. Example, WPS 401, (The World Bank, March 1990). Gregory, Gene. “ The Profits of Harassment,” Far Eastern Economic Review (October 26, 1979), pp. 74-79. Grinols, Earl L. “ Procedural Protectionism: The American Trade Bill and the New Interventionist Mode,” Weltwirtschaftliches Archiv (Heft 3, 1989), pp. 501-21. Hamilton, Colleen, and John Whalley. “ Safeguards,” Jeffrey J. Schott, ed., in Completing the Uruguay Round: A Results-Oriented Approach to the GATT Trade Negotiations (Institute for International Economics, September 1990), pp. 79-92. Jackson, John H. The World Trading System (MIT Press, 1990). Kaplan, Seth. “ Injury and Causation in USITC Antidumping Determinations: Five Recent Approaches,” in P.K.M. Tharakan, ed., Policy Implications of Antidumping Measures (North-Holland, 1991), pp. 143-73. JULY/AUGUST 1991 18 Lande, Stephen L., and Craig VanGrasstek. The Trade and Tariff Act of 1984: Trade Policy in the Reagan Administration (Lexington Books, 1986). Messerlin, Patrick A. “Antidumping,” in Jeffrey J. Schott, ed., Completing the Uruguay Round: A Results-Oriented Ap proach to the GATT Trade Negotiations (Institute for Interna tional Economics, September 1990), pp. 108-29. Moore, Michael O. “ Rules or Politics?: An Empirical Analysis of ITC Antidumping Decisions,” Working Paper (George Washington University, 1990). FEDERAL RESERVE BANK OF ST. LOUIS Office of the United States Trade Representative. 1991 Trade Policy Agenda and 1990 Annual Report (GPO, 1991). Powell, Jim. “ Why Trade Retaliation Closes Markets and Impoverishes People,” Policy Analysis (Cato Institute, November 30, 1990). Richardson, J. David. “ Empirical Research on Trade Liberalization with Imperfect Competition,” OECD Economic Studies (Spring 1989), pp. 7-50. U.S. International Trade Commission. Annual Report 1986 (GPO, 1987). 19 Jeffrey D. K arrenbrock Jeffrey D. Karrenbrock was an economist at the Federal Reserve Bank of St. Louis when this paper was written. David H. Kelly provided research assistance. The Behavior of Retail Gasoline Prices: Symmetric or Not? S in c e DEREGULATION in the early 1980s, crude oil prices have been allowed to move freely with market conditions. Because of oil supply shocks and seasonal movements in gaso line demand, retail gasoline prices often fluctuate more widely than consumer prices in general. Some analysts and politicians have criticized these retail gasoline price movements, alleging that they do not respond symmetrically to price changes at earlier stages of the marketing chain. In particular, they believe that retail gasoline prices do not reflect decreases in oil and whole sale gasoline prices as rapidly and fully as they do price increases. The shaded insert on page 20 contains comments typical of this line of criticism. The perceived asymmetry in retail gasoline price movements is of special concern to consumers who believe that they are being “gouged" by the oil industry. Much of the perception of possible asymmetry focuses on the relationship between the price of oil and the retail price of gasoline. This suggests that oil producers or refineries are principally responsible for the asymmetry. In fact, a survey undertaken by the American Petroleum Institute concluded that 80 percent of Americans be lieved that oil companies artificially raised the price of gasoline after Iraq's invasion of Kuwait on August 2, 1990.1 This statistic suggests that many Americans believe retail gasoline stations are owned and operated by the oil refiners. In some cases this is true, but much of the gaso line sold at the retail level is sold through out lets that are not owned by the oil producers and refiners. The fact that many retail outlets are “independent" suggests that they have some autonomy in setting the retail price. The role these retailers play in the perceived asymmetry is largely ignored, even though they are as much a possible source of such an asymmetry as are the oil producers and refiners.2 This arti cle analyzes the role that retailers may play in the perceived asymmetric movement of retail gasoline prices. Specifically, we test whether 1McKenzie (1991). 2An article by Solomon (1990), however, does point out the potential role of retail outlets. See “ Gasoline Prices Resist Crude Behavior.” JULY/AUGUST 1991 20 What Goes Up Need Not Come Down? “Those who are doing the gouging will hear from the president.” —Treasury Secretary Nicholas Brady. The Wall Street Journal, (Shribman and McQueen) August 9, 1990. “Retail (gasoline) prices go up much faster than they come down.” —a spokesman for the Automobile Association of America. The Wall Street Journal, (Solomon) August 9, 1990. “Pump prices are fast to respond to rising prices but slower to fall when crude prices fall.” —Antonio Szabo, oil consultant with Bonner & Moore. The Wall Street Journal, (Business Bulletin) August 3, 1989. wholesale gasoline price increases are passed along to the retail customer more fully and rapidly than are wholesale gasoline price decreases. GASOLINE D ISTRIBUTIO N, PR IC IN G A N D M ARGINS The purchase of gasoline at the retail pump is the end of a long and complicated marketing chain. A simplified illustration of how oil, after undergoing refining, reaches the consumer as gasoline is shown in figure 1. From the oil fields, oil is moved to the refineries either by tanker, pipeline, or a combination of the two. The re finery receiving the oil may be owned by the company that produced the oil or may be in dependent. On January 1, 1990, 205 U.S. refin eries, owned by over 100 companies, were in operation. At the refinery, oil is distilled into a variety of products including gasoline, home heating oil, diesel oil, jet fuels, asphalt, kerosene and lubri cants. One barrel of oil (42 U.S. gallons) yields about 43 percent gasoline.3 Gasoline is trans ported from the refinery by truck, pipeline, tanker or barge. Some is moved directly from the refinery to retail outlets; some is moved from the refinery to terminal storage areas closer to final consumption. From these storage 3See Anderson (1984, p. 216). FEDERAL RESERVE BANK OF ST. LOUIS "Whenever oil prices fall, there is always this stickiness in gasoline prices on the way down. You never see this stickiness on the way up.” —Ed Rothschild, energy expert at Citizen Ac tion. New York Times, (Wald) July 2, 1990. “When crude prices go up, product prices tend to rise with crude prices. But when crude prices go down, product prices tend to lag— they go down slowly.” —John Hilton, oil in dustry analyst for Argus Research Corp. St. Louis Post-Dispatch, (Crudele) June 19, 1990. areas, the gasoline is generally moved to the point of final sale by truck. Once the gasoline reaches its final destination before purchase, it is usually stored in large underground tanks. Refiners may sell gasoline directly to "end users” such as large trucking firms, industrial m anufacturers and utilities. They may also sell directly to retail gasoline outlets. Retail gasoline stations owned by the refining company are classified as "end users.” Retail gasoline stations not owned by refining companies are known as “independents.” As figure 1 shows, sales to end users accounted for about 19 percent of refin ers’ gasoline sales, by volume, in 1988, with 17 percent of the sales to company outlets and 2 percent to other end users. The other 81 percent of refiners’ gasoline sales are made to either “jobbers” or indepen dent retail outlets. Jobbers purchase gasoline from the refiners which they in turn sell and distribute to retail stations and large users. Gas oline sales made by refiners to the non-companyowned retail outlets and to jobbers are referred to as “sales for resale.” Several different entities are involved in the pricing of gasoline as it is moved from the oil field to the retail gasoline outlet. When oil is sold to the refinery, the price for this transac tion is called the producer price. The price charged for gasoline by the refiner or jobber to 21 Figure 1 Oil and Gasoline Distribution Channels 1. Based, in part, on information provided in Dougher and Jones (1990), p. 7. the retail gasoline station is called the wholesale price.4 The price the gasoline station charges the consumer is called the retail price. The dif ferences between prices at various levels in the marketing chain are called "margins.” The dif ference between the retail and wholesale price is called the wholesale-retail margin. The dif ference between the wholesale price and pro ducer price is called the producer-wholesale margin. The overall difference is called the producer-retail margin. 4The price that the jobber pays the refiner is included in the “ sales for resale” price series used in this study. JULY/AUGUST 1991 22 DEFINITIONS OF ASYM M ETRIC GASOLINE PRICE MOVEMENTS Retail price movements are defined as asym metric if an increase in the wholesale price af fects the retail price differently than an equal sized decrease. Three types of asymmetry are defined. The first deals with the length of time in which a wholesale price change works its way through to the retail level. For example, is an increase in the wholesale price passed along more quickly to the retail level than an equal sized wholesale price decrease? The second type of asymmetry deals with the amount of a wholesale price change that passes through to the consumer. For example, does a 10-cent increase in the wholesale price lead to a 7.5-cent increase in the retail price, while a 10cent decrease in the wholesale price leads to only a 5-cent decline in retail price? The third type of asymmetry is a combination of time and amount. The pattern of retail price response may differ for wholesale price increases and decreases. Although the retail price may ad just to a wholesale price increase and decrease by an equal total amount and length of time, the amount of adjustment in each period may not be equal for price increases and decreases. For example, in cases where the wholesale price increases and decreases 10 cents per gallon, retail prices may require two months to completely respond to both wholesale price changes. Assume the retail price increases and decreases 9 cents per gallon in response to the wholesale price increase and decrease, respec tively. In such a situation, symmetry exists with respect to both the timing and amount of retail price movements. The pattern of the retail price response might be to increase (decrease) 7 cents in the initial month and increase (decrease) 2 cents in the month following a wholesale price increase (decrease). This pattern is symmetric. However, the pattern could be such that the retail price rises 7 cents and 2 cents in the first two months for a wholesale price increase, while the retail price falls only 3 cents in the initial month and 6 cents in the second month in response to a wholesale price decrease. This pattern is not symmetric. 5Pick et al (1990), Kinnucan and Forker (1987), Ward (1982), Heien (1980) and Hahn (1990) all find asymmetry in the agricultural markets. Domowitz et al (1988), Bils (1987) and Morrison (1988) find asymmetric markups in the manufacturing sector. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Since producers, wholesalers and retailers all play a role in the determination of the retail price of gasoline, the perceived asymmetric price movements in the industry could be oc curring between the producer and wholesale level or the wholesale and retail level. As noted earlier, many discussions of the perceived price asymmetry in the gasoline industry focus on the producer-retail price margin. Such a focus on the producer-retail margin tends to mask the role that retailers play in determining the producer-retail margin. Indeed, the perceived asymmetry may as readily be due to retailer behavior. In this case, simply observing the producer-retail price margin would not allow us to determine who is responsible for any asymmetry. Price-movement asymmetry has been found to exist in several commodity markets, including oranges, lemons, dairy products, some fresh vegetables, pork and beef. In addition, the markup of price over cost in durable and non durable manufacturing has been found to vary over the business cycle.5 Thus, a finding that price movement asymmetry exists in the retail gasoline market would not be unique. Many of the works cited above indicate the importance of industry concentration as a factor in explain ing the existence of asymmetry in these m ar kets. Kinnucan and Forker (1987) note that "because of industry concentration . . ., it is commonly asserted that middlemen use market power to employ pricing strategies which result in complete and rapid pass-through of cost in creases but slower and less complete transmis sion of cost savings.” GASOLINE PRICES AN D CO NSUM PTIO N The U.S. average retail and wholesale prices of gasoline are shown in figure 2 for the period examined (January 1983 to December 1990). Several intervals of relatively large and rapid wholesale price changes are shown in the fig ure. In early 1986, following the collapse of oil prices, wholesale gasoline prices dropped sharp ly. In the spring of 1989, gasoline prices rose sharply due "in part because of the temporary closing of the port of Valdez, Alaska, at the ter minus of the Trans-Alaska pipeline, after the 23 Figure 2 U.S. Average Retail and Wholesale Gasoline Prices1 Cents per Gallon 160 Cents per Gallon 160 0 1983 84 85 86 1Retail prices include federal and state tax. Exxon Valdez oil spill in March.”6 The jump and subsequent decline in prices in late 1990 are associated with an OPEC oil price increase prompted by Iraq in late July 1990, the subse quent Iraqi invasion of Kuwait and the world embargo of Iraq-Kuwait oil. In all instances, the retail price appears to parallel the wholesale price quite closely. A more detailed and system atic analysis is necessary to determine if there is indeed a symmetric response in retail prices to a wholesale price increase and decrease. Al though not shown in figure 2, the wholesale and retail prices of different grades of gasoline (pre mium, unleaded regular and leaded regular) also exhibit similar parallel movements with wholesale price changes. Since the analysis below examines asymmetry for different gasoline grades, it is useful to note the relative importance of these fuels. The mix of different grades of gasoline has changed sub stantially during the last 30 years. Prior to 1975, leaded gasoline accounted for over 50 percent of all motor gasoline fuel sales. Leaded gaso line's market share began to decline, however, after the enactment of environmental laws that required automobiles to burn unleaded gasoline and refiners to reduce the lead content of their 6See Wald (1990). JULY/AUGUST 1991 24 gasoline. Today, leaded gasoline accounts for only about 17 percent of total motor gasoline consumption, while unleaded regular and pre mium gasoline account for 59 percent and 24 percent, respectively.7 TESTING THE W HOLESALER ETAIL M AR GIN FO R SYM M ETRY The hypothesis considered is that movement in the wholesale-retail margin in the gasoline market is symmetric. We test to see if decreases in wholesale gasoline prices are passed along to consumers as rapidly and as fully as are whole sale gasoline price increases. We test only for symmetry in the wholesale-retail price margin because the model used for this test may be best suited for this margin. The model assumes a markup method is used to set the retail price of gasoline.8 To test for symmetric movements in retail prices, we use a model in which the current retail gasoline price (Rt) is a function of the wholesale gasoline price (W,); both prices are measured in cents per gallon. This relationship is summarized as (1) R, = a0 + aiWt. The effect of a change in the wholesale price on the retail price is (2) R , - R t_! = a j I W . - W n ) . In order to examine how the affect of a wholesale price increase differs from that of a decrease, periods of wholesale price increases and decreases must be separated. Following an approach similar to that developed by Wolfram (1971), this segmentation can be achieved using the model 7Based on volumes of first sales of motor gasoline in the Petroleum Marketing Annual [U.S. Department of Energy, (1988)], p. 216. 8This approach seems to more accurately represent the pricing behavior of retail outlets than oil refiners. Refiners with several oil products are perhaps more likely to employ a more sophisticated pricing mechanism than the retailer with a narrower range of oil products. One could make the argument, however, that retail outlets also have a multi product pricing function, especially if the station is associated with a convenience store. Dougher and Jones (1990) note suggestions that low margins on gasoline may be offset by higher margins on convenience foods. 9W olfram’s procedure uses the level of the dependent vari able, while we use the first difference of the dependent FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis (3) ARt = a,W It + a2WD, + e t, where ARt = Rt - Rt_j, WIt = Wt-W tM, if (W .-W ,.,) > 0, and = 0 otherwise, WDt = W .-W ,^, if (Wt-W t_j) < 0, and = 0 otherwise, et = a random error term .9 All WIt are positive or zero and all WD, are negative or zero. If retail prices respond sym metrically to wholesale price increases and de creases, then one would expect to find a! = a2. In order to allow for lags in adjustment time, a more general specification is where p and q are the specified num ber of lags for the wholesale price increases and decreases, respectively (p need not equal q). An intercept, a^, could be positive, negative or zero and need not be included on theoretical grounds. Follow ing Heien (1980) and Boyd and Brorsen (1988), however, we include it to avoid biasing the co efficient estimates if the intercept is not truly zero. This variable captures the average in fluence of all other factors besides raw material price changes that influence the retail price.10 Differences in the timing of price pass-through would be indicated by differences in the num ber of lags for increases (p) and decreases (q). The test of interest for the amount of pass through now becomes testing the equality i?0ai,i = ,?0a2„. In other words, is the cumulative variable. The model was also run for unleaded gasoline using the natural logs of all variables. The results are similar to those using the first-difference data. 10ln some studies, a variable to measure changes in other major marketing margin cost components, such as labor, transportation and packaging materials, has been included in equation 4. Preliminary estimates for this study that in cluded transportation wages and/or service station wages showed that neither variable was statistically significant. 25 effect of a wholesale price increase equivalent to that of a wholesale price decrease? If wholesale price changes are fully reflected in the retail price, we would expect to see Z0ai,i = 1 and Zfla2i = 1. Symmetry in the pattern of retail price response cannot be rejected if p equal q and all a j, = a2i. D A T A A N D ESTIM ATION PROCEDURE January 1983 through December 1990, a pe riod of relatively little government intervention in the gasoline market, was chosen as the period of analysis. Honeycutt (1985) notes that a ". . . factor that influenced gasoline marketing, beginning in August 1971 and continuing to Jan uary 1981, was extensive federal intervention in the marketplace.”11Furthermore, he notes that "statements by several major refiners that any changes in gasoline marketing would be phased in gradually suggest that not all important re sponses to decontrol had occurred by Septem ber 1981.”1Z In order to allow time for these "important responses” to have little or no effect on the results, the period studied here starts in January 1983. During the period analyzed, the number of months with price increases and decreases for retail and wholesale prices was roughly equal across all grades of gasoline. 11Honeycutt (1985), p. 108. 12lbid., p. 113. 13The unadjusted data are calculated by the U.S. Bureau of Labor Statistics and reported in the U.S. Energy Informa tion Agency’s Monthly Energy Review. These prices in clude all federal, state and local taxes paid at the time of sale. For the period 1978 forward, prices were collected from a sample of service stations in 85 urban areas se lected to represent all urban consumers—about 80 percent of the total U.S. population. Service stations are selected initially, and on a replacement basis, in such a way that they represent the purchasing habits of the Consumer Price Index population. Service stations in the current sample include those providing all types of service (i.e., full, mini and self-serve). See Monthly Energy Review, February 1989, p. 106. Retail prices are collected at dif ferent stations during the month of estimation. 14Taxes were removed from the retail price using informa tion provided in the U.S. Department of Energy’s Petrol eum Marketing Monthly. Federal and state motor fuel taxes are reported by the agency about twice a year (generally those effective on January 1 and July 1). ,5Handling the tax rate changes in this manner could bias the results because tax rate changes that occur between reported tax rate changes are not accounted for until the next reporting month. The retail prices used are tax-adjusted U.S. City Average Retail Prices of Motor Gasoline.13 The prices used were reduced by the sum of the federal gasoline tax and a simple average of the 50 states’ gasoline tax.14 No attempt was made to interpolate tax rates between months where tax rates were actually observed. The most current reported tax rates were used until new tax data became available.15 Wholesale prices are those from data referred to as "Sales for Resale.”16 These are sales of re fined petroleum products to purchasers who are "other-than-ultimate consumers.” This series does not include refined petroleum product sales made directly to end users, such as agri culture, industry and utility consumers or sales made by refiners to company-operated retail outlets. Wholesale prices are reported exclusive of taxes. RESULTS Equation 4 was estimated for premium, un leaded regular and leaded regular gasoline. Pre liminary estimates of lag lengths were selected using Akaike’s (1970) Final Prediction Error (FPE) criterion.17 The FPE procedure used to estimate the “best” lag length requires the user to specify a maximum lag length. For our data, the lag lengths selected by the FPE procedure were sensitive to alternative maximum lag lengths.18 Information Agency by firms responding to two separate surveys. The first survey, EIA-782A, “ Refiners'/Gas Plant Operators’ Monthly Petroleum Product Sales Report,” is sent to a census of about 200 refiners and gas plant operators. The second survey, EIA-782B, “ Reseller/Re tailers’ Monthly Petroleum Product Sales Report,” is sent to about 3,000 resellers and retailers. Some of the firms in this survey are replaced on an annual basis. In both surveys, firms are surveyed on a monthly basis and are asked to report prices on a monthly volume-weighted basis. 17Batten and Thornton (1984) note that the FPE criterion at tempts to balance the “ risk” due to bias when shorter lag lengths are selected against the “ risk” due to the increase in variance when longer lag lengths are chosen. Thornton and Batten (1985) point out that the FPE procedure gives relatively more importance to a lack of bias than efficiency. They argue that the procedure is asymptotically inefficient in that, on the average, it selects lags that are too long in large samples. 18Maximum lag lengths of six, nine and 12 months were specified in the FPE procedure. Results for the six-month and nine-month maximum were identical, although the 12month maximum model chose longer Wl lags (10 months) for premium and unleaded regular gasoline. Lag lengths suggested by the six- and nine-month maximum lag length models were used in estimating equation 4. 16See the U.S. Department of Energy's Petroleum Marketing Annual and the Petroleum Marketing Monthly. This price series is based on information provided to the Energy JULY/AUGUST 1991 26 Table 1 Symmetry Tests For Different Grades of Gasoline from January 1983December 1990 Amount Timing Type of gasoline Number of inonths lagged Wl WD Price parameter estimates Increase Decrease la ,! = 0 I a 2.i = 0 t-value for test of £ a ij= Za2 i t-value for test of £ a ij = i l a 2,i = 1 R2 D.W. Premium 1 1 .98* (18.28) .90* (16.98) .88 .34 1.79 .91 1.94 Unleaded regular 1 1 1.03* (18.64) .99* (17.64) .46 .53 .22 .92 2.15 Leaded regular 1 2 1.10* (19.58) 1.05* (16.93) .49 1.70 .80 .92 2.29 Note: Numbers in parentheses are the absolute values of the t-statistics. ’ Indicates statistical significance at the 5 percent level. After estimating the model with the lag lengths suggested by the FPE procedure, F-tests and t-tests were performed to see if any of the lags (incrementally or as a group) could be elimi nated as statistically insignificant. Only the sig nificant lags are reported below.19 Significant first-order autocorrelation was not present in any of the estimated equations. oline, however, wholesale price decreases affect the retail price for three months. Thus, the hy pothesis that the length of time in which retail prices completely respond to a wholesale price change is symmetric cannot be rejected for premium and unleaded regular gasoline but can be for leaded regular gasoline. T im ing S y m m e try Since the impact of the wholesale price change on the retail price is distributed over more than one month, the test for symmetry in the amount of pass-through examines w hether the total response to a wholesale price increase is equal to the total response topa wholesale price decrease. In other words, is I a , i = I a 2i? The results of this test are shown in table 1. For all grades of gasoline, the cumulative response of retail prices to a wholesale price increase is no different from that to a wholesale price de crease. In addition, the hypotheses that Z a1(i = 1 and Zoa2>i = 1 cannot be rejected for any grade of gasoline. This implies that whole sale price decreases are fully passed along to The ordinary least squares estimates for equa tion 4 are summarized in table 1. Lag lengths used for periods of wholesale price increases were the same across all grades of gasoline; lag lengths used for periods of wholesale price decreases were the same for premium and un leaded regular. Leaded regular gasoline had a slightly longer lag length for wholesale price decreases. These models suggest that wholesale price increases affect retail prices for two months (the initial month plus a lagged month). Similarly, wholesale price decreases affect the retail price of premium and unleaded regular gasoline for two months. For leaded regular gas 19The reported lag lengths are those suggested by the FPE criterion except for premium’s Wl (for which the FPE pro cedure suggested a lag length of three months), and unleaded’s WD (for which the FPE procedure suggested a lag length of two months). FEDERAL RESERVE BANK OF ST. LOUIS A m ou n t S y m m e try 1= 0 1 1= 0 * 27 Table 2 The Pattern of Retail Gasoline Price Response Equation 4 Parameter Estimates Wholesale increases a-1,0 Premium Unleaded regular Leaded regular Wholesale decreases aV 32,0 32,1 a22 ao .64* (14.39) .34* (7.62) .29* (6.43) .62* (11.57) -.1 6 (1.02) .68* (14.97) .35* (7.71) .30* (6.41) .69* (12.65) __ - .0 8 ( -50) .76* (17.21) .34* (7.69) .23* (5.13) .68* (12.85) .14* (3.05) -.0 7 (0.41) Note: Numbers in parentheses are the absolute values of the t-statistics. ‘ Indicates statistical significance at the 5 percent level. consumers, as are wholesale price increases. In short, the hypothesis that the amount of pass through in the retail gasoline market is sym metric cannot be rejected for the period of investigation. P a ttern S y m m e try Even though the time it takes retail prices to respond fully to wholesale price changes and the total amount that retail prices respond to wholesale price changes are symmetrical, there is a difference in the pattern of response to wholesale price increases and decreases. The coefficient estimates for equation 4 are graphi cally shown, by grade of gasoline, in figure 3. For wholesale price increases, the largest retail response occurs in the current month for all grades of gasoline. But, for wholesale price de clines, retail prices respond relatively little in the first month, and make their largest adjust ment in the month following the wholesale price decline. Using the premium gasoline model as an ex ample, a direct interpretation of the coefficients, as reported in table 2, is as follows: a 10-cent increase in the wholesale price leads to a 6.4- cent increase in the retail price during the in itial month, while a 10-cent wholesale price de cline leads to a 2.9-cent decline in the initial period. For premium gasoline, there is about a 3.5-cent per gallon difference in the amount that the retail price responds to a 10-cent whole sale price increase and decrease during the in itial month. For unleaded regular and leaded regular, the difference is about 3.8 cents and 5.3 cents per gallon for every 10 cent change in the wholesale price, respectively. Indeed, a test for equality of the a10 and a20 coefficients is re jected for all grades of gasoline, indicating asym metry in the amount of price response during the initial month of the wholesale price change. Wholesale gasoline price increases are passed along more fully in the initial month than are wholesale price decreases. The amount of the total retail adjustment occurring in the initial month ranges from 65 percent to 69 percent for wholesale price increases, and from 22 per cent to 32 percent for wholesale price decreases.20 During the second month, between 31 percent and 35 percent of the total retail adjustment oc curs for wholesale price increases, and between 20The initial month percent respons^for a wholesale price increase was calculated as [ai,0/( 2EJai,i|) 1*100. JULY/AUGUST 1991 28 Figure 3 Asymmetry in the Pattern of Retail Price Response (Estimated Coefficients for Equation 4) Retail Response (Coefficient) Premium Gasoline Retail Response (Coefficient) Response (Coefficient) Unleaded Regular Gasoline Response (Coefficient) 0.8 Retail Response (Coefficient) 0.8 0.8 Wholesale Price Increase Wholesale Price Decrease Leaded Regular Gasoline Wholesale Price Decrease Wholesale Price Increase FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis t-2 Retail Response (Coefficient) 0.8 29 65 percent and 70 percent for wholesale price decreases. The hypothesis that a^, = a21 is re jected for all grades of gasoline, indicating asym metry in the amount of price response during the month following a wholesale price change. For leaded gasoline, a third month is needed before the impact of a wholesale price decline is fully reflected in the retail price. CONCLUSION This paper has tested for symmetric retail gasoline price responses to changes in wholesale gasoline prices. The results show that the length of time in which a wholesale price increase is fully reflected in the retail gasoline price is the same as that of a wholesale price decrease for premium and unleaded regular gasoline. Whole sale gasoline price increases for leaded regular gasoline are passed along to the consumer more quickly than price decreases. Although the time in which retail prices fully respond to increases and decreases in wholesale prices is the same for both premium and unleaded gasoline, the pattern of retail price adjustment is such that consumers will experience the bulk of a whole sale price change sooner for price increases than they do for decreases. However, contrary to the popular belief that consumers do not benefit from wholesale gasoline price decreases, wholesale gasoline price decreases are even tually passed along to consumers as fully as are wholesale gasoline price increases. REFERENCES Domowitz, Ian, R. Glenn Hubbard, and Bruce C. Petersen. “ Market Structure and Cyclical Fluctuations in U.S. Manufacturing,” The Review of Economics and Statistics (February 1988), pp. 55-66. Dougher, Rayola S., and Russell O. Jones. Gasoline Distribution and Service Station Margins: An Assessment of EPA Assumptions and Implications for Methanol (American Petroleum Institute, September 1990). Fraser, R.W. “ Uncertainty and the Theory of Mark-up Pricing,” Bulletin of Economic Research (1985), pp. 55-64. Hahn, William F. “ Price Transmission Asymmetry in Pork and Beef Markets,” The Journal of Agricultural Economics Research (1990), pp. 21-30. Heien, Dale M. “ Markup Pricing in a Dynamic Model of the Food Industry,” American Journal of Agricultural Economics (1980), pp. 11-18. Honeycutt, T. Crawford. “ Competition in Controlled and Uncontrolled Gasoline Markets,” Contemporary Policy Issues (Spring 1985), pp. 105-18. Kinnucan, Henry W., and Olan D. Forker. “Asymmetry in Farm-Retail Price Transmission for Major Dairy Products,” American Journal of Agricultural Economics (1987), pp. 285-92. McKenzie, Richard B. “ Did 'Big O il’ Gouge Prices?” The Journal of Commerce, March 6, 1991, p. 8A. Morrison, Catherine J. “ Markups in U.S. and Japanese Manufacturing: A Short Run Econometric Analysis,” Work ing Paper Series No. 2799, National Bureau of Economic Research, Inc. (1988). Pick, Daniel H., Jeffrey D. Karrenbrock, and Hoy F. Carman. “ Price Asymmetry and Marketing Margin Behavior: An Example for California-Arizona Citrus,” Agribusiness, Vol. 6, No. 1, (1990), pp. 75-84. Shribman, David, and Michel McQueen. “ Office-Seekers Revive 1970s Campaign Strategy of Bashing Oil Com panies Over Spike in Prices,” The Wall Street Journal, August 9, 1990. Solomon, Caleb. “ Gasoline Prices Resist Crude Behavior,” The Wall Street Journal, May 2, 1990. ________“ Oil Companies Bend in Wake of Public Outcry," The Wall Street Journal, August 9, 1990. Akaike, Hirotugu. “ Statistical Predictor Identification,” Annals of the Institute of Statistical Mathematics (1970), pp. 203-17. Thornton, Daniel L., and Dallas S. Batten. “ Lag-Length Selection and Tests of Granger Causality Between Money and Income,” Journal of Money, Credit, and Banking (May 1985), pp. 164-78. Anderson, Robert O. Fundamentals of the Petroleum Industry (University of Oklahoma Press, 1984). U.S. Department of Energy, Energy Information Agency. Monthly Energy Review, Washington, D.C., various issues. Batten, Dallas S., and Daniel L. Thornton. “ How Robust Are the Policy Conclusions of the St. Louis Equation?: Some Further Evidence,” this Review (June/July 1984), pp. 26-32. ________Petroleum Marketing Annual, Washington D.C., various issues. ________Petroleum Marketing Monthly, Washington D.C., various issues. Bils, Mark. “ The Cyclical Behavior of Marginal Cost and Price,” The American Economic Review (December 1987), pp. 838-55. Wald, Matthew L. "Prices of Gasoline Fail to Reflect Drop in Cost of Crude Oil,” New York Times, July 2, 1990. Boyd, Milton S., and B. Wade Brorsen. “ Price Asymmetry in the U.S. Pork Marketing Channel,” North Central Journal of Agricultural Economics (1988), pp. 103-09. Ward, Ronald W. “Asymmetry in Retail, Wholesale, and Shipping Point Pricing for Fresh Vegetables,” American Journal of Agricultural Economics (1982), pp. 205-12. “ Business Bulletin,” The Wall Street Journal, August 3, 1989. Wolfram, Rudolph. “ Positivistic Measures of Aggregate Supply Elasticities: Some New Approaches—Some Critical Notes,” American Journal of Agricultural Economics (May 1971), pp. 356-59. Crudele, John. “ Gasoline Up, Oil Down in Price Paradox,” St. Louis Post-Dispatch, June 19, 1990. JULY/AUGUST 1991 30 Michael T. Belongia Michael T. Belongia is an assistant vice president at the Federal Reserve Bank of St. Louis. James A. Chalfant, Dennis Jansen, Thomas Grennes, Richard A. King, John S. Lapp and David Orden made many helpful comments on earlier drafts. Dawn Peterson, Lora Holman and Lynn Dietrich provided research assistance. Monetary Policy and the Farm/Nonfarm Price Ratio: A Com parison of Effects in Alternative Models S i n c e 1974, f o l lo w in g p u b lic a tio n of Schuh’s "The Macroeconomics of Agriculture/’ much research effort has been devoted to deter mining whether and how monetary policy af fects the farm sector. One of the more active areas of interest has been the question of wheth er changes in the money stock affect the farm/ nonfarm product relative price ratio. The reason for this particular interest, as described by Tweeten (1980), is that declines in the relative price ratio represent a "cost-price squeeze” for farmers; thus, he suggests, if contractionary monetary policy causes farm prices to adjust downward more quickly than farm input prices, farm income will decline as well. Penson and Gardner (1988), surveying the relevant literature, conclude that the agricultural sector has borne the brunt of adjustment costs whenever slower money growth has occurred. ’ Chambers (1984); Starleaf, Meyers and Womack (1985); Falk, Devadoss and Meyers (1986); Taylor and Spriggs (1989); and Tegene (1990) found similar results. Doo Bong Han, Jansen and Penson (1990) reaffirm the significance of this linkage by reporting that the conditional means and variances of agricultural prices are more closely related to FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis These conclusions received considerable atten tion in policy discussions during the mid-1980s when real farm incomes, exports and asset values were falling sharply. As those discussions inten sified, additional research reported that the quantitative effect of monetary actions on rela tive farm prices was not only large but persis tent, if not permanent. Among recent studies, Devadoss and Meyers (1987), for example, report that negative money supply shocks . . . "harm farmers because farm product prices decrease relatively more than nonfarm product prices” (p. 842). Many other studies found similar results.1 In sum, the notion that contractionary mone tary policy affects agricultural prices and in come differently than comparable measures in the nonfarm sector has become one of the the conditional means and variances of M1 than those of industrial prices. Orden (1986a), Lapp (1990), Gardner (1981) and Grennes and Lapp (1986), in contrast, did not find the relative price of farm products to be related to nominal macroeconomic variables. 31 stylized facts" of agricultural economics.2 Because neoclassical theory implies that changes in money growth have no real consequences in the long run, however, the large and sometimes permanent effects of monetary actions on agricultural prices reported in the literature seem to present an anomaly. Recent episodes, moreover, seem to run counter to the view that contractionary monetary policy selectively hurts farmers. First, real farm income rose during the late 1980s, a period some analysts would charac terize as one of substantial monetary contrac tion.3 Second, although the dollar's decline since early 1985 would help expand U.S. farm ex ports, all other things the same, the exchange rate depreciation has occurred at an odd time: when monetary policy has been contractionary and federal budget deficits have been expanding. Although the conventional wisdom links both factors to lower farm sector prices and income, this result is supposed to be transmitted through a rising value of the dollar.4 This article reviews the previous literature linking monetary actions to the relative price of farm products and attempts to reconcile the conflicting theoretical and empirical approaches that have been applied to this issue. Because previous studies derive their empirical models from a variety of generally noncomparable theo retical models, this paper highlights cases in which the direction or significance of a particu lar variable’s impact differs across models. By estimating each model with the same data and testing each model’s implications directly, we can better assess monetary policy’s effects on relative farm prices and the agricultural sector. A REVIEW OF THE LITER ATUR E Table 1 lists the important features of studies that examined the effects of monetary actions on the farm/nonfarm relative price ratio. The most common measure of farm prices used in these studies is the index of prices received by farmers. The relative price issue is typically in 2Chambers (1985). 3Between IV/1986 and IV/1990, for example, the 12-quarter moving average growth rate of M1 declined from 10.3 per cent to 3.0 percent. A “ trend” growth rate of M1 this low has not been seen in nearly three decades. “ See, for example, Belongia and Stone (1985). 5See, for example, Granger and Newbold (1974), Plosser and Schwert (1978) and Dickey and Fuller (1979, 1981). vestigated by dividing this index by another in dex of either the aggregate price level or the prices of some commodity bundle composed of nonfarm products; in some instances, farm and nonfarm nominal price indexes have been re gressed on a monetary measure individually to identify different speeds of adjustment and thereby infer the net impact of changes in mon ey growth on the selected relative price ratio. Ml has been used almost uniformly as the in dicator of monetary actions. Annual, quarterly and monthly data have been used to estimate the empirical relationship be tween Ml and the relative price measure. Most studies specify this relationship as one between the natural logarithms of the two series, some thing which, in view of the more recent litera ture on common trends in data and spurious regression relationships, may have given rise to significant associations where none actually existed.5 With the exceptions of Lapp (1990), and Grennes and Lapp (1986), these studies found Ml to have short-run effects on the farm/nonfarm relative price ratio. Unfortunately, in many cases, it is not easy to categorize the significance, magnitude or persistence of these effects. Where tested, the verdict seems about evenly split be tween those studies that find monetary actions to be neutral in their effect on the long-run relative price ratio and those that find the effects to be permanent. The only general conclusion that emerges from the studies summarized in table 1 is that the wide diversity among sample periods, relative price measures, variable specifi cations and results makes it difficult to tell whether and by how much monetary actions af fect the farm/nonfarm relative price ratio. A R EVIEW OF THE D A T A Figure 1 shows quarterly values for the an nualized percentage changes in the indexes of prices received and prices paid by farmers since 1976.6 These indexes are based on the bundle qualitative conclusions of this section. The plot starts in 1976 to avoid the price volatility associated both with OPEC and U.S. farm policies in the 1973-74 period. Moreover, the empirical work to follow begins after the system of flexible exchange rates was adopted and most of the one-time adjustments to new exchange rate levels—especially trade flows—are presumed to have taken place. of 6Use monthly data or producer price indexes for farm and industrial (nonfarm) commodities does not affect the JULY/AUGUST 1991 32 Table 1 A Summary of Results from Studies of the Monetary Policy-Relative Farm Price Question Author(s) Relative price measure Monetary policy indicator Sample period Data frequency Specification of variables M1 1976.05-1982.05 Monthly Logs Not tested Inflation rate 1930-83; 1930-53; 1954-70; 1971-83 Annual Percentage changes Not tested M1 1960.01-1985.12 Monthly Logs Long-run neutrality of money “ Im portant” Monetary Effects Chambers (1984) Starleaf, et al. (1985) Devadoss and Meyers (1987) CPI-food CPI-nonfood Prices received, Prices paid Prices received Non-neutral Index of Industrial Prooduct Prices Taylor and Spriggs (1989) Index of Canadian Farm Product Prices M1 (Canada) 1959.1-1984.4 Quarterly A Logs M1 had largest effect of domestic variables, but all foreign variables had larger effects. Tegene (1990) PPI farm output, PPI nonfarm output M1 1934-87 Annual Logs In the short run, a change in M1 affects farm prices more quickly than manufacturing prices. Index of Farm Prices M1 1960.1-1985.4 Quarterly ALogs M1 1951-81 Annual Logs; A logs Neutral M1 1951-85 Quarterly A Logs Neutral Han, et al. (1990) Conditional mean and variance of farm prices are more sensitive to changes in M1 than are the conditional mean and variance of industrial prices. “ Sm all” Monetary Effects Grennes and Lapp (1986) Lapp (1990) Prices received CPI Prices received PPI or CPI Im portance of Monetary Effects is Subject to Interpretation Orden (1986) Prices received GNP deflator M1; Interest rates 1960.1-1984.3 Quarterly Logs Monetary effects are small if represented by M1, but larger if represented by interest rates. Orden and Fackler (1989) Prices received GNP deflator M1; Interest rates 1975.1-1988.1 Quarterly Logs Small, but significant, effect over four quarter period; longrun neutrality. New Zealand farm output prices, Manufacturing prices New Zealand M1 1963.1-1987.1 Quarterly Logs Clear short-run effect over four quarters; long-run neutrality holds. Robertson and Orden (1990) of farm products that farmers produce and sell and commodities that farmers purchase as pro duction inputs, respectively.7 Quarterly values for the growth rate of Ml also are shown in this figure. Data for the individual series are summarized in table 2. As the figure and table show, farm product prices have been more vol 7The prices received measure is a weighted index of about 112 farm product prices; the prices paid measure is a weighted index of about 450 farm input prices. For more RESERVE BANK OF ST. LOUIS FEDERAL atile than farm input prices. A test of the equali ty of variances, for example, produces an Fstatistic of 7.62 against a 5 percent critical value of 1.53. Figure 2 shows changes in the ratio of the in dexes plotted against changes in M l growth. In detail, see Handbook #365, U.S. Department of Agriculture (1970). 33 Figure 1 Growth of M1, Prices Paid by Farmers, and Prices Received by Farmers Table 2 Descriptive Statistics for Farm and Nonfarm Prices, I/1976-IV/1990 (annualized first differences of logarithms, quarterly data)__________________________________________ Mean Prices received by farmers Prices paid by farmers Prices received/prices paid M1 Standard deviation Minimum Maximum 2.22% 4.74 -2 .5 2 7.03 14.30% 5.18 12.38 4.49 - 24.75% -4 .9 6 - 26.35 -4 .0 9 46.74% 15.92 32.18 16.85 JULY/AUGUST 1991 34 Figure 2 First Difference of the Ratio of Prices Received to Prices Paid and the Change in M1 Growth 20 20 16 16 12 12 -4 -4 -8 -8 -12 -12 1976 77 78 79 80 81 82 very simple terms, these series represent the logic in much of the literature that links mone tary policy to the relative price of farm pro ducts. For example, accelerations in money growth are thought to be associated with in creases in the farm/nonfarm product price ratio. Over this sample period, however, the simple correlation coefficient for these series, 0.13, is not significantly different from zero at the 5 percent significance level. Finally, it is interesting to abstract from the short-run volatility in these series and examine the data for longer-run trends. Since 1976, the average growth rate of prices received by farm ers has been about one-half that of farm input prices; as a consequence, the relative farm price ratio has fallen at an annual rate of more than 8See Meltzer (1990) for a thorough review of U.S. monetary policy since the mid-1960s, with special emphasis on its tendency to produce increasing rates of money growth. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 83 84 85 86 87 88 89 1990 2.5 percent. Conversely, Ml has grown over the sample period at an annual average rate of 7.03 percent. From a long-run perspective, the down ward trend in the relative price ratio is consis tent with what Tweeten has called a “cost-price squeeze” for farmers. The origin of this squeeze, however, does not seem to be related to the relatively expansionary long-run course of monetary policy.8 MONEY G R O W T H A N D R ELATIVE PRICES: ALT E R N A T IV E THEO RETICAL RESULTS The research attempting to link monetary ac tions to relative farm price changes has not questioned whether changes in money growth 35 Figure 3 Graphical Representation of a Barro-type Model MODEL I. Farm Commodities Market Non-farm Commodities Market Qf q nf Model Assumptions and Predicted Result:An unanticipated decrease in money growth causes the demands for both farm and nonfarm commodities to fall. Because the income elasticity of farm commodity demand is assumed to be lower than that for nonfarm products, the decrease in farm product demand is smaller. Assuming identical supply elasticities in the two markets, APF < APN and (PF /PN rises. F F) affect the farm/nonfarm price ratio; instead, the direction, size and persistence of this effect have been its primary focus. Because alternative theo retical models produce different empirical speci fications and, quite possibly, different results, some attempt must be made to distinguish among these alternatives. For guidance on these issues, the testable implications of three models used to investigate the money-relative price question are developed below. M odel 1: A Change in M on ey G row th as a S h o ck to A ggregate D em an d Equilibrium “Barro-type” models assume that anticipated changes in the money stock affect all nominal prices equi-proportionally and there fore leave relative prices unchanged.9 Relative prices are affected in these models only by an unexpected change in the money stock. In model #1, illustrated in figure 3, an unanticipated decline in the money stock produces a negative shock to aggregate demand as people find themselves with a shortage of real money bal ances and an excess supply of goods. Their col lective actions to restore equilibrium by reduced spending shifts aggregate demand to the left. This shift lowers output and income temporarily and the price level permanently. If supply elasticities in the farm and nonfarm sectors are identical, this demand shift will af fect relative prices only if the income elasticities 9See, for example, Barro (1976). JULY/AUGUST 1991 36 of demand for farm and nonfarm products dif fer. If the income elasticity for farm products is lower than that for nonfarm products, an unex pected decrease in the money stock would in crease temporarily the relative price of farm products.10 This interpretation of the model, therefore, predicts a response that is contrary to the story embedded in the "stylized facts” of agricultural economics. Because the direction of relative price change will vary with the par ticular assumptions about shifts in supply and demand across markets, the “sign” on this effect in a regression equation offers a direct way to test the implications of this one interpretation of the model. The predictions of this model, however, deny that monetary contractions are a source of longlasting harm to the farm sector. In this case, as in the other examples that follow, the real in come effect is a short-run phenomenon. When people realize that the real demand for in dividual products has not changed fundamental ly but that, instead, the monetary contraction caused a general decline in aggregate demand, the aggregate price level will fall to restore the original equilibrium and relative price ratio. Thus, because the decrease in the money stock is reflected only in a lower aggregate price level in the long run, the neutrality of monetary ac tions is preserved. In this model, as well as in model #2 that follows, whether the relative price ratio rises or falls and whether it returns to its original value in the long run are the model's testable hypotheses. As shown in table 1, however, the long-run neutrality proposition has not been tested in many previous studies or, when violated, often has not been discussed.11 M odel 2: R e la tive P rice C hanges C aused b y D iffere n t E la sticities o f S u p p ly A slightly different variant of the equilibrium Barro model, which predicts a relative price change in the opposite direction from the pre vious discussion, is based on different assump tions about the structure of the farm and nonfarm goods markets. In this model, illustrated in figure 4, the short-run elasticity of supply of farm products is argued to be less than that of ’ “ Historically, this assumption has been supported by the data with estimates of the income elasticity for food de mand near 0.2 and higher estimates for nonfood items; see King (1979) for a review of this literature. "E xceptions are Bessler (1984) and Robertson and Orden (1990). http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis nonfarm products because of differences in the production processes.12 With long lags between planting and breeding decisions and product marketings, the ability to adjust farm output in the short run is assumed to be limited. Other things the same, this characteristic of farm pro duction would cause the farm/nonfarm relative price ratio initially to fall in response to a negative aggregate demand shock. Again, the ex istence of long-run neutrality is a testable pro position and the length of the adjustment pro cess must be determined empirically. This model’s predictions, however, are consistent with the argument that the relative price of farm products will fall under a contractionary monetary policy. M odel 3: P rice S tick in ess an d “O ve rsh o o tin g ” So far, prices in both the farm output and in put markets were assumed to be flexible in response to changes in other variables that af fect them. Thus, changes in the relative price ratio depended on the relative magnitudes of shifts in supply and demand and the slopes of those curves; they were not influenced by dif ferent speeds of adjustment in the two markets. Another approach to this question has relied on some degree of price-stickiness in nonfarm prices to explain changes in the relative price ratio. By adapting the overshooting model from the exchange rate literature, as illustrated in figure 5, this analysis assumes that prices in the flexi ble price (farm) sector adjust to a monetary change more quickly than other prices in the fixed price (nonfarm) sector.13 So, for example, while long-term contracts prevent nonfarm prices from adjusting downward immediately in response to a monetary contraction (as they would in the Barro-type model), the auction market characteristics of the determination of farm prices force them to fall quickly and, con sequently, temporarily reduce the farm/nonfarm relative price ratio as well. Thus, again in the short run, a negative monetary change causes a temporary reduction in farm prices relative to nonfarm prices. 12See, for example, Starleaf (1982). 13See, for example, Frankel (1986) or Rausser (1985). 37 Figure 4 Graphical Representation of Differing Supply Elasticities Model As in Model I, an unanticipated decrease in money growth decreases the demands both for farm and nonfarm products; here, however, the decreases are assumed to be equal. Under these conditions, a lower elasticity of supply for farm products will cause APP > APN and (PF /PN ) will fall. Note that combining the F F results of Model I with Model n produces an ambiguous result because differences in the sizes of demand shifts may be large enough to cause an increase, a decrease, or no change in (PF F /PN ). Although this predicted direction of relative price change is the same as in model #2, the mechanics of price stickiness allows the possibility that fully anticipated monetary changes (as well as u n expected changes) can af fect the relative price ratio. Thus, testing for the significance of expected monetary changes on the relative price ratio provides a direct way to discriminate between the two models. Unfor tunately, the converse is not true: failing to find significant effects from anticipated monetary changes does not necessarily reject an over shooting type of model because its mechanics can be set in motion solely by monetary sur prises as in the previous cases. RECONCILING ALTE R N A TIV E THEORIES The foregoing discussion showed that farm prices are significantly more variable than non farm prices, and that the farm/nonfarm relative price ratio has declined persistently over time. Unfortunately, the implications of our three models differed considerably in terms of the ex p ected direction of change in the relative p rice of farm products to nonfarm products as well as the mechanism by which a monetary action influenced this ratio. To resolve this conundrum, each model was estimated using identical data sets to see which one’s implications are best supported by the results. These estimations are intended to provide evidence on three aspects of the possible mone tary influences on the farm/nonfarm relative price ratio. The first piece of evidence is the direction of relative price change; this will discriminate between model vs. models *2 and #3. The second piece of evidence is the statistical significance of the relationship, where the significance of a variable can be viewed as evidence for or against a particular model. The JULY/AUGUST 1991 38 Figure 5 The Overshooting Model1 Schedule EE shows all possible equilibrium relationships between PF and PN prior to the change in monetary F policy. Slower money growth, which is disinflationary, shifts EE to E'E' where farm (flex) prices (PF decrease ) but nonfarm (fix) prices (PN ) do not adjust; thus, PF F “ o v e rs h o o ts ” fro m p o in t A to p o in t B. A s fix -p ric e markets adjust, (P f/P Nf) gradually returns to the long-run equilibrium at point C. ’ A dopted from R. D ornbusch, Open Econom y M acroeconom ics, Fig. 11-8, p. 208. third piece of evidence is the magnitude of the impact that a given monetary change has on the relative price ratio; here, it is recognized that monetary effects may be statistically significant and yet still be quantitatively unimportant. VAR E stim ation As a first step in the investigation, a vector autoregressive (VAR) model was estimated. A VAR, which can be used to determine the FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis amount of variation in the relative price ratio that one might attribute to monetary shocks, is useful in gauging the strength of the hypothe sized relationship. It has the additional advan tage of not requiring the specification of any particular functional form among the variables included in the model. The VAR and other equations that follow were estimated with quar terly, seasonally adjusted data over the 1/1976IV/1990 sample interval. Thus, by way of in 39 Table 3 Single-Equation Results for the VAR Estimation, I/1976-IV/1990 Explanatory Variables Dependent variable Relative prices M1 Exchange rate Industrial production Relative price 0.206 (1.089) 0.031 (0.064) -0 .2 3 3 (1.735) M1 0.015 (0.257) 0.626 (4.192) Exchange rate 0.451 (1.874) Industrial production 0.131 (1.690) R1 DW -0 .1 4 7 (0.413) .12 1.98 -0 .0 2 0 (0.488) -0 .2 5 7 (2.340) .28 1.96 -0 .0 4 6 (0.075) 0.499 (2.929) -0 .1 3 7 (0.302) .09 2.04 0.344 (1.737) 0.003 (0.063) 0.515 (3.541) .29 2.01 NOTE: t-statistics in parentheses apply to sums of lag coefficients and apply to the null hypothesis that the sum is equal to zero. troduction to the more specific testing to follow, the VAR can offer some insights to the strength of the money-relative price link. The VAR model included four variables: the farm/nonfarm price ratio (as measured by the ratio of the indexes of prices received by farm ers to the producer price index), Ml, the index of industrial production and the real tradeweighted exchange rate. Variables other than Ml were included because observed changes in the relative price ratio may have other origins. For example, technological changes in the non farm sector (which would affect industrial prices) or export demand (which could have varying effects across the farm and nonfarm sectors) could affect the relative price ratio in isolation from monetary changes.14 While these other measures do not exhaust the list of in fluences on the relative price ratio, they do cap ture other influences affecting prices in the farm and nonfarm product markets so that the remaining variation can be explained by changes in Ml growth and the past history of the relative price ratio itself. These variables also were chosen because they have been used in previous work and our interest is in stressing comparability with other studies. All variables were specified as first differences of logarithms.15 Sums of lagged coefficients and t-statistics for these sums for the single equation estimation are reported in table 3. The results of interest indicate that Ml growth is not related signifi cantly to changes in the relative price of farm products.16 One possible explanation for this result is that flows from farm inventories, which were historically large over most of the sample 14Evaluating the impact of the industrial production measure here also will serve as an additional check on the over shooting model, which includes an output measure as an explanatory variable and predicts a positive relationship with the relative price ratio. of prices received by farmers divided by either the index of input prices paid by farmers or the all-item CPI. The possible effects of exports on relative prices also was in vestigated by replacing the real exchange rate index with the real quantity of farm exports. In no case, however, were the qualitative conclusions discussed in the text affected by this change: M1 growth never had a significant effect on the relative price ratio and the effects of trade flows were significant only if significance levels beyond the standard 5 percent level were used. 15A likelihood ratio test suggested by Sims (1980) was employed to select a single lag length for all variables in each of the four equations in the VAR representation. This test indicated a choice of two quarters. U.S. 16The estimations reported in tables 3 and 4 also were per formed using relative price measures defined as the index JULY/AUGUST 1991 40 Table 4 VAR Variance Decomposition: Four-, Eight- and 12-QuarterAhead Forecast Error Variances Innovations Series Dependent variable Relative price Relative prices M1 Exchange rate Industrial production 81.62 81.28 81.28 0.72 0.93 0.93 8.11 8.15 8.15 9.54 9.64 9.65 M1 5.42 6.87 6.90 79.14 76.11 76.04 2.81 2.75 2.75 12.63 14.27 14.30 Exchange rate 6.30 6.47 6.48 7.14 7.33 7.33 84.78 84.24 84.20 1.79 1.96 1.98 29.59 28.92 28.93 13.51 15.10 15.12 2.75 3.05 3.05 54.16 52.93 52.91 Industrial production NOTE: Row 1 = Four-quarter-ahead forecast error Row 2 = Eight-quarter-ahead forecast error Row 3 = 12-quarter-ahead forecast error period, offset any relative price change caused by an aggregate shock. Only the real exchange rate, which has a marginally significant and negative coefficient, is statistically associated with the relative price ratio. While this result is consistent with many of the arguments raised by agricultural econo mists about how restrictive monetary policy could raise the exchange rate, reduce exports and depress farm prices, this line of reasoning is not shown by line 3 of table 3, which indi cates no statistically significant relationship be tween Ml and the real exchange rate. Thus, these reduced-form estimates suggest that monetary changes have little, if any, effect on the relative price ratio. V ariance D eco m p o sitio n Further evidence about the effect of monetary shocks on relative prices is found in table 4, which presents the percentage of four-, eight17Because variables tains the potential VAR results are sensitive to the ordering of [e.g., Cooley and LeRoy(1985)], the table con results from the ordering that gives the largest influence for M1. FEDERAL RESERVE BANK OF ST. LOUIS and 12-quarter-ahead forecast error variances explained by past innovations in the relative price ratio and the other variables in the model.17 Monetary shocks explain less than 1 percent of the relative price forecast error variance, while about 81 percent is attributable to past in novations in the relative price series itself. These findings are generally consistent with those reported by Chambers (1984) and Orden (1986b), who found less than 10 percent of the error variance could be attributed to monetary shocks and more than half could be attributed to past behavior of the relative price ratio.18 Moreover, both the real exchange rate and in dustrial production explain substantially more of the variation in the relative price ratio than does Ml. As noted earlier, other analysts (most notably Schuh) have argued monetary effects are trans18Devadoss and Meyers (1987), who reported large and quite persistent monetary effects on relative prices, did not report a variance decomposition, so their results are not directly comparable. 41 mitted to agriculture through the real exchange rate and its impact on farm exports. Some in sight into this notion is found in table 4, which shows that innovations in the real exchange rate series account for about 8 percent of the variance in relative farm prices. Moreover, mon etary shocks apparently explain only about 7 percent of the variance in real exchange rate movements, a result consistent with the small or non-significant effects of monetary shocks on the real exchange rate reported by Batten and Belongia (1986). Thus, all things and potential avenues of influence considered, these results indicate a statistically weak and numerically small relationship between monetary shocks and movements in the farm/nonfarm price ratio. The shaded insert on page 42 discusses possible changes in these relationships if the thrust of monetary policy is measured by different indicators. E stim a tes F rom a B a rro -T yp e M odel While the foregoing results suggest a fairly weak relationship between monetary shocks and relative price changes, the VAR method is not appropriate for testing the relevant struc tural hypotheses that characterize the models discussed above. In a model that treats a mone tary shock as a shock to aggregate demand, assuming a lower income elasticity for farm products would imply that, in the short run, the farm/nonfarm relative price ratio is inversely related to innovations in Ml. Moreover, because neoclassical models of this nature recognize that nominal shocks affect real or relative magni tudes only in the short run, the sum of the coefficients for lagged innovations in Ml should not be significantly different from zero. The persistence of any short-run nonneutralities, however, remains to be determined. The basic predictions of this model can be ex amined by estimating an equation of the form: P (1) A(p^) = a + i I 0 bj E(m),_j + j I= 0 c, [m-E(m)]t ; *NF = P + Q £ t< where E(m) is the expected growth rate of Ml, [m - E(m)] is the unexpected component of Ml 19Equation 1, in many respects, is the one Devadoss and Meyers (1987), among others, estimated after placing zero restrictions on the bj coefficients. If, however, their results are explained by a fix-price/flex-price (overshooting) economic structure, fully anticipated changes in the money stock could affect the relative price ratio and there is no justification for the restrictions (see equations 2 and 3 growth, and a, b( and c, are coefficients to be estimated over undetermined lag lengths p and q, respectively. Under the assumptions about market structure discussed earlier, the b, coeffi cients should be zero and the c, coefficients for the initial lags of unexpected changes in money growth should take negative values; the model’s general prediction about the long-run neutrality of monetary shocks implies that the sum of c, coefficients should be zero. Lapp (1990), who recently discussed and reported results from a model of this form, found monetary actions to have small, short-lived effects on the relative price ratio.19 Before equation 1 can be estimated, the re quisite values for the aggregate demand shock (the unexpected component of Ml growth) must be obtained. An autoregressive model was fit to the first differences of logarithms of Ml and in spection of the autocorrelation functions in dicated an AR(6) was an adequate representa tion of this series. The null hypothesis that the residuals from this representation were white noise could not be rejected. These residuals were employed in equation 1 as the measure of monetary shocks; the fitted values were used to represent anticipated money growth. A final prediction error (FPE) criterion sug gested estimating a model with contem poraneous and three lagged values for the unanticipated component of money growth and excluding the anticipated portion of money growth entirely. Before estimating equation 1 in this form, it first was estimated using contem poraneous and three lagged values for both monetary variables in order to test more di rectly whether anticipated money growth had any effect on relative prices. As the first row of table 5 indicates, neither component of money growth is related significantly to the farm/nonfarm product ratio. The second row of the table, which reports the results of the model chosen by the FPE criterion, shows that monetary shocks have no permanent effects on the relative price ratio. Moreover, none of the individual lag coefficients (not reported) is significantly different from zero indicating the below); rather, the significance of those coefficients is a key hypothesis to be tested. Moreover, by failing to specify a theoretical model, Devadoss and Meyers also miss the chance to rule out a Barro-type model as an explanation for relative price behavior on the basis of “ wrong” (positive) signs for the Cj coefficients. JULY/AUGUST 1991 42 Is M l The "Right’' Measure Of Monetary Actions? Investigations of how monetary actions af fect economic activity have been influenced in the 1980s by financial deregulation and in novation. Most notable among these financial changes was the introduction of interestbearing checkable deposits, which, many economists believe, has distorted the behavior of Ml since 1981.1 Other research has argued that another measure of Federal Reserve actions—the federal funds rate—is more close ly related to economic activity than money growth; indeed, Orden and Fackler (1989), in a similar study of monetary actions and the relative price of farm products, speculate about whether interest rates are a better gauge of monetary actions than money growth.2 To investigate these issues, the analysis reported in tables 3 and 4 was repeated replacing M l growth with the growth rate of MIA and the first difference of the federal funds rate. MIA, which is Ml less interestbearing checkable deposits, presumably deletes idle savings-type balances from what is intended to be a transactions-based measure of the money stock. Indeed, Darby, et al. (1989) found MIA to be a better em pirical measure of monetary actions in the 1980s than Ml. Using the federal funds rate can be defended by arguing that it examines the influence of an interest rate that is direct ly under the control of the Federal Reserve. The relevant results of these estimations are reported in the table below.3 Revised VAR Estimates Using M1A and the Federal Funds Rate As Monetary Indicators: I/1976-IV/1990_____________________ Revised Reduced-Form Estimates Relative price Monetary indicator Exchange rate Industrial production Relative price 0.215 (1.124) 0.0191 (0.036) -0 .2 3 7 (1.488) Relative price 0.213 (1.148) -2 .7 1 8 2 (1.404) -0 .2 1 0 (1.694) Dependent variable R2 DW -0 .1 6 7 (0.471) .12 1.97 0.034 (0.091) .15 1.96 Revised Variance Decompositions Dependent variable Relative price Monetary indicator Exchange rate Industrial production Forecast horizon Relative price 81.70 81.05 81.04 1.48’ 2.21 2.21 7.53 7.47 7.47 9.29 9.27 9.27 4 Qtr. 8 Qtr. 12 Qtr. Relative price 78.19 77.81 77.81 8.942 9.09 9.09 3.60 3.62 3.62 9.28 9.47 9.47 4 Qtr. 8 Qtr. 12 Qtr. ’ Monetary indicator is A In M1A. 2Monetary indicator is A fed funds rate. 1See, for example, Belongia and Chalfant (1990) for a review of some of the issues. 2See Friedman and Kuttner (1990) for arguments and evidence on the federal funds rate as an indicator of monetary policy. FEDERAL RESERVE BANK OF ST. LOUIS 3Complete results are available from the author, 43 The top portion of the table, which reports the revised reduced-form equations using the growth rate of MIA and changes in the federal funds rate as monetary indicators, shows no qualitative change to the results in table 3: neither indicator of monetary actions is related significantly to movements in the absence of short-run effects as well. If monetary actions have any effect on the farm/nonfarm price ratio, the results in table 5 reject the no tion that they are transmitted through the mech anism described in figure 3. The O versh o o tin g M odel The implications of the overshooting model, derived in Frankel (1986), can be stated in a straightforward manner. The testable hypotheses implied by the model shown in equation 2 are that a change in the log level of the money stock (Am,) or in the expected growth rate of money (A^t) will have larger-than-proportional effects on farm prices. That is, in a regression of the form, (2) APFt = c0 + Cj Am, + c2 A^t + £„ where APFt is the change in the log level of the index of prices received by farmers, the ex pected results are that Cj > 1 and c2 > 1. Thus, “overshooting” occurs because farm prices respond initially by a larger percentage than either the actual level of the money stock or the expected rate of money growth. The standard interpretation of this model, which relative price ratio. Similarly, the bottom por tion of the table indicates that neither monetary indicator explains more than 10 percent of the forecast error variance and that movements in the relative price ratio continue to be dominated by past shocks to the ratio itself. assumes that nonfarm prices are fixed in the short run, also would imply that contractionary monetary policy would temporarily depress the farm/nonfarm relative price ratio, a result op posite to that from model Finally, some analysts have tried to account for business cycle effects on farm prices by ad ding the change in the log level of real output (Ay,) to equation 2. This gives equation 3: (3) APf, = c„ + c, Am, + c2A^, + c3 Ay, + Note that equation 3 maintains the two original overshooting hypotheses implied by equation 2 (c, > 1; c2 > 1). The hypotheses embodied in equations 2 and 3 were tested over the same periods reported earlier. Real output was measured by industrial production. The change in the expected money growth rate, h\x, was calculated as the first dif ference of fitted values from the money growth autoregression discussed earlier. The equations first were fit only with contemporaneous values for right-hand-side variables and then, again, allowing for lags. Results of the estimations are reported in table 6. Table 5 Effects of Monetary Shocks on Relative Prices in a Barro Model Intercept Expected M Unexpected M H1 DW -2 5 .8 6 2 (1.604) 3.371 (1.473) 0-3 -2 .2 9 8 (0.909) 0-3 - .0 3 1.51 -2.211 (1.237) _ 0.493 (0.535) 0-3 -.0 1 1.48 NOTE: The third line of numbers in each row of the table indicates lags estimated. The numbers in parentheses are t-statistics for the sum of the lag coefficients. JULY/AUGUST 1991 44 Table 6 Results from Overshooting Models, I/1976-IV/1990 R2 DW -.0 1 1.56 Intercept Am -0.07 1 (0.20) 0.350 (0.80) 0.381 (0.49) -1 .6 6 2 (0.47) 0.248 (0.59) 0.283 (0.38) 0.793 (2.58) .08 1.60 5.26 (1.36) - 0.485 (1.01) 0-1 0.120 (0.17) 0 0.685 (1.77) 0-1 .23 1.38 Am Ay NOTE: The third line of numbers in the bottom row of the table indicates lags estimated. Numbers in parentheses for bottom regression are t-statistics for the sums of the lag coefficients. Although none of the results for the restricted model shows any effects from either monetary variable, the more general form of the over shooting model indicates a significant contem poraneous relationship between the growth rate of industrial production and index of prices received by farmers; the sum of this effect and the coefficient for the lagged effect, however, is not significantly different from zero. The crucial question for the overshooting model, however, is whether the coefficients associated with the growth rate of Ml and the change in the expected growth rate of Ml are significantly greater than one. For Am, its coeffi cient in each of the three regressions is num er ically less than one and is significantly less than one in the last regression. This rejects a predic tion of the overshooting model. Similarly, the coefficient associated with A^, the change in the expected growth rate of money, is numerically less than one in each case and significantly so in the last equation. The implication is a rejec tion of the overshooting model. CONCLUSIONS Because many studies have found monetary shocks to have positive and persistent effects on the farm/nonfarm relative price ratio, the pur suit of a contractionary monetary policy to reduce inflation has been blamed for causing widespread financial distress in agriculture. Although an understanding of this literature is FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis certainly important to the debate about whether farm programs can or should be used to cushion the sector from changes in macroeconomic policies, the evidence on the response of farm prices and income to monetary policy actually has been widely mixed. In part, this diversity has been due to the different theoretical models and empirical techniques that have been employed. Following a research strategy suggested by King (1979) to distinguish among alternative models and empirical results, a revised set of "stylized facts” emerges on the relationship bet ween monetary actions and the relative price of farm products: • Farm prices are significantly more variable than nonfarm prices. • VAR results consistently show that monetary innovations explain less than 10 percent of the forecast error variance of the farm/non farm price ratio, whereas past innovations in the relative price ratio itself explain 80 per cent or more of the error variance. Thus, while monetary effects may be statistically significant, they are economically unimportant. • Although the flex-price/fix-price model is widely asserted to represent the economic structure generating the farm/nonfarm price series, its main hypotheses are rejected by the data. The standard interpretation of a Barro-type model also is rejected. 45 • Tests find the behavior of farm prices to be consistent with the neoclassical prediction of long-run neutrality; the "long run” for ad justments in the farm/nonfarm price ratio to a monetary change is less than one year. Falk, Barry, S. Devadoss and William H. Meyers. “ Money, Inflation, and Relative Prices: Implications for U.S. Agriculture,” Center for Agriculture and Rural Development, Department of Economics Working Paper Number 86-WP1 (Iowa State University, 1986). Frankel, Jeffrey A. “ Expectations and Commodity Price Dynamics: The Overshooting Model,” American Journal of Agricultural Economics (May 1986), pp. 344-48. Friedman, Benjamin M., and Kenneth Kuttner. “ Money, In come, Prices and Interest Rates After The 1980s,” Federal Reserve Bank of Chicago Working Paper 90-11 (July 1990). REFERENCES Gardner, Bruce L. “ On the Power of Macroeconomic Linkages to Explain Events in U.S. Agriculture,” American Journal of Agricultural Economics (December 1981), pp. 871-78. Akaike, Hirotugu. “ Statistical Predictor Identification,” An nuals of The Institute of Statistical Mathematics (1970, no. 2), pp. 203-17. Granger, C.W.J., and P. Newbold. “ Spurious Regressions in Econometrics,” Journal of Econometrics (July 1974), pp. Barnett, Richard C., David A. Bessler and Robert L. Thomp son. “ The Money Supply and Nominal Agricultural Prices,” American Journal of Agricultural Economics (May 1983), pp. 303-37. Grennes, Thomas, and John S. Lapp. “ Neutrality of Inflation in the U.S. Agricultural Sector,” Journal of International Money and Finance (June 1986), pp. 231-43. Barro, Robert J. “ Rational Expectations and The Role of Monetary Policy,” Journal of Monetary Economics (January 1976), pp. 1-32. Han, Doo Bong, Dennis W. Jansen, and John B. Penson, Jr. “ Variance of Agricultural Prices, Industrial Prices, and Money,” American Journal of Agricultural Economics (November 1990), pp. 1066-73. Batten, Dallas S., and Michael T. Belongia. “ Monetary Policy, Real Exchange Rates, and U.S. Agricultural Ex ports,” American Journal of Agricultural Economics (May 1986), pp. 422-27. King, Richard A. ’’Choices and Consequences,” American Journal of Agricultural Economics (December 1979), pp. 839-48. Batten, Dallas S., and Daniel L. Thornton. “ Lag Length Selection and Tests of Granger Causality Between Money and Income,” Journal of Money, Credit and Banking (May 1985), p. 164-78. Belongia, Michael T., and James A. Chalfant. “Alternative Measures of Money as Indicators of Inflation: A Survey and Some New Evidence,” this Review (November/December 1990), pp. 20-33. Belongia, Michael T., and Courtenay C. Stone. “ Would Lower Federal Deficits Increase U.S. Farm Exports?” this Review (November 1985), pp. 5-19. Bessler, David A. “ Relative Prices and Money: A Vector Autoregression on Brazilian Data,” American Journal of Agricultural Economics (February 1984), pp. 25-30. Chambers, Robert G. “ Credit Constraints, Interest Rates, and Agricultural Prices,” American Journal of Agricultural Economics (May 1985), pp. 390-95. _______ . “Agricultural and Financial Market Interdependence in the Short Run,” American Journal of Agricultural Economics (February 1984), pp. 12-24. Cooley, Thomas F., and Stephen F. LeRoy. “Atheoretical Macroeconometrics: A Critique,” Journal of Monetary Economics (November 1985), pp. 283-308. Darby, Michael R., Angelo R. Mascaro, and Michael L. Marlow. “ The Empirical Reliability of Monetary Aggregates as Indicators: 1983-87,” Economic Inquiry (October 1989), pp. 555-85. Dickey, D. A., and W. A. Fuller. “ Distribution of the Estimators for Autoregressive Time Series with a Unit Root,” Journal of the American Statistical Association (June 1979), pp. 427-31. 111- 20 . Lapp, John S. “ Relative Agricultural Prices and Monetary Policy,” American Journal of Agricultural Economics (August 1990), pp. 622-30. Meltzer, Allan H. “ The Fed at Seventy-Five,” in, Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System (Kluwer Academic Publishers, 1990). Obstfeld, Maurice. “ Overshooting Agricultural Commodity Markets and Public Policy: Discussion,” American Journal of Agricultural Economics (May 1986), pp. 420-21. Orden, David. “ Money and Agriculture: The Dynamics of Money-Financial Market-Agricultural Trade Linkages,” Agricultural Economics Research (Summer 1986a), pp. 14-28. _______ . "Agriculture, Trade, and Macroeconomics: The U.S. Case,” Journal of Policy Modeling Spring (1986b), pp. 27-51. Orden, David, and Paul L. Fackler. “ Identifying Monetary Im pacts on Agricultural Prices in VAR Models,” American Journal of Agricultural Economics (May 1989), pp. 495-502. Penson, John B., and Bruce L. Gardner. “ Implications of the Macroeconomic Outlook for Agriculture,” American Journal of Agricultural Economics (December 1988), pp. 1013-22. Plosser, C.I., and G.W. Schwert. “ Money, Income and Sunspots: Measuring Economic Relationships and the Ef fects of Differencing,” Journal of Monetary Economics (November 1978), pp. 637-60. Rausser, Gordon C. “ Macroeconomics and U.S. Agricultural Policy,” U.S. Agricultural Policy: The 1985 Farm Legislation, ed. Bruce L. Gardner (American Enterprise Institute for Public Policy Research, 1985), pp. 207-52. _______ . “ The Likelihood Ratio Statistics for Autoregressive Time Series with a Unit Root,” Econometrica (July 1981), pp. 1057-72. Rausser, Gordon C., James A. Chalfant, H. Alan Love, and Kostas G. Stamoulis. “ Macroeconomic Linkages, Taxes, and Subsidies in the U.S. Agricultural Sector,” American Journal of Agricultural Economics (May 1986), pp. 399-412. Devadoss, S., and William H. Meyers. “ Relative Prices and Money: Further Results for the United States,” American Journal of Agricultural Economics (November 1987), pp. 838-42. Robertson, John C., and David Orden. “ Monetary Impacts on Prices in the Short and Long Run: Some Evidence from New Zealand,” American Journal of Agricultural Economics (February 1990), pp. 160-71. JULY/AUGUST 1991 46 Schuh, G. Edward. “ The Exchange Rate and U.S. Agriculture,” American Journal of Agricultural Economics (February 1974), pp. 1-13. Taylor, J. S., and J. Spriggs. “ Effects of the Monetary Macro-economy on Canadian Agricultural Prices,” Cana dian Journal of Economics (May 1989), pp. 278-89. Sims, Christopher. “ Macroeconomics and Reality,” Econometrica (January 1980), pp. 1-48. Tegene, Abebayehu. “ The Impact of Macrovariables on the Farm Sector: Some Further Evidence,” Southern Journal of Agricultural Economics (July 1990), pp. 77-85. Starleaf, Dennis R. “ Macroeconomic Policies and Their Im pact Upon the Farm Sector,” American Journal of Agricultural Economics (December 1982), pp. 854-60. Starleaf, Dennis R., William H. Meyers and Abner W. Womack. “ The Impact of Inflation on the Real Income of U.S. Farmers,” American Journal of Agricultural Economics (May 1985), pp. 384-89. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Tweeten, Luther G. "Macroeconomics in Crisis: Agriculture in an Underachieving Economy,” American Journal of Agricultural Economics (December 1980), pp. 853-65. _______ . “ Impact of Federal Fiscal-Monetary Policy on Farm Structure,” Southern Journal of Agricultural Economics (July 1983), pp. 61-68. U.S. Department of Agriculture. “ Major Statistical Series of the U.S. Department of Agriculture: How They are Used and Constructed,” Handbook #365 (1970). 47 Michelle R. Garfinkel and Daniel L. Thornton Michelle R. Garfinkel, an assistant professor at the University of California at Irvine, was a senior economist at the Federal Reserve Bank of St. Louis while this paper was written. Daniel L. Thornton is an assistant vice president at the Federal Reserve Bank of St. Louis. Richard I. Jako and Scott Leitz pro vided research assistance. The Multiplier Approach to the Money Supply Process: A Precautionary Note T T h E MULTIPLIER MODEL of the money supply, originally developed by Brunner (1961) and Brunner and Meltzer (1964), has become the standard paradigm in macroeconomics and money and banking textbooks to explain how the policy actions of the Federal Reserve in fluence the money stock. It also has been used in empirical analyses of money stock control and the impact of monetary policy actions on other economic variables. One important feature of this model is that it decomposes movements in the money supply in to the part that is due directly to Federal Re serve policy actions (the adjusted monetary base) and the part that is due to changes in technology and the tastes and preferences of depository in stitutions and the public (the money multiplier). In this decomposition, the multiplier is assumed to be independent of the policy actions of the central bank. The independence is implicitly predicated on the assumptions that the demands 1The notion that the multiplier is independent of Federal Reserve actions— implicit in the work of Brunner and Meltzer (1964, 1968) and, more recently, in Plosser (1991) — has never been demonstrated rigorously with micro-eco for both checkable deposits and currency are determined by the same factors, and that indi viduals can quickly and costlessly alter their holdings of currency and checkable deposits to achieve the desired proposition of the two alter native forms of money.1 Open market purchases, for example, increase reserves and consequently checkable deposits; but the public simply shifts from checkable deposits to currency until the (unchanged) desired ratio of currency relative to checkable deposits is once again achieved. Be cause policy actions have no impact on the pub lic’s holdings of currency relative to checkable deposits, the multiplier does not depend directly on the policy actions of the Fed. This article investigates the theoretical and empirical validity of the key feature of the mul tiplier approach. In theory, the multiplier is in dependent of the policy actions of the Federal Reserve only if the demands for currency and checkable deposits are determined by identical nomic principles. The argument presented here that would suggest such independence is implicit in works as early as Fisher (1911). JULY/AUGUST 1991 48 factors and if, conditional on these factors, these demands are strictly proportional. From an em pirical perspective, this condition is necessary but not sufficient; the degree to which the mul tiplier is influenced by policy actions also de pends on the strength of the relationship be tween policy actions and checkable deposits. An empirical analysis shows that most of the variability of the observed ratio of currency to checkable deposits is due to variation in check able deposits, and thereby suggests that the de mand for currency is not strictly proportional to the demand for checkable deposits. Prior to the Monetary Control Act of 1980 (MCA), how ever, the link between reserves and checkable deposits was quite loose—so much so, that the notion that the multiplier is independent of pol icy actions was operationally valid. Nevertheless, the empirical relevance of this notion has weak ened considerably since the implementation of the MCA in the early 1980s. Since then, the re lationship between Fed policy actions and checkable deposits and, thus, the multiplier has tightened markedly. The evidence presented here, that the multi plier is not independent of Federal Reserve ac tions in the post-MCA period, raises some ques tions about the appropriateness of using the monetary base as an indicator of the effects of policy actions on the money stock. More impor tant from a policy perspective, it also suggests a modification of the standard approach to money stock control that might yield substantial im provements in effective monetary aggregate targeting. THE M ONEY M ULTIPLIE R A P P R O A C H : A SIMPLE EXAM PLE As a starting point for understanding the de composition of the money supply into the mone tary base and the multiplier, note that the nar row money stock, Ml, is defined as (1) Ml, = TCD, + C„ where TCD denotes total checkable deposits and C denotes the currency held by the nonbank public. The monetary base (MB), not adjusted for changes in reserve requirements, is simply the sum of currency and reserves (including 2Since the Fed eliminated reserve requirements on all non transaction deposits in December 1990, this representation approximates the current system. For convenience of ex FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis cash in the vaults of depository institutions) in the banking system, R: (2) MBt= Ct+ Rt. Currency, supplied by the Federal Reserve on demand, reflects the portfolio decisions of the public rather than monetary policy actions. Re serves, in contrast, can be affected directly by the Fed’s sales or purchases of government se curities in the open market. For simplicity, assume that the Federal Re serve has a simple system of reserve require ments, with required reserves, RR, given by (3) RR,= rTCD,, 0 < r < 1, where r denotes the ratio of reserves that must be held against TCD.2 A change in the reserve requirement ratio, r, also would constitute a monetary policy action by the Fed. Furthermore, for simplicity, assume that ac tual reserves always equal required reserves so that excess reserves are identically zero. With this simplifying assumption, equation 3 can be rewritten as (4) R, = rTCD,. The model is completed by assuming that cur rency is held in some proportion, k, of TCD. That is, (5) C, = kTCD,, where the proportion k, hereafter called the kratio, is the public’s desired ratio of currency to TCD holdings. Combining equations 1, 2, 4 and 5 produces the monetary base-multiplier representation of the money supply: (6) M lt = m MB,, where m, the money multiplier, is given by (7) m = (1 +k)/(r + k). According to this representation, a policy action that increases R by one dollar, through open position, the discussion to follow abstracts from reserves that depository institutions must hold on government and foreign transactions balances. 49 market purchases of government securities, in creases MB by one dollar and the money stock by m dollars.3 In this representation, policy actions are reflected not only in MB, through changes in R, but in m, through changes in r. With a simple adjustment to MB, however, the effects of pol icy actions on the money supply can be isolated in one measure. This alternative measure of the monetary base, called the adjusted monetary base, AMB, reflects both changes in R and r. It is constructed by calculating the hypothetical level of reserves that would have been required under the reserve requirements in existence dur ing a chosen base period for the current (actual) level of reservable deposits. With the chosen base period, changes in required reserves due to changes in reserve requirements, r, are added to the monetary base.4 Specifically, the AMB is given by (8) AMB, = MB, + RAM,, where the reserve adjustment magnitude, RAM, is defined as (9) RAM, = (r*-r)TCD,. This adjustment measures the reserves released or absorbed by changes in r relative to r*, the required reserve ratio during the base period. In the base period, RAM is zero and AMB = MB. A decrease in r from its base-period level (r*) releases reserves into the banking system and thereby increases RAM and AMB. Conversely, an increase in r reflects the reserve drain by reducing RAM and AMB. Combining equations 1, 4, 5, 8 and 9 yields the following decomposition of Ml, 3Note that because r< 1 , m > 1 . If the assumption that ex cess reserves are not held were replaced by the assump tion that they are held in a fixed proportion, e, of TCD, then the denominator of the multiplier would include e as well, so that the multiplier would be smaller than that shown in equation 7. 4See Tatom (1980), for example, who discusses the issue of choosing the appropriate base period in light of changes in the structure of reserve requirements. This theoretic discussion focuses on the measure of the adjusted mone tary base constructed at the Federal Reserve Bank of St. Louis. See Garfinkel and Thornton (1991) for a more de tailed discussion of this measure and a similar one con structed by the Federal Reserve Board. (10) Ml, = m* AMB,, where (11) m* = (l + k)/(r*+k). In this characterization of the money supply pro cess, all changes in monetary policy, through changes in r or R, are reflected in the AMB. Changes in the multiplier reflect only changes in the public’s desire to hold currency relative to checkable deposits, changes in the k-ratio.5 Because, in this model, the k-ratio is not directly influenced by the policy actions of the Fed, the multiplier is independent of policy. THE D EM AND FOR CURRENCY, CHECKABLE DEPO SITS AN D NEAR-MONIES: W H A T IS THE k-RATIO? Interest in the currency-deposit ratio dates back to Fisher (1911), who was concerned that the two forms of money had different income velocities. He realized that these two monies are imperfect substitutes: currency is especially use ful for making small, "face-to-face" transactions, while checkable deposits provide a convenient means for making large, “out-of-town” transactions. Fisher reasoned, however, that individuals achieve an "equilibrium” in their holdings of the two forms of money. The notion of a desired or optimal k-ratio is based on the assumption that individuals decide how much of their money holdings they will allocate between currency and checkable deposits, based on both the rela tive advantages of each in undertaking an indiv idual’s planned transactions and their relative holding cost. This ratio was assumed to be a ment and foreign transaction balances, m * = (1 + k)/ (r*(1 + g + f) + k), where g and f denote the ratios of gov ernment and foreign transactions accounts to TCD, re spectively. If, in addition, excess reserves were held, as described in footnote 2, m* = (1 +k)/(r*(1 + g + f) + k + e). These complications can be ignored, however, because movements in the observed ratio of currency to TCD ex plain most of the movements in the multiplier, as will be discussed shortly. 5ln a slightly more realistic model, which allows for the fact that depository institutions must hold reserves on govern JULY/AUGUST 1991 50 function of a num ber of economic and social variables.6 Given these variables, the demands for cur rency and checkable deposits were thought to be strictly proportional to each other. More over, because individuals are free to adjust their holdings of the two monies quickly and costless ly, it was assumed that the actual currencydeposit ratio would deviate from the desired ratio for only a short period of time.7 According to this line of reasoning, all changes in the observed currency-to-deposit ratio, denoted here by the K-ratio, are to be interpreted as changes in the desired ratio caused by one of these fac tors. While not numerically constant, as it was assumed to be in the previous analysis, the kratio was viewed as not being directly affected by monetary policy actions. The following discussion, supported by subse quent empirical analysis, suggests, however, that the observed ratio of currency to checkable deposits can be and has been affected directly by the policy actions of the Federal Reserve.8 This effect can emerge without changing the relative advantages of currency and checkable deposits or their relative holding cost. S u b stitu ta b ility , H oldin g C osts a n d th e O ptim al k-R atio There are a num ber of reasons why one might question the assumption that changes in the observed currency-to-deposit ratio necessar 6Fisher assumed that the optimal k-ratio depended on real income or wealth, the degree of development of the business sector, population density, relative holding costs and custom and habit. (Checkable deposits were thought to be “ superior” to currency, although money in any form was a superior good.) Cagan (1958) extends the list of determinants of the k-ratio considerably. Both he and Hess (1971) attempt to quantify the effects of such factors. 7Cagan (1958) recognized that, at times, restrictions might prevent the adjustment of the currency ratio. He explicitly considered the case of financial crises where banks sus pended convertibility. He noted, “ At these times individ uals could not exchange deposits for currency, and the desired currency ratio undoubtedly exceeded the actual ratio...” Without such restrictions, however, individuals are free to adjust their currency holdings to the desired level very quickly and at a low cost. The assumption that there exists a desired currency-todeposit ratio and that individuals adjust their actual hold ings of currency to their desired level was made opera tional for models of the money supply process by Karl Brunner and Allan Meltzer in a series of articles. See Brunner (1961) and Brunner and Meltzer (1964, 1968). 8lt has long been recognized that policy actions can have an indirect effect on the multiplier through the presumed http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis ily reflect changes in the optimal k-ratio—that is, changes in the relative holding cost and ad vantages of currency and checkable deposits. First and perhaps foremost among these is that the demand for either of these forms of money might depend on a number of special factors that are unrelated to the demand for the other. Thus, changes in the relative advantages of these two forms of money might not be empiri cally important in explaining changes in the ratio of currency to checkable deposits. For example, many believe that currency has no rival for illegal transactions. The same is true for foreign demand for U.S. currency by countries that need “hard currencies” for their domestic transactions.9 To the extent that cur rency is held for these reasons, independent of factors that determine the demand for check able deposits, policy actions can induce changes in TCD without affecting currency demand. Con sequently, policy can alter the ratio of currency to TCD and, hence, the multiplier.10 One might also argue that changes in the rela tive holding cost of the two monies are not es pecially relevant for explaining observed changes in the currency-to-deposit ratio. The relative holding cost of the alternative monies is given by the difference between the rates of return on the two forms of money.11 The return on holding currency is zero.12 Although non-interest-bearing checking accounts (demand deposits) have an explicit return of zero, they can yield a effect of policy actions on economic variables such as real income or interest rates. It is argued that such variables influence the k-ratio or the other ratios that make it up— particularly, the ratio of excess reserves to total checkable deposits. See Mishkin (1989) for a more detailed discus sion of this indirect effect. 9Because it is difficult to account for a relatively large amount of the total stock of U.S. currency outstanding, one should not be too surprised to find that, in the ag gregate, demand for currency and checkable deposits are not closely related. See, for example, Avery, Elliehausen, Kennickell and Spindt (1986, 1987). 10This potential influence is illustrated below with an exam ple. To be sure, longer-run movements in the K-ratio might be attributable to some factors that affect the relative ad vantages for the two forms of money. 11The discussion to follow focuses on the nonbank public’s perspective. The relative holding costs to depository in stitutions generally will differ. 12Adjusted for inflation, it is minus the expected rate of infla tion. Note that currency used for illegal transactions yields a greater return because of the tax avoidance. For foreign ers, currency can yield a return that differs from zero due to the appreciation or depreciation of their home currency relative to the U.S. dollar. 51 positive implicit return—for example, free toast ers for new customers, subsidized accounting and payment services, etc. The return on hold ing interest-bearing checking accounts is the net interest paid on these accounts plus free pay ment services.13 The relative holding cost of currency and demand deposits, however, is unresponsive to movements in market interest rates because the explicit returns to both assets are identically zero. Surprisingly, the same seems to be true for currency and interest-bearing checking ac counts, even since the elimination of Regulation Q ceiling rates in 1986. Interest rates paid on interest-bearing checkable deposits included in TCD have been unresponsive to movements in short-term interest rates.14 Despite the fact that the explicit holding cost of currency relative to that of checkable deposits has changed little, the observed ratio of currency to checkable depos its exhibited sharp swings during the 1980s. Thus, it is unlikely that changes in the public’s holding of currency and checkable deposits are due primarily to changes in their relative hold ing cost. The H oldin g o f C u rren cy, C h ecka ble D ep o sits a n d O th er F inancial A ssets Thus, it would not appear that individuals simply shift their money holdings between cur rency and checkable deposits in response to variations in their relative advantages or holding cost. This conjecture would be reinforced by the fact that currency and especially checkable deposits are substitutes for other “near-money” stores of wealth, for example, money market mutual funds. From this broader perspective, the demands for currency and checkable depos 13Net interest is interest net of service charges. For a dis cussion of these, see Carraro and Thornton (1986). This explicit return also could be adjusted for inflation. '■•Indeed, interest rates on the interest-bearing portion of TCD, called other checkable deposits (OCD), have changed little during the 1980s. The rate on OCD fluctuated be tween 5 percent and just over 5.5 percent during our sam ple period. 15This consideration raises a fundamental question— namely, what constitutes an appropriate monetary aggregate? In theory, monetary aggregation requires the “ monetary” ag gregate to be “ weakly separable.” That is to say, it must behave as a fundamental commodity with respect to con sumption and other financial assets. There can be substitution between assets that compose the aggregate, but not between those that compose the aggregate and its are seen as being determined simultaneously with the demand for near-money assets.15 An important part of the determination of the ratio of currency to checkable deposits, there fore, is the degree of substitutability between currency and demand deposits on the one hand and between each of these money assets and near-money assets on the other. Although the explicit rates paid on TCD are relatively unre sponsive to changes in market interest rates, rates paid on near-money assets can vary mark edly with variations in other market interest rates. The effect of these variations on the pro portion of Ml held in the form of currency, of course, depends on the degree of substitutabili ty between near-money assets and the two forms of money. If currency is a relatively poor substi tute for such assets while TCD is a relatively good one, the ratio of currency to TCD will change with changes in rates paid on such nearmoney assets because of changes in TCD. The relevance of this substitutability between TCD and other near-money assets appears to have been heightened by the nationwide intro duction of interest-bearing checking deposits in January 1981. Since then, the cross-price or in terest elasticity of the demand for checkable deposits has increased. This increase is hardly surprising because the payment of interest on checkable deposits has made them closer substi tutes for interest-bearing time and savings de posits. Indeed, some evidence suggests that in dividuals have shifted a significant portion of their "savings” balances into interest-bearing checking accounts.16 Because these saving bal ances are substitutes for savings and money market accounts that have higher explicit re turns, the interest elasticity of the demand for checkable deposits should have risen, while the interest elasticity of currency demand should not have changed.17 those that do not. Some evidence suggests that, while cur rency and demand deposits satisfy this condition for weak separability, these two assets plus interest-bearing transac tion balances do not. See, Fisher (1989), for example. Belongia and Chalfant (1989), among others, however, find that the data are consistent with the notion that the assets included in M1 and a grouping broader than M1 (currency and total checkable deposits) satisfy the weak separability condition. Thus, the empirical results in this line of research are not conclusive. 16See Sill (1990). 17See Thornton and Stone (1991) for a derivation of this result. These results are borne out empirically by simple linear regressions of the monthly change in both currency and other checkable deposits on a scale measure and the three-month T-bill rate. JULY/AUGUST 1991 52 Thus, changes in interest rates, whether pol icy induced or not, can have an asymmetric ef fect on the demands for currency and checkable deposits, with a direct effect on the proportion in which the alternative monies are held.18 Al though this asymmetric effect is likely to have played a larger role since the introduction of interest-bearing checking accounts in generating fluctuations in the ratio of currency to TCD, policy has induced changes in this ratio more directly since the MCA (as discussed below). D EPO SIT SU BSTITU T IO N AN D THE M ONEY M ULTIPLIE R Provided that the demands for currency and checkable deposits are determined by factors that are independent of one another, monetary policy actions can have a direct influence on the relative holdings of each and, thus, the multi plier.19 The channel of influence is most easily illustrated in the extreme case where the de mand for currency is completely independent of the demand for checkable deposits. That is, equation 5 is replaced with (5') C, = C, where C is a constant. Equation 1 also can be rewritten as (10 Ml, = (1 + K,)TCDt, where, as defined previously, Kt = (C/TCD)t is the observed ratio of currency to TCD.20 Using equations 1' and 5' in place of 1 and 5, the money supply can be written as (12) M l, = m ,' AMB,, where m*' = [(l + K,)/(r*+K,)]. The crucial difference between this expression and equation 10 is that, here, policy actions af18The same would be expected for changes in the level of income. Indeed, Hess (1971) presents estimates indicating asymmetric effects of both changes in interest rates and income on the demands for currency and checkable deposits. It should be emphasized that this effect of changes in interest rates on the k-ratio is not the same as that which was alluded to earlier-i.e., through the relative holding cost of currency and checkable deposits (see foot note 8). 19Many researchers who have estimated currency demand equations have abstracted from the relationship of curren cy to TCD. For example, see Hess (1971) and Dotsey (1988). http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis feet both the adjusted monetary base and the money multiplier. To see why, consider a policy action involving the purchase of T-bills in the market by the Fed. This policy action increases the stock of reserves and, assuming zero excess reserves, TCD. In the earlier formulation of his model, the K-ratio was assumed to be unchang ed; the increase in TCD would be accompanied by a proportionate increase in currency, so that the observed ratio of currency to TCD, K, would not change. Thus, the effect of this policy action on the money stock would be isolated in the monetary base—the multiplier would be unaffected. In the modified model, however, TCD increases while currency is constant. Consequently, the Kratio falls and the multiplier, m*', rises. In this instance, the change in monetary policy is re flected both in the adjusted monetary base, be cause of a change in R, and in the multiplier because of a policy-induced change in K. Al though this argument is made in terms of a stat ic model, the main point, that policy can in fluence the multiplier, would carry over into a more realistic dynamic model. Two of the more salient features of the longer-run consequences of this analysis are taken up in the shaded in sert on page 54. THE RECENT BE H A V IO R OF THE K -R ATIO Figure 1 shows the K-ratio and the observed adjusted monetary base multiplier, Ml/AMB, from January 1970 to November 1990. Note that the multiplier is essentially the m irror image of the K-ratio; the K-ratio accounts for much of the multiplier’s short-run (month-to-month) variability and for the significant shifts in its longer-run “trends.” Indeed, as shown in table 1, changes in the K-ratio alone explain over 80 percent of the month-to-month variability in 20Because k is meaningless in this formulation, K, will not equal k. More generally, currency demand can be thought of as having two components, one related to TCD as em bodied in the k-ratio and the other unrelated to TCD. That is, C, = C + kTCD,. In this more general formulation, the k-ratio is determined solely by the relative holding cost of currency and TCD and the substitutability between them as discussed above. In this case, K , = — Q + k. The restriction in (5’), that k = 0, is imposed only for illustrative purposes. 53 Figure 1 The K-Ratio and the M1 Multiplier January 1970-November 1990 Ratio R a tio Table 1 Regression Estimates of Changes in the Multiplier on Changes in the K-ratio Period Constant K-ratio SEE R2 D.W. 1/1970-12/1980 .000 (0.35) -4 .3 5 5 * (13.94) .007 .598 2.47 1/1981-11/1990 .001 (1.84) -3 .7 1 4 * (23.02) .005 .818 2.49 3/1984-11/1990 .001 (1.35) -3 .5 0 4 * (21.66) .004 .854 3.02 * indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. JULY/AUGUST 1991 54 The Long-Run Multiplier? That the K-ratio does not appear to be sta tionary (in the sense of being mean-reverting) raises an interesting question of how the magnitude of the multiplier is determined in the long run. Two examples are presented to show that the multiplier is not invariant to monetary policy and the "long-run multiplier” is critically dependent on the specifications of the demands for currency and TCD. The first example assumes that the demand for real currency is determined solely by real income, while the demand for real TCD is determined both by real income and the nom inal interest rate, where TCD is inversely related to the latter. This example captures the notion that the demand for TCD is deter mined simultaneously with the demands for other near-monies. For simplicity, these demands are assumed to be linear in natural logs, and the income elasticities of the demands for currency and TCD are assumed to be equal. Under these assumptions, the natural log of the K-ratio, InK, depends solely on the natural log of the nominal interest rate and is positively related to it. By influ encing the nominal interest rate, which equals the real rate plus a premium for ex pected future inflation, monetary policy would affect the long-run multiplier. To see why, assume that a change in policy raises reserve growth permanently. If this in crease results in a permanent increase in the actual and anticipated rates of inflation, the nominal interest rate will rise. The perman ently higher level of the nominal interest rate will increase the level of the K-ratio causing a permanent reduction in the multiplier. The second example assumes that the de mand for currency is driven largely by forces external to the domestic economy, say, for eign demand for U.S. currency. It also as sumes that the domestic demand for curren cy is determined solely by the relative holding cost of currency vs. TCD and that this cost is constant. Again, these relation ships are assumed to be linear in the natural 'The parameter h might be thought of as the log of the op timal k-ratio, reflecting only the relative advantages and costs of currency and TCD. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis logs. If the foreign demand grows at a con stant rate, b, then the log of the demand for currency is given by InC, = b t + h lnTCDt, where t denotes a time trend and h is a con stant.1 What happens to the the K-ratio in the long run is determined by the relative growth rates of foreign and domestic demands for currency. For example, assume that the de mand for nominal TCD is determined solely by nominal income and that nominal income grows at a constant rate, d—at least in the long run. If d is less than or equal to b/(l-h), the K-ratio will rise without bound and the multiplier will approach unity. If d is greater than b/(l-h), the K-ratio will approach zero and the multiplier will approach 1/r*, where r* is the base-period reserve requirement (see the text for more details). Note, however, that, in this example, the long-run multiplier is not independent of pol icy actions. For example, assume that d>b/(l-h) so that the multiplier is ap proaching 1/r*. Now assume that a change in policy reduces the growth rate of TCD and, thus, the rate of inflation and the growth rate of nominal income. At the very least, this policy change would cause the multiplier to approach its long-run equilibrium value more slowly, as it drives down d. Indeed, if the growth rate of TCD slowed to the point where d<b/(l-h), the long-run multiplier could converge to 1 rather than 1/r*. Of course, there are a number of other in teresting possibilities. The major points that, even in the long run, the multiplier depends on monetary policy and that the exact value of the long-run multiplier between 1 and 1/r* depends critically on the specifications of the demands for currency and TCD are nonethe less valid. Before a meaningful “long-run” representation of the multiplier can be ob tained, it is necessary to specify carefully both the demand for currency and the de mand for TCD.2 2Note that, because the multiplier is bounded, M1 and AMB must be cointegrated. 55 Figure 2 The K-Ratio, Currency and TCD Ratio changes in the multiplier since the implementa tion of the MCA.21 The MCA tightened the link between the K-ratio and the multiplier by reducing or eliminating other sources of varia tion in the multiplier.22 While the MCA was im plemented in a series of steps from November 1980 to September 1987, its major features were almost fully implemented by February 1984.23 Since then, changes in the K-ratio alone explain over 85 percent of changes in the multiplier. 21The Durbin-Watson statistic for each of the equations in dicates significant, negative first-order serial correlation. Because we are primarily interested in the explanatory power of changes in the K-ratio as measured by the ad justed R-square, however, maximum likelihood estimates of the equations adjusted for serial correlation are not reported here. Nonetheless, there are no substantive dif ferences between the maximum likelihood estimates and those reported in table 1. B illio ns o f D ollars The R ela tio n sh ip B etw een T otal C h ecka ble D ep o sits a n d the K -R atio Figure 2 shows the K-ratio, currency and TCD. The behavior of these series suggests that changes in the trend of the K-ratio are associ ated more closely with changes in the trend of TCD than with changes in the trend of curren cy growth. For example, the sharp rise in the 23The MCA was first implemented in November 1980 and was fully phased-in by September 1987. The empirically significant features of the act were completed with the Fed’s adoption of contemporaneous reserve requirements in February 1984, so the sample was broken at this point. See Garfinkel and Thornton (1989) for a discussion of these changes and their effect on the multiplier. “ See Garfinkel and Thornton (1989) for details. JULY/AUGUST 1991 56 Figure 3 Deviations of the Growth Rates of the K-Ratio and Currency from Their Means January 1970 thru November 1990 Currency K -R a tio 40 40 30 K-Ratio Growth Rate 20 10 -10 -2 0 -3 0 -3 0 1970 72 74 76 78 K-ratio in the early 1970s is associated with a slowing in the growth of TCD. The decline in the K-ratio in the early 1980s and its subse quent rise are clearly associated with a sharp acceleration in the growth of TCD followed by a sharp deceleration in its growth. That TCD accounts for much of the shortrun variation in the K-ratio also is evidenced by figures 3 and 4, which show, respectively, devi ations of the growth rate of the K-ratio from its mean and deviations of the growth rates of cur rency and TCD from their respective means. As FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis 80 82 84 86 88 1990 shown in the figures, the month-to-month vari ability in the growth of TCD is considerably larger than that of currency. The variability of TCD more closely matches the variability of the K-ratio than does the variability of currency. While the growth rates of the K-ratio and TCD are highly, inversely related, there is little positive association between the growth rate of the K-ratio and the growth rate of currency. This observation is verified in table 2, which shows the simple correlations between the monthly annualized growth rates for currency 57 Figure 4 Deviations of the Growth Rates of the K-Ratio and Total Checkable Deposits From Their Means January 1970 thru Novem ber 1990 K -R a tio 40 and the K-ratio and for TCD and the K-ratio for four periods of roughly equal length between January 1970 and November 1990. If variation in the K-ratio were simply due to shifts between currency and TCD, its variation would be equally attributable to variation in currency and TCD. This is not the case, however. The growth rates of currency and the K-ratio were positive ly correlated during only two of the four peri ods. They were negatively correlated in the other two, although the correlations are not sig nificantly different from zero. In contrast, there is a strong, consistent negative correlation be tween the growth rate of TCD and the K-ratio during all four of the periods. Figures 3 and 4 and the correlations reported in table 2 clearly suggest that month-to-month variability in the Table 2 Correlations Between the Monthly Growth Rate of the K-Ratio and the Monthly Growth Rates of Currency and TCD Period 1/1970-12/1974 1/1975-12/1979 1/1980-12/1984 1/1985-11/1990 K-ratio and currency .368* -.0 1 6 -.1 1 2 .265* K-ratio and TCD -.9 0 1 * -.9 1 1 * - .955* - .951 * * indicates statistical significance at the 5 percent level. JULY/AUGUST 1991 58 K-ratio is driven largely by movements in TCD. Finally, as shown in figure 4, periods of per sistent deviations in the growth rate of TCD above (below) its mean are associated with per sistent deviations of the growth rate of the Kratio below (above) its mean. Consequently, both the short and long-run movements of the Kratio are associated with movements in TCD rather than currency. The apparent importance of TCD in influencing the K-ratio suggests that K-ratio changes have not occurred simply because of variations in the relative advantages and holding cost of currency and TCD. That is to say, changes in the K-ratio have not been a simple result of the public’s desire to shift the composition of Ml between currency and checkable deposits. The L ink B etw een T otal C h ecka ble D ep o sits a n d R e se rv e s Movements in the multiplier appear to be deter mined primarily by movements in the K-ratio, which, in turn, appear to be determined pri marily by changes in TCD. The question that re mains is what determines the stock of TCD out standing? The models of the money supply pre sented above provide a simple answer: given the reserve requirement ratio, TCD is deter mined solely by the amount of reserves supplied by the Federal Reserve. This strong link arises in this model because reserves are assumed to be held only to support checkable deposits.24 Prior to the MCA, commercial member banks were required to hold reserves against all time 24ln reality, of course, depository institutions hold excess reserves and are required to hold reserves on transaction deposits other than those included in M1. 25The total reserves measure used here is total reserves ad justed for reserve requirement changes, prepared by the Federal Reserve Board. 26The Durbin-Watson statistic for the first time period sug gests that there is significant first-order positive serial cor relation, as would be expected given the likelihood of misspecification (see the appendix). Maximum likelihood estimates of this equation adjusting for first-order serial correlation confirm this result. The estimated coefficient of first-order serial correlation is -.314 with a t-statistic in ab solute value of 3.29. Nevertheless, the parameter estimates after adjusting for serial correlation are generally close to those reported in table 3, and they are statistically significant. More important, the adjusted R-square only in creases to .147; hence, the dramatic rise in the adjusted R-square in the 1980s is not due to the fact that total reserve captures the autoregressive part of TCD. FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis and saving deposits, while non-member banks and other depository institutions were not re quired to hold reserves against their transaction deposits in Ml. Because of these factors, the link between TCD and reserves was not particu larly strong. In reducing or eliminating reserve requirements on a num ber of non-transaction accounts and extending reserve requirements to all depository institutions, however, the MCA significantly strengthened the relationship be tween TCD and reserves. The effect of the MCA is illustrated in table 3, which shows the results of simple linear regres sions of changes in TCD on changes in total re serves, TR, for several periods between January 1970 and November 1990.25 The regression equa tions in this table (and in subsequent ones) are intended to be illustrative and should not be in terpreted as alternative models for the money supply process. (See the appendix for details.) In all cases but the initial phase-in of the MCA, there is a statistically significant relationship between changes in TCD and TR. The strength of the relationship, as measured by the adjusted R-square, however, increases after the imple mentation of the MCA.26 The adjusted R-square increases from .06 before the MCA to .67 after the MCA. All of this improvement emerges in the period after February 1984, when the ad justed R-square increases further to .83.27 More over, the reciprocal of the estimated coefficient on TR is .124, very close to the marginal re serve requirement of .12 during the latter peri od. Indeed, the null hypothesis that this coeffi cient is equal to 1/.12 cannot be rejected at the 5 percent significance level (the t-statistic is 0.62). 27The switch from lagged to contemporaneous reserve ac counting in February 1984 might explain some of this ap parent improvement. To account for this possibility, the change in TCD was regressed on both the contemporan eous and lagged change in TR. In no case was the coeffi cient on lagged TR statistically significant from zero at the 5 percent level. Indeed, the results differed little from those reported in table 3. That the switch from lagged to contemporaneous reserve requirements is of no significant consequence is consistent with the conjecture of Thornton (1983) and the empirical evidence presented by Garfinkel and Thornton (1989). The relationship between TR and TCD will likely become even stronger given the recent elimination of reserve requirements on all time and sav ings deposits. 59 Table 3 Regression of Changes in TCD on Changes in Total Reserves Period Constant Total reserves SEE R2 D.W. 1/1970-12/1980 .795* (5.72) 1.870* (3.10) 1.418 .062 1.35 1/1981-11/1990 .742* (3.42) 7.264* (15.64) 2.081 .674 2.23 1/1981-2/1984 1.765* (3.81) 2.704 (1.92) 2.414 .067 2.06 3/1984-11/1990 .441* (2.09) 8.082* (19.90) 1.676 .832 1.94 * indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. Table 4 Regression Estimates of Changes in the Multiplier on Changes in Total Reserves Period Constant Total reserves SEE R* D.W. 1/1970-12/1980 - .003* (2.86) -.0 0 3 (0.67) .011 .004 1.70 1/1981-11/1990 - .004* (3.28) .016* (6.77) .011 .275 2.12 1/1 981 -2/19 84 .002 (0.77) -.0 0 3 (0.42) .014 - .0 2 3 1.82 3/1984-11/1990 -.0 0 6 * (6.05) .020’ (11.07) .007 .603 2.31 * indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. The E ffec ts o f P o lic y A ction s on the M u ltiplier a n d th e M o n ey S to ck The above analysis suggests that policy actions could exert a strong effect on the multiplier in the 1980s. Table 4 shows that this is the case. Changes in TR account for 60 percent of the variation in the multiplier since March of 1984. The table also shows that, because of the loose link between reserves and total checkable de posits, the assumption that policy actions had no effect on the multiplier was a reasonable working assumption before the adoption of the MCA. Indeed, changes in the multiplier are uncorrelated with changes in TR during the pe riod ending December 1980. These results suggest that there should be a dramatic change in the relationship between Ml and TR in the 1980s. Simple regressions of changes in M l on changes in TR and changes in JULY/AUGUST 1991 60 Table 5 Regression Estimates of the Change in M1 on the Change on Total Reserves and the Change in the Adjusted Monetary Base Period Constant Total reserves 1/1970-12/1980 1.328* (8.86) Adjusted monetary base 2.079* (3.20) -.1 8 5 (0.74) 1/1981-11/1990 1.868* (8.34) -.2 8 0 (0.53) 1 /1 981 -2/19 84 2.613* (5.33) 1.113 (1.51) 3/1984-11/1990 1.696* (7.73) 1.839* (3.20) 7.995* (18.89) -1 .0 2 3 (1.43) 3.158* (7.48) 1.28 1.312 .312 1.54 .659 2.13 2.939 .361 1.39 .070 2.00 2.369 .200 1.86 1.746 2.910 (1.95) .066 2.554 2.773* (8.23) D.W. 2.149 7.251 * (15.12) R! 1.529 2.513* (7.73) SEE .817 1.73 3.138 .407 1.33 * indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. the AMB, reported in table 5, bear this out.28TR explains a relatively small amount of the varia tion in changes in Ml before MCA and over 80 percent of the variation of changes in Ml since early 1984. The table also shows that the ex planatory power of the monetary base has in creased since the MCA, as would be expected.29 Nonetheless, the explanatory power of the AMB declined significantly relative to that of TR. IM PLIC ATIO N S FOR M O NE TAR Y PO LICY Prior to the MCA, when it appeared that the multiplier was independent of policy actions, 28Again, the Durbin-Watson statistics indicate significant serial correlation, especially when the AMB is used as the independent variable. In no case did an adjustment for seriai correlation using a maximum likelihood technique alter any of the substantive results presented in table 5. That is, these results too suggest that there is a marked increase in the explanatory power of TR in the 1980s and FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis a rather simple, straightforward approach to money stock control was implied—namely, to target the level of the adjusted monetary base consistent with a money-stock target conditional on a forecast of the multiplier, where the multi plier forecast was not conditional on the target setting for the monetary base. This notion also implied that the adjusted monetary base is the best indicator of the effects of policy actions on the money stock. The realization that the multiplier is not in dependent of policy actions suggests that the monetary base might not be the best indicator of policy actions on the money stock and that revising the simple empirical models of the that changes in TR explain much more of the variation in changes in M1 in the 1980s than do changes in the AMB, even allowing for significant first-order serial correlation. 29See Garfinkel and Thornton (1989) for a discussion of this point. 61 money supply process to account for the effects of policy actions on the multiplier could result in improved money stock control. These issues are discussed briefly in this section. The A d ju ste d M o n eta ry B ase as an In d ica to r o f P o lic y A ction s on the M o n ey S to ck The adjusted monetary base continues to reflect all policy actions—changes in both re serves and reserve requirements; however, it does not fully capture the effects of these ac tions on the money stock. Indeed, changes in Ml are now more closely linked to changes in TR than to changes in the AMB. Consequently, it now appears that total reserves, adjusted for reserve requirement changes, is a better indica tor of the effects of monetary policy actions on the money supply than is the adjusted monetary base. Furthermore, the quantity of currency out standing is demand-determined. Consequently, unlike adjusted reserves, the adjusted monetary base can give misleading signals of the course of monetary policy when there are exogenous shifts in the demand for currency. To take a concrete example, currency growth accelerated markedly beginning about December 1989.30 This acceleration was accompanied by a sharp acceleration in the growth of the ad justed monetary base from 3.4 percent in 1989 to 8.4 percent in 1990. Such a sharp rise in base growth would tend to indicate that mone tary policy had eased. But the growth of ad justed reserves and, thus, TCD indicate a sub stantially weaker easing of policy. TCD increas ed at a 1.2 percent rate in 1990 compared with a -1.3 percent rate in 1989. Of course, the ap parent exogenous increase in the demand for 30While the exact cause of this acceleration remains un clear, many attribute it (at least in part) to currency exports to South American and Eastern bloc countries. 31An equally interesting, but less frequently discussed, epi sode occurred during 1989 when, after remaining fairly constant, currency growth slowed abruptly. During this period, the K-ratio rose rather than fell, as one might ex pect given the apparent shift in the demand for currency. The increase in the K-ratio was driven by negative growth in reserves and, hence, TCD during this period. 32See Balbach (1981), Hafer, Hein and Kool (1983), and Johannes and Rasche (1979, 1987) for a discussion of this approach and for alternative methods that have been used to forecast the money multiplier. 33Note that the multiplier approach can be more difficult to implement. Most notably, the control problem becomes currency caused the K-ratio to rise and the mul tiplier to fall, so that Ml grew slowly relative to the monetary base during the period.31 Because there is now a closer link between TR and Ml than between the AMB and Ml and because TR is less likely to give misleading signals, TR is likely to be a better indicator of both monetary policy and the effects of policy changes on the money stock. The M u ltiplier A p p ro a ch to M on ey S to ck C on trol That the multiplier is not independent of policy actions also has important implications for the multiplier approach to money stock con trol. Taking this approach, the target level of Ml is achieved by forecasting the multiplier, then supplying the amount of the adjusted mon etary base necessary to hit the desired Ml tar get.32 If, however, the multiplier is a function of open market operations, policymakers must also predict the effect of their actions on the multi plier. That is to say, the multiplier approach to money control should be modified to take ac count of the effects of policy actions on the multiplier. Taking account of such effects un doubtedly will improve money control over the simple approach that assumes independence be tween the multiplier and policy actions. The magnitude of this improvement depends on how accurately the effects of policy actions on the multiplier can be forecast. To the extent that variations in the multiplier are largely explained by variations in the k-ratio and these variations are, in turn, largely influenced by policy (espe cially in the post-MCA period), such a modifica tion could produce a substantial improvement in money stock control.33 nonlinear. One alternative approach would be to simply forecast the level of currency, then supply the reserves necessary to hit a target level of TCD. The target level of TCD would have to be consistent with both the M1 target and the forecast level of currency. Whether this or the multiplier approach, suitably modified to account for the effect of policy actions on the multiplier, would provide greater monetary control is an empirical issue well beyond the scope of this paper. Both approaches will produce forecast errors when there are unexpected shifts in the de mand for currency. The real issue is whether better esti mates of the K-ratio can be obtained by estimating the numerator and denominators separately than estimating them together. This is an empirical issue. Nevertheless, this alternative approach could be simpler to implement and might provide superior control if reasonably accurate forecasts of currency can be made. JULY/AUGUST 1991 62 SUM M AR Y This article has examined closely the standard multiplier model of the money supply process, specifically questioning the view that the ad justed monetary base multiplier is independent of the policy actions of the central bank. Be cause the demand for currency depends on a number of factors that are unrelated to the de mand for checkable deposits (and vice versa) and because the stock of checkable deposits has been more closely tied to the quantity of re serves supplied by the Federal Reserve since the implementation of the MCA, changes in mone tary policy result in changes in the ratio of cur rency to checkable deposits and, consequently, changes in the multiplier. Hence, the Federal Reserve’s monetary policy actions are reflected both in the adjusted monetary base and the money multiplier. Theoretical considerations suggest that the multiplier has never been independent of policy. The elimination of reserve requirements on some non-transaction accounts and the exten sion of Federal Reserve reserve requirements to all depository institutions has greatly increased the association between checkable deposits and reserves. These changes have increased signifi cantly the association between changes in mone tary policy actions and changes in the multiplier. That the multiplier is affected by policy actions suggests that money stock control using the multiplier model would be enhanced by taking the effect of policy actions on the multiplier into account. How much improvement can be expec ted with this modified approach and how effec tive alternative approaches to monetary control can be is left as a topic for further research. REFERENCES Avery, Robert B., George E. Elliehausen, Arthur B. Kennickell, and Paul A. Spindt. "The Use of Cash and Transaction Accounts by American Families.” Federal Reserve Bulletin (February 1986), pp. 87-108. _____ . “ Changes in the Use of Transactions Accounts and Cash for 1984 to 1986.” Federal Reserve Bulletin (March 1987), pp. 179-96. Balbach, Anatol B. “ How Controllable is Money Growth?,” this Review (April 1981), pp. 3-12. Belongia, Michael T., and James A. Chalfant. “ The Chang ing Empirical Definition of Money: Some Estimates from a Model of the Demand for Money Substitutes,” Journal of Political Economy (April 1989), pp. 387-97. Brunner, Karl. “A Schema for the Supply Theory of Money,” International Economic Review (January 1961), pp. 79-109. http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis Brunner, Karl, and Allan H. Meltzer. “ Some Further Investigations of Demand and Supply Functions for Money,” Journal of Finance (May 1964), pp. 240-83. ______ “ Liquidity Traps for Money, Bank Credit, and Interest Rates,” Journal of Political Economy (January/February 1968), pp. 1-37. _____ . “ Money Supply,” in B. M. Friedman and F. H. Hahn eds., Handbook of Monetary Economics (Elsevier Science Publishers, 1990). Cagan, Phillip. “ The Demand for Currency Relative to the Total Money Supply,” Journal of Political Economy (August 1958), pp. 303-28. Carraro, Kenneth C., and Daniel L. Thornton. “ The Cost of Checkable Deposits in the United States,” this Review (April 1986), pp. 19-27. Dotsey, Michael. “ The Demand for Currency in the United States,” Journal of Money, Credit and Banking (February 1988), pp. 22-40. Fisher, Douglas. Money Demand and Monetary Policy (The University of Michigan Press, 1989). Fisher, Irving. The Purchasing Power of Money (Augustus M. Kelley, 1911). Garfinkel, Michelle R., and Daniel L. Thornton. “ The Link Between M1 and the Monetary Base in the 1980s,” this Review (September/October 1989), pp. 35-52. _____ . “ Measuring the Monetary Base,” mimeo (February 1991). Hafer, R. W., Scott E. Hein and Clemens J. M. Kool. “ Fore casting the Money Multiplier: Implications for Money Stock Control and Economic Activity,” this Review (October 1983), pp. 22-33. Hess, Alan C. “An Explanation of Short-Run Fluctuations in the Ratio of Currency to Demand Deposits,” Journal of Money, Credit and Banking (August 1971), pp. 666-79. Johannes, James M., and Robert H. Rasche. "Predicting the Money Multiplier,” Journal of Monetary Economics (July 1979), pp. 301-25. _____ . Controlling the Growth of Monetary Aggregates (Kluwer Academic, 1987). Mishkin, Fredric S. The Economics of Money, Banking and Financial Markets, 3rd. ed. (Little, Brown and Company, 1989). Plosser, Charles I. "Money and Business Cycles: A Real Business Cycle Interpretation,” in Michael T. Belongia, ed., Monetary Policy on the 75th Anniversary of the Federal Reserve System (Klumer Academic Publishers, 1991). Sill, Keith. “An Empirical Examination of Money Demand in an Intertemporal Optimizing Framework,” unpublished paper (University of Virginia, 1990). Stone, Courtenay C., and Daniel L. Thornton. “ Solving the 1980s’ Velocity Puzzle: A Progress Report,” this Review (August/September 1987), pp. 5-23. Tatom, John A. “ Issues in Measuring An Adjusted Monetary Base,” this Review (December 1980), pp. 11-29. Thornton, Daniel L. ’’Simple Analytics of the Money Supply Process and Monetary Control,” this Review (October 1982), pp. 22-39. _____ . “ Lagged and Contemporaneous Reserve Accoun ting: An Alternative View,” this Review (November 1983), pp. 26-33. Thornton, Daniel L., and Courtenay C. Stone. “ Financial Innovations: Causes and Consequences,” in Kevin Dowd and Mearyn K. Lewis, eds., Current Issues in Monetary Analysis and Policy, (MacMillan Publishers, 1991). 63 Appendix A Model of the Money Supply Process One might be tempted to interpret the regres sion equations in the text as representing alter native models of the money supply process; how ever, the reader is cautioned not to do so. In deed, as the article suggests, some existing mod els of the money supply process are misspecified. This appendix illustrates the bias of some of the regression equations estimated in the article. The discussion in the paper suggests that, since the MCA, there is a very simple linear relation ship between TCD and TR of the form TCD, = a + bTR, + et, where the coefficients a and b are constants and e is a residual error that is assumed to be white noise. The error term arises because some reserves are held against transaction deposits not included in TCD and because depository in stitutions hold excess reserves. The constant term, a, enters the equation because a lower reserve requirement for a tranche of checkable deposits exists and because some of the vari ables omitted from this equation might have non-zero means. If TR is correctly adjusted for changes in reserve requirements, including the annual change in the deposit tranche, then the coefficients a and b should be constant, where b is the reciprocal of the marginal reserve requirement—that is, b = 1/.12 = 8.33. The discussion and the empirical evidence in the paper further suggest that currency holdings are independent of TCD, so that C, is simply ex ogenous from the perspective of money stock control. If this representation is true, then a regres sion of Ml on the monetary base is misspecified, because it imposes a restriction that is in consistent with the process generating the data. To see this, consider the following regression specification: (A.2) C, + TCD, = g + hTR, + jC, + q,. With the restriction h = j, equation A.2 is iden tical to equation A.I. The above analysis, how ever, suggests that the coefficient h should equal 8.33 and the coefficient j should equal 1. If this is the case, imposing the restriction that these coefficients are equal will be resoundingly rejected by the data. To test this hypothesis, first-difference specifi cations of equations A.l and A.2 are estimated using monthly data for the period from March 1984 through November 1990. These estimates use Federal Reserve Board data for the adjusted monetary base and total reserves, adjusted for changes in reserve requirements. These data come close to satisfying the identity that the monetary base is equal to the currency compo nent of the money supply plus total reserves. These estimates are presented in table A.l. In the unrestricted version of the equation, neither the null hypothesis that h = 8.33 nor the null hypothesis that j = l can be rejected at the 5 percent significance level. The t-statistic for the test that h = 8.33 is .59 and the t-statistic for the test that j = l is .25. Hence, it is not surprising that the restriction that h = j is soundly rejected by the data. It is interesting to note that imposing this restriction biases the coefficient of the mone tary base multiplier away from its true value. The estimated multiplier of 4.005 is nearly 50 percent larger than its average value during this period. This bias emerges because of an omitted variable. To see this, note that equation A.2 could be rewritten as either (A.3) Ml, = g + hMB, + (j-h)C, + q, or (A.l) M l, = g + sMB, + q,. (A.4) Ml, = g + jMB, + (h-j)TR, + q,. Given the definitions of Ml and MB, this equa tion can be rewritten as Hence, equation A.l can be obtained by omit ting C,from equation A.3 or TR, from equation JULY/AUGUST 1991 64 Table A.1 Estimates of Equations A.1 and A.2 Constant AAMB -2 .0 9 9 * (2.49) 4.005* (7.48) .590 (1.21) ATR AC D.W. .407 1.10 1.690 .908* (2.51) R2 3.138 8.090* (19.67) SEE .828 1.93 'indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. Table A.2 Estimates of Regression of ATCD on ATR and AC: March 1984 - November 1990 Constant .581 (1.19) ATR 8.069* (19.67) AC SEE R* D.W. -.1 1 5 (0.32) 1.685 .830 1.93 'indicates statistical significance at the 5 percent level. Absolute values of t-statistics are in parentheses. A.4. In the former case the estimate of h is bias ed downward (4.005 vs. 8.33); in the latter case the estimate of j is biased upward (4.005 vs. 1). Furthermore, the equation exhibits serial cor relation, a common indicator of misspecification. These results are not too surprising given that the demand for currency appears to be indepen dent of the demand for TCD, as illustrated in table A.2, which shows the results of a regres sion of changes in TCD on changes in TR and changes in C. The coefficient on the change in C is negative, indicating a substitution between http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS FEDERAL Federal Reserve Bank of St. Louis TCD and currency, but is not significantly dif ferent from zero. Given this independence, it is hardly surprising that regressions of changes in Ml on TR and changes in TCD on TR produce nearly identical results. Comparing the results in table A.l with those in table 5 shows that the coefficient is biased downward slightly when Ml is regressed on TR. This occurs because Ct is omitted from the right-hand side of the equa tion and because of the weak negative associa tion between changes in both Ct and TCDt and, hence, changes in TRt. 65 Cletus C. Coughlin and Thomas B. M andelbaum Cletus C. Coughlin is a research officer and Thomas B. Mandelbaum is an economist at the Federal Reserve Bank of St. Louis. Thomas A. Pollmann provided research assistance. Measuring State Exports: Is There a Better Way? Te „ RISING LEVEL OF U.S. EXPORTS in re cent years has caused jobs and incomes in many states to become more closely tied to ex ports. To assess the economic effects of state exports, it is essential to have reliable informa tion on the level of export activity by firms with in the individual states. Such information is essential for numerous other purposes as well. For example, policymakers and others interested in state economic development require export data to assess the effectiveness of programs de signed to stimulate export activity; they also re quire such data to assess the effects of trade policy changes, such as the proposed free trade agreement with Mexico.1 Unfortunately, no ideal measure of state export activity currently exists. This article describes the two available state export series and compares their estimates of manufactured exports. Such a comparison was 1As reported in Business America (1991), state governments engage in a wide variety of activities to promote exports. These activities include overseas trade missions, technical assistance (such as seminars on the legal and financial aspects of trade), and the dissemination of trade leads. Seven states have export finance programs; 41 states maintain offices in 24 countries to promote trade. These promotional activities raise the issue of whether interna tional exports by firms within a state generate different economic results than domestic exports (or exports to other states by firms within a state). Empirical evidence to not possible until recently because the two se ries were not available for the same year. Our comparison for 1987 reveals that the two series provide conflicting information about export ac tivity in many states. The most prominent deficiency of both mea sures is that they are based on the value of ex port shipments by firms within a state rather than on the value of goods produced within a state that are exported. While this distinction may sound arcane, the discussion below indicates that it is not. Moreover, income and employ m ent in a state a re dep en d en t on the latter measure, not on the value of export shipments. To address this deficiency, a third estimate of state manufactured exports is developed in this article. Comparisons show the differences be tween this new measure and one of the existing measures of export activity. Such a comparison further illuminates the shortcomings of the two assess whether such a distinction is meaningful in an economic sense is scarce. See Webster et al. (1990) for evidence that the employment effects of international ex ports exceed those of domestic exports for many industries. JULY/AUGUST 1991 66 available series and the advantages of a series like that developed here. EXISTING MEASURES OF STATE EXPORTS Historically, the focus of U.S. trade data has been on country-to-country trade flows (that is, U.S. exports to and imports from individual countries). Recently, increasing attention has been focused on trade flows involving individual states. Exports of Boeing aircraft from Wash ington and imports of foreign cars by Missouri residents are just two examples of traded goods that have attracted attention to the fact that state jobs and incomes are related to the inter national economy. Our focus is restricted to ex port activity at the state level. To date, those in terested in the magnitude of these flows have relied on two data sources published by the U.S. Census Bureau: Exports From Manufacturing Establishments (EME) and the Origin o f Move ment o f Commodities (OMC). E x po rts F rom M anu factu ring E sta b lish m en ts Approximately 56,000 of 220,000 manufactur ing establishments are asked in the Annual Sur vey of Manufactures to report the total value of products shipped for export.2 Since many estab lishments do not know the final destination of their products, the reported exports understate the value of all manufacturing export shipments. To compensate, the total amounts reported are adjusted to include estimates of exports by other distributors, such as wholesalers. 2At five-year intervals, a more comprehensive coverage of manufacturing establishments occurs with the Census of Manufactures. See appendix A of U.S. Census Bureau (1991) for details on the 1987 Census of Manufactures and the 1986 Annual Survey of Manufactures. 3For details, see appendix C of the U.S. Census Bureau (1991). 4A Shipper’s Export Declaration must be filed for all export shipments except for those going to Canada. Effective November 30, 1990, this document was no longer required for Canadian shipments because of a decision to substitute Canadian import statistics for U.S. export statistics. See Ott (1988) for an explanation why Canadian import data are considered more accurate than U.S. ex port data. 5Processed food, forestry, petroleum and coal products that originate in these primary sectors are included as manufactured exports. 6Such decisions are also complicated by the fact that, until 1987, export data were generally unavailable at the three- FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis Differences between the directly reported values and the national total derived from Ship per’s Export Declarations are allocated to states.3 A Shipper’s Export Declaration is a document that exporters must file which includes the value of each export shipment.4 The allocation procedure is complicated slightly because the in dustry classification scheme used in Shipper’s Export Declarations differs from that used in the Annual Survey of Manufactures. An additional complication is that the value of export ship ments in Shipper’s Export Declarations includes freight and wholesale margins. Since the value of export shipments in the EME is reported as shipments leave the plant, the costs associated with transportation and wholesaler activity must be removed from the values reported in Ship per’s Export Declarations. EME was first produced in 1960 as the Origin o f Exports o f Manufactured Products. It was pro duced at varying intervals until it became an annual report in 1983. This series possesses some significant shortcomings. First, the series is restricted to manufactured exports. It pro vides no information for establishments engaged in exporting services or unprocessed commodi ties produced by the agricultural, mining, fores try and fishing sectors.5 Second, this series is available with a two-year or more delay. For example, data for 1985 and 1986 became available in early 1989 and data for 1987 became available in 1991. Many ana lysts view these data as having only historical value because information on recent activity is not available for use in current decisions, such as those involving targeting export promotion expenditures.6 digit Standard Industrial Classification (SIC) level. The SIC is the standard by which establishment-based U.S. govern ment economic statistics are classified by industry. For details, see U.S. Office of Management and Budget (1987). For manufacturing, 20 industries are identified at the two-digit SIC level. The industry becomes more nar rowly defined as the number of digits for an SIC level in creases. Prior to 1987, the export data were presented at the two-digit SIC level, or only for broad industries. An ex ample of the disaggregation offered by the use of threedigit SIC codes is chemicals and allied products (SIC 28) which has eight industry groups: industrial inorganic chemicals (SIC 281); plastics materials and synthetics (SIC 282); drugs (SIC 283); soaps, cleaners and toilet goods (SIC 284); paints and allied products (SIC 285); industrial organic chemicals (SIC 286); agricultural chemicals (SIC 287) and miscellaneous chemical products (SIC 289). 67 The final and most important shortcoming is that this series reports the value of shipments instead of what is termed "value added.” Value added is the value of a firm's sales minus the value of the goods and services it purchases from other firms to make its products. As the term implies, value added measures the dollar value a firm adds to the value of purchased in puts in its production process. One way to calculate the market value of final goods and services produced during a year is to sum the value added at each stage of pro duction by the firms in an economy.7 To illus trate, assume an automobile producer had total sales of $18 billion, of which $10 billion reflect the value of steel, tires, plastics, electricity and other inputs used by the producer to make auto mobiles. The cost of these intermediate inputs is subtracted from the producer’s revenue to cal culate value added, so the automobile pro ducer’s value added was $8 billion. This pro cedure is repeated for each firm in the econ omy; the sum of all firms’ value added equals the total value of production within an economy. Using the value of export shipments rather than the value added related to exports might be a misleading indicator of export activity in a state. Some manufactured products are not ex ported directly, but are combined as inputs with other resources to produce an export. If these inputs were produced in one state and trans ported to another for final processing, the value of export shipments from the latter manufactur ing establishment in the exporting state would overstate the value added that actually occurred in that state.8 The value of export shipments in cludes value added in both states. A state’s value of export shipments will exceed its export value added if its exporting firms rely heavily on inputs produced elsewhere or if its TThe value added approach is one of three standard methods for calculating the market value of production. The other methods focus on income and expenditures. The income approach sums the incomes derived from economic activity, which are primarily wage, profit and in terest incomes from employment of labor and capital resources. The value added in an establishment is the in come generated by the establishment’s activity. The ex penditure approach sums four general categories of spen ding on goods and services: consumption, investment, government and net exports (that is, exports minus im ports). The income and expenditure approaches are used more extensively in the United States than the value add ed approach. In the European Community, however, the value added approach is used extensively in the ad ministration of taxes. firms produce relatively few inputs used by ex porting firms in other states. On the other hand, the value of export shipments from a state will fall short of its export value added if its export ing firms produce more inputs that are used by exporting firms in other states than its firms purchase from elsewhere. Export value added and the value of export shipments will only be the same if the value of shipments used to pro duce exports in other states exactly offsets the value of inputs from other states that are used to produce exports. Overall, a state’s value of export shipments may overstate, understate or equal the value added that actually occurred in the state. The empirical importance of this dif ference is examined below. O rigin o f M o vem en t o f C o m m o d ities Prompted by a request from the transporta tion industry, a second export series, the OMC, began in 1987. The goal of this series is to iden tify where merchandise begins its export jour ney so that it can be tracked to its port. In the case of a manufactured good, the so-called “point of origin” does not require that the location of production of all component parts be identified, but rather where a completed manufactured good began its export journey. According to the instructions that accompany the Shipper’s Ex port Declaration, the point of origin could be any of the following: 1) the state in which the merchandise actually began its journey to the port of export (indicated by the two-digit U.S. Postal Service abbreviation); 2) the state of origin of the commodity with the greatest share of value in a bundle of exports; or 3) the state of consolidation (the state where goods are con solidated by an intermediary for overseas ship ment). In practice, the ports from which goods are shipped overseas are frequently used to iden tify the point of origin. This discretion in identi8Although such intermediate products are identified by the state of production as “ supporting exports” in the EME, the state from which they are ultimately exported is not in dicated. In addition, adding a state’s supporting exports to its final shipments would result in some “ double counting” of exports and overstate the value added associated with manufactured exports. Note that the national value of ex port shipments is a theoretically appropriate measure of value added because the sum of export shipments across all establishments does measure the market value of these manufactured exports. At the national level, there is no double counting. The shipments of intermediate inputs used for the exports are already included in the value of export shipments and are not added again in the calculation of manufactured exports. JULY/AUGUST 1991 68 fying the point of origin reflects the fact that determining the location of production is not a primary objective of this data series.9 Origin of movement totals are determined by sorting Shipper's Export Declarations by the state where a commodity became an export. A problem, however, is that Declarations for many shipments contain no point of origin. For exam ple, in 1987 about 25 percent of the 9.7 million Declarations for shipments contained no state code. To make the data more useful, the Census Bureau contracted with the Massachusetts In stitute for Social and Economic Research to de velop estimates for the origin of shipments lack ing state codes.10 This expanded series is used in the following discussion and is referred to as the OMC series. This newer series has some desirable charac teristics relative to the export data provided di rectly by manufacturing establishments in the EME, although the older series is generally viewed as the more reliable of the two series.11 One attractive feature of the OMC is that the data are available with a lag of months rather than years. In addition to manufactured ex ports, shipments data on nonmanufactured m er chandise exports are provided. The initial for eign destination of these goods is provided as well. Consequently, information about state-tocountry export flows is available for the first time. Like the EME, however, this series does not approximate the extent of value added in a state resulting from manufactured exports. A C O M PAR ISO N OF THE T W O SERIES While there are several reasons why the two export series might differ, it is possible that the actual differences are small enough to allow the data to be used interchangeably. No comparison of the two series has been possible previously because 1987 was the first year for which the 9Smith (1989, 1990) notes that identifying the production locations of exported goods is especially difficult for agricultural and mined commodities. Small shipments of these commodities are often combined at storage facilities prior to reaching their port of embarkation. Shippers tend to report either the state of consolidation or the port as the state of origin. 10Details on the methods to generate these estimates can be found in Lerch (1990). FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis data in the OMC were available and the 1987 EME was just released in April 1991. First, we compare each state's level of exports as in dicated in the EME and the OMC. Next, we in vestigate whether a state’s rank differs between the two measures. To complete the analysis, a particular facet of the linear relationship be tween the two export series is examined. See appendix A for details on the three methods used to compare the two 1987 series, as well as the two 1986 export series discussed later. If the two series are closely related, then the OMC data, which are available after a considerably shorter lag, could be used in place of the EME data. C om parin g Levels: 1 98 7 E xport S eries Table 1 shows the value of 1987 manufac tured exports according to the two series for the 51 "states” (50 states and the District of Co lumbia) and the total of the states. One reason the two series differ is that the data in the OMC include transportation costs and wholesale m ar gins, while the data in the EME are the value of exports at the producing plant. This accounts for the bulk of the $22.3 billion excess of the OMC state total over the EME state total in table l.12 If these items were the only source of dif ference, the export value in the OMC for each state would be higher than the value in the EME. Also, the difference would be greater for those states farthest from major ports or a for eign border, reflecting the higher transportation costs. Table 1 shows that exports according to the OMC are higher than the level according to the EME in just 20 of the 51 states. This is in sharp contrast to the expectation that the OMC measure should be higher based on differences in its coverage and on the difference in the state totals. This discrepancy occurs primarily because of the OMC’s focus on where merchandise began its export journey. Since this location is often "S e e Farrell and Radspieler (1990) and Little (1990). ,2Appendix B of the Exports from Manufacturing Establishments: 1987 shows the difference between the value of exports at the port of export and the estimated plant value to be $28 billion. 69 T a b le 1 Manufactured Exports by State for 1987 1987 E xports' EME OMC State Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total Differences Levels2 Percentage3 Rank EME OMC $2,138.6 889.6 2,086.0 1,353.0 22,996.1 1,818.4 4,741.1 519.7 100.7 4,803.0 3,561.1 175.3 765.8 8,687.8 5,001.1 2,552.6 1,858.2 2,906.4 3,408.7 779.7 1,927.9 6,347.9 12,412.0 4,733.3 1,712.1 5,148.6 135.7 768.7 170.3 1,160.4 3,982.5 206.0 11,824.0 5,670.6 222.6 13,041.1 1,355.4 2,121.1 6,717.3 691.9 3,234.7 270.3 3,567.2 14,046.3 589.1 598.6 3,656.6 10,841.7 1,126.2 4,108.5 39.5 $1,896.8 1,516.0 2,772.7 636.1 30,448.7 1,623.4 3,096.9 842.0 265.1 9,602.8 3,380.1 153.5 462.0 8,471.5 4,102.9 1,756.3 1,448.2 1,930.9 5,865.2 636.1 1,881.1 8,093.9 17,618.2 3,850.1 1,122.3 2,851.0 167.0 693.5 356.8 835.0 6,347.6 148.9 17,614.8 4,898.0 217.4 8,991.1 991.3 2,294.6 5,734.5 408.5 2,159.6 56.0 2,309.3 22,662.3 757.0 704.3 4,750.8 11,793.9 920.0 3,500.4 227.2 $ -2 4 1 .8 626.4 686.7 -7 1 7 .0 7,452.6 -1 9 5 .0 - 1,644.2 322.3 164.4 4,799.8 -1 8 1 .0 -2 1 .8 -3 0 3 .8 -2 1 6 .3 -8 9 8 .2 - 796.3 -410.1 -9 7 5 .5 2,456.5 -1 4 3 .6 -4 6 .8 1,746.0 5,206.2 - 883.2 -5 8 9 .8 -2,297.6 31.3 -7 5 .2 186.5 -3 2 5 .4 2,365.1 -57 .1 5,790.8 - 772.6 - 5 .2 -4,0 5 0 .0 -364.1 173.5 -9 8 2 .8 -2 8 3 .4 -1,075.1 -2 1 4 .3 - 1,257.9 8,616.0 167.9 105.7 1,094.2 952.2 - 206.2 -608.1 187.7 -1 1 .3 % 70.4 32.9 -5 3 .0 32.4 -1 0 .7 -3 4 .7 62.0 163.3 99.9 -5 .1 -1 2 .4 -3 9 .7 - 2 .5 -1 8 .0 -3 1 .2 -22.1 -3 3 .6 72.1 -1 8 .4 - 2 .4 27.5 41.9 -1 8 .7 -3 4 .4 -4 4 .6 23.0 - 9 .8 109.5 -2 8 .0 59.4 -2 7 .7 49.0 -1 3 .6 - 2 .3 -3 1 .1 -2 6 .9 8.2 - 14.6 -4 1 .0 -3 3 .2 -7 9 .3 -3 5 .3 61.3 28.5 17.7 29.9 8.8 -1 8 .3 - 14.8 475.2 25 36 27 33 1 30 14 43 50 13 20 47 39 7 12 24 29 23 21 37 28 9 4 15 31 11 49 38 48 34 17 46 5 10 45 3 32 26 8 40 22 44 19 2 42 41 18 6 35 16 51 26 30 21 41 1 29 19 35 45 6 18 49 42 8 15 28 31 25 11 40 27 9 3 16 32 20 48 39 44 36 10 50 4 13 47 7 33 23 12 43 24 51 22 2 37 38 14 5 34 17 46 $193,571.0 $215,863.5 $22,292.5 11.5 — — Sources: EME\ U.S. Department of Commerce, Bureau of the Census, Exports from Manufacturing Establishments: 1987 (GPO, 1991). OMC: Massachusetts Institute for Social and Economic Research, University of Massachusetts, "U.S. Exports by State of Origin of Movement,” data tape (1990). ’ Millions of dollars. 2OMC value minus EME value. H(OMC - EME)!EME)'\ 00. JULY/AUGUST 1991 70 identified as a port, the OMC estimates of ex ports are more concentrated in states that con tain, or are near, major ports. Also, in some states where transportation costs might be ex pected to be relatively high, such as Nebraska and Kansas, the export value in the OMC is lower than the value in the EME, again con trary to the expectations based on transporta tion costs alone. The importance of ports in the OMC data is further illustrated in figure 1, which plots the level of exports in each series. If each state's ex ports were identical in both series, all points would fall on the line labeled "line of equality.” The points below the line of equality indicate that states’ exports in the OMC often are lower than reported in the EME. In seven of the states labeled in figure 1— California (CA), Florida (FL), Louisiana (LA), Michigan (MI), New Jersey (NJ), New York (NY) and Texas (TX)—the value using the OMC is much higher than the value using the EME. In these states, total exports using the form er measure exceed exports using the latter measure by almost $37 billion. This pattern is consistent with the fact that data in the EME in dicate the value of exports shipped from a state’s manufacturers, while the data in the OMC are more likely to indicate the value of exports ship ped from the state of consolidation or the port. Therefore, using the value in the OMC as a measure of a state’s export activity can be misleading. As table 1 shows, the percentage differences are also considerable for many states. For exam ple, the value of exports from Wyoming mea sured in the OMC is nearly six times higher than that reported in the EME, while the OMC esti mate for South Dakota is 79.3 percent lower. On average, the absolute value of the difference for a state is 44 percent; excluding Wyoming reduces the average difference to 35.3 percent. The overall correspondence between the dol lar levels of the two series is highlighted by cal culating the simple correlation between them. This measure ranges from negative one to posi tive one and equals one when the two measures are perfectly correlated. In the present case, the correlation of .95 is high. Thus, when a state’s OMC export value is higher than the average OMC export value using all states, the state’s EME export value also tends to be higher than the average EME export value. Figure 1 shows this general correspondence by plotting the states’ 1987 exports as indicated FEDERAL http://fraser.stlouisfed.org/RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis by the two series. Most observations cluster around the line of equality. Still, the substantial difference between the two series for several states indicates that the two measures are not identical. C om parin g R an ks: 1 987 E x po rt S eries Another useful comparison between the two series involves ranking the states to see if states with larger (smaller) export values using one measure also have larger (smaller) export values using the other measure. Ranks are often used as a summary measure of a state’s relative ex port performance. If each state had the same or, at least, a similar rank among the 51 states using either series, then the more current data in the OMC could be used to rank the states in a more recent year. In view of the high simple correlation between the two series, it is no surprise that table 1 indi cates a general similarity between a state’s ex port ranks using the two measures. California and Texas, for instance, rank first and second in both series. This general impression is cor roborated by the calculation of a Spearman rank correlation that allows for pairwise comparisons of the alternative proxies. This coefficient ranges from negative one to positive one; it equals one when measures yield identical rank ings and minus one when the rankings are iden tically inversely related. The correlation be tween the two series’ export ranks is .96, which is very close to one. Although this high correlation indicates a close overall correspondence between the rank ings according to the two series, policymakers or researchers who rely on the more current ranking available from the OMC as an indicator of the relative scope of export activity in a specific state can easily be misled. The ranking of each state in the OMC is not identical to the more reliable ranking in the EME. Florida and Louisiana, for example, rank considerably higher according to the OMC, due to the major ports in those states from which a large volume of merchandise is shipped. Missouri, on the other hand, ranks only 20th using the more current OMC measure, but is 11th according to the EME. 71 Figure 1 A Graphical Comparison of Two State Export Series: 1987 Billions of Dollars \ \ s ' ✓ Actual Linear / * Association y TX • ✓ \ / / / / Line o f Equality \ FL • l a . n j ✓/ / NY Ml / • • / / / / > / / • OH / & •V* 0 •• 5 10 15 20 25 Exports from Manufacturing Establishments (EME) A C loser L o o k a t th e L in ear A sso cia tio n B etw een th e 1987 E x po rt S eries A more rigorous criterion to assess the inter changeability of the two measures reveals a substantial difference between the two series. This criterion, termed difference preservation, requires that the two export series differ by no more than some constant across states. If this criterion is met, one export series could be reliably used as an index for the other. If the OMC data preserved the difference in the EME data, the association between the two series could be illustrated by a line indicating equality of a state’s exports, give or take some constant. In figure 1, such a line would be par allel to the line of equality. This is not the case, however. The dashed line, based on the actual linear association between the two series, is clearly not parallel to the line of equality. Con JULY/AUGUST 1991 72 sequently, one measure is not interchangeable for the other. This means that researchers and other users of state export data in statistical studies should not use one measure as a proxy for the other because the results can vary de pending upon which measure is used. In prac tice, this finding applies to the use of the more timely OMC-based measure as a proxy for the EME-based measure. A NEW STATE EXPORT MEASURE RASED ON VALUE ADDED Existing state export series indicate the value of export shipments rather than export value added. As such, they reflect both the value add ed in a state’s factories as well as the value add ed embodied in intermediate goods which may have been produced in other states. For exam ple, an airplane assembled and exported from the state of Washington may have components manufactured in California and Texas. Conse quently, these series fail to identify the true amount of state economic activity used to pro duce manufactured exports. To address this problem, we estimate a mea sure of each state’s value added associated with manufactured exports. In conjunction with the EME, the Census Bureau provides data for each state regarding the number of manufacturing workers producing manufactured exports in each industry as well as the number of nonmanufacturing workers in jobs related to the production of manufactured exports. In fact, ap proximately the same number of nonmanufac turing jobs as manufacturing jobs are related to manufactured exports. This reflects the fact that manufacturing requires the productive ef forts of workers (such as lawyers, accountants and transportation and communication workers) from various nonmanufacturing industries. Unlike the value of export shipments, the level of export-related employment is directly related to the value added of exports in a state; such employees directly generate the value add ed. This employment information is used to esti 13This is consistent with the Washington State input-output model for 1982 (Bourque, 1987). For example, in W ashington’s largest export sector, aerospace, inputs from other states equal 56.2 percent of the sector’s shipments. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis mate state export value added. The estimates are based on the assumption that the productiv ity (output) of each export-related employee is no different than the average w orker’s productivity in that industry and state. Consequently, export value added in a state is equal to the sum over all industries of the num ber of export-related employees in a state multiplied by their productivity. One data series necessary for such estimates, gross state product—the market value of the goods and services produced within a state dur ing a year—is not currently available for 1987. Since this precludes calculating export value added for 1987, our measure of exports is estimated for 1986. Appendix B provides a detailed discussion of the methodology used in estimating state export value added. C om parin g E x po rt Value A d d e d Fs. EME Figure 2 and table 2 compare the value of the newly constructed series of manufactured ex ports, export value added, with the value of state export shipments from the EME. Sum ming over all states, the export value added total ($196,656.2 million) exceeds the total in the EME ($159,374.5 million) by $37,281.7 million. Threefourths of this difference is due to transporta tion costs and trade margins that are included in our calculations, but are not in the EME total. The differences between the two measures at the level of individual states, however, reflect much more than transportation costs and trade margins. Rather, they reflect the fundamental distinction between value added and value of shipments accounts. The state of Washington is especially noteworthy as the level of export value added is approximately one-half the level of exports in the EME. This suggests that manu facturing export shipments from Washington contain a large percentage of intermediate in puts produced elsewhere.13 Using the export shipments value as a measure of this state's ex port activity is clearly misleading. 73 Figure 2 A Graphical Comparison of Two State Export Series: 1986 30 Billions of Dollars CA • Actual Linear Association / / s / T3 20 * a > T3 ■ a (0 a) D n > ro NY • / a x / in / / / '•TX OH / / / * ✓ / s / Line of Equality / • 10 NJ PA* / / / • WA 5 10 15 20 Exports from Manufacturing Establishments (EME) Using export shipments values also can cause inaccurate inferences in terms of understating a state’s export activity. For example, the export value added in 12 states exceeds the values in the EME by more than 50 percent. Wyoming, with an export value added that is more than nine times its E/WE-based export value, is by far the most extreme example. The primary reason is that firms in Wyoming process large quanti ties of oil and coal that are shipped to other states for use in manufactured exports. While Wyoming is a small exporter regardless of the measure used, the large percentage dif ferences are not restricted to relatively small ex porters. California, the nation’s leading exporter by both measures, is estimated to export 53.5 percent more on the basis of value added than it does on the basis of shipment data. Thus, California firms are supplying large amounts of goods and services ultimately exported in the form of manufactured exports from other states. These differences for many states between export value added and the EME-based measure of state exports raise the issue of the general association across all states between the mea sures. As was done above, the ranking of states' export value added was compared with the ranking of exports reported in the EME to JULY/AUGUST 1991 74 T a b le 2 Manufactured Exports by State for 1986______________ 1986 E xports' Value EME Added State Levels2 Rank Percentage3 EME Value Added 29 38 27 33 1 30 13 42 50 16 20 45 40 7 11 24 26 23 18 36 28 9 3 14 31 12 49 37 48 35 15 47 6 10 44 4 32 25 8 41 22 46 19 2 39 43 21 5 34 17 51 22 43 24 33 1 28 13 39 49 14 19 51 42 6 11 29 31 26 23 38 27 9 5 15 32 17 48 37 47 34 8 44 2 10 45 4 30 25 7 40 21 46 20 3 36 41 18 12 35 16 50 - — $1,684.9 712.9 1,755.8 1,065.4 17,216.4 1,477.7 3,996.4 429.5 91.0 3,372.6 2,826.7 214.3 502.6 7,209.2 4,787.4 1,932.4 1,835.0 1,939.8 3,020.3 800.6 1,740.5 5,513.8 10,878.0 3,691.9 1,337.1 4,267.9 101.2 753.3 167.1 892.6 3,548.1 177.7 9,412.4 5,260.8 214.7 10,653.0 1,084.6 1,862.7 6,026.6 481.9 2,398.0 212.7 2,910.4 10,981.5 668.5 384.0 2,704.0 9,862.8 983.2 3,313.5 19.1 $2,427.1 545.8 2,283.7 1,299.2 26,421.4 2,087.0 4,968.0 794.5 194.3 4,966.2 3,685.0 160.8 548.1 10,107.3 5,352.8 1,893.0 1,690.3 2,191.2 2,374.9 799.6 2,162.7 7,139.8 10,273.9 4,327.7 1,401.7 3,896.3 232.8 808.0 243.5 1,193.0 7,248.5 335.4 15,660.5 5,916.7 289.0 11,561.7 1,819.6 2,264.5 9,373.2 748.7 2,451.8 250.8 3,212.1 13,195.1 944.0 651.3 3,701.8 5,176.1 1,049.1 4,163.6 173.2 $742.2 -167.1 527.9 233.8 9,205.0 609.3 971.6 365.0 103.3 1,594.0 858.3 -5 3 .5 45.4 2,898.0 565.4 -3 9 .4 -1 4 4 .7 251.4 -6 4 5 .4 -1 .1 422.2 1,626.0 -604.1 635.8 64.5 - 371.6 131.6 54.7 76.4 300.4 3,700.0 157.7 6,248.0 655.9 74.3 908.7 735.0 401.8 3,347.0 266.8 53.8 38.1 301.7 2,214.0 275.5 267.3 997.8 -4,6 8 7 .0 65.9 850.1 154.1 44.1% -2 3 .4 30.1 21.9 53.5 41.2 24.3 85.0 113.5 47.3 30.4 -2 5 .0 9.0 40.2 11.8 - 2 .0 - 7 .9 13.0 -2 1 .4 -0 .1 24.3 29.5 - 5 .6 17.2 4.8 - 8 .7 130.1 7.3 45.7 33.7 104.3 88.7 66.4 12.5 34.6 8.5 67.8 21.6 55.5 55.4 2.2 17.9 10.4 20.2 41.2 69.6 36.9 -4 7 .5 6.7 25.7 806.9 $159,374.5 $196,656.2 $37,281.7 23.4 Alabama Alaska Arizona Arkansas California Colorado Connecticut Delaware District of Columbia Florida Georgia Hawaii Idaho Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina South Dakota Tennessee Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming Total Differences Sources: EME: U.S. Department of Commerce, Bureau of the Census, Exports from Manufacturing Establishments: 1986 (GPO, 1989). Export Value Added: Authors' calculations. 'Millions of dollars. 2Export Value Added minus EME value. 3((Export Value Added - EME)IEME)100. FEDERAL RESERVE BANK OF ST. LOUIS 75 determine whether states with larger (smaller) export values using one measure in 1986 also had larger (smaller) export values using the other measure in the same year. The two measures yield a general similarity between a state's export ranks. The Spearman rank cor relation is .98, which is virtually one. The ranks of a number of states, however, differ substantially across the two measures. Nine states have ranks that differ by five places or more. The largest changes involve Washing ton, which drops from fifth place using export shipments to twelfth place using export value added, and New Jersey and Alabama, both of which moved up seven places (New Jersey from 15 to 8 and Alabama from 29 to 22) when using value added. The overall correspondence between these two measures is also indicated by the simple correla tion between the two measures. The simple cor relation is .95. This close correspondence is evi dent when the two series for each state are plotted as in figure 2. Many observations cluster around the line of equality reflecting the linear association between the two measures. This strong association, however, does not mean that the two measures are interchange able. In terms of figure 2, the actual linear asso ciation (identified by the dashed line) varies sig nificantly in a statistical sense from the line of equality.14 Consequently, one measure is not a reliable proxy for the other when used in statistical studies. CONCLUSION Despite some improvement in available infor mation on state export activity in recent years, the two existing state export series are deficient in several ways. Their most important limitation is that they both measure the value of shipments and not the extent of a state’s economic activity (value added) related to manufactured exports. Nonetheless, as general indicators of export ac tivity across all states, the two measures provide similar information. Despite this overall similari ty, the two series can lead to substantially dif ferent conclusions when used for some states. Furthermore, the OMC series, which is available on a more timely basis, is not a satisfactory proxy for the more accurate EME data on ex port shipments according to the criterion used in this article. The estimate of a state’s export value added generated in this article is inherently superior to the existing measures available for assessing state export performance. This new measure can produce different conclusions than shipments-based data when used for some states or when used in statistical studies. Consequently, users should reconsider their use of the existing export series when they desire an accurate mea sure of a state’s value added related to manufac tured exports. The evidence presented here on export value added and its deviation from the EME-based ex port shipments measures is for one year only, however. The behavior of this discrepancy over time is unknown. This reinforces the importance of developing historical data on state export value added for analyses involving state export activity. REFERENCES Bourque, Philip J. The Washington State Input-Output Study for 1982, Graduate School of Business Administration, University of Washington (March 1987). Coughlin, Cletus C., and Phillip A. Cartwright. “An Examina tion of State Foreign Exports and Manufacturing Employ ment,” Economic Development Quarterly (No. 3, 1987), pp. 257-67. Farrell, Michael G., and Anthony Radspieler. “ Origin of Merchandise Exports Data,” unpublished manuscript, Foreign Trade Division, Census Bureau, U.S. Department of Commerce (March 20, 1990). Jackson, John D., and James A. Dunlevy. “ The Interchange ability of Alternative Measures of Permanent Income in Regression Analysis,” Applied Economics (October 1982), pp. 455-68. _______ . “ The Orthogonal Least Squares Slope Estimator: Interval Estimation and Hypothesis Testing Under the Assumption of Bivariate Normality,” Proceedings of the Business and Economic Statistics Section of the American Statistical Association (1981), pp. 294-97. Lerch, Stephen C. ‘‘State of Origin Export Data: An Interpre tive View,” unpublished manuscript, Office of Financial Management, State of Washington (1990). Little, Jane Sneddon. “ New England’s Links to the World Economy,” Federal Reserve Bank of Boston New England Economic Review (November/December 1990), pp. 33-50. Malinvaud, Edmond. Statistical Methods of Econometrics (North-Holland, 1980). 14The dashed line indicates that a state’s value added tends to be higher than its shipments. In particular, each dollar increase in state export shipments in 1986 was associated with a $1.36 rise in export value added, on average. JULY/AUGUST 1991 76 Ott, Mack. "Have U.S. Exports Been Larger Than Reported?” this Review (September/October 1988), pp. 3-23. Smith, Tim R. “ Regional Export Growth: Lessons from the State-Level Foreign Trade Data,” Regional Science Perspec tives (No. 1, 1990), pp. 21-38. _______ . “ Regional Exports of Manufactured Products,” Federal Reserve Bank of Kansas City Economic Review (January 1989), pp. 21-31. U.S. Department of Commerce, Bureau of the Census. Exports from Manufacturing Establishments: 1985 and 1986 (GPO, January 1989). ________Exports from Manufacturing Establishments: 1987 (GPO, February 1991). ________“ State Offices That Provide Export Assistance,” Business America (No. 2, 1991) pp. 24-25. U.S. Office of Management and Budget. Standard Industrial Classification Manual 1987. Webster, Elaine, Edward J. Mathis, and Charles E. Zech. “ The Case for State-Level Export Promotion Assistance: A Comparison of Foreign and Domestic Export Employment Multipliers,” Economic Development Quarterly (August 1990), pp. 203-10. Appendix A Interchangeability of Alternative Measures The existence of alternative export measures raises the issue of the extent to which the measures are interchangeable. In other words, do these measures provide virtually identical in formation about state export performance? Dif ferent criteria exist for assessing this issue. Three criteria are used here: 1) rank correla tion criterion; 2) simple correlation criterion; and 3) orthogonal regression criterion. The rank correlation criterion focuses on the degree to which measures have identical rankings for cor responding observations. As a first step, states (including the District of Columbia) are ranked from 1 (the state with the largest value of ex ports) to 51 (the state with the smallest value of exports) for each export measure. To make pair wise comparisons of the rank-order, a Spearman rank correlation coefficient, Rs is calculated as follows: (1) Rs = 1 - [6/N(N2- 1)] I (Oj-Ai)2, where i is a subscript denoting specific states (i = l, 2,. . ., 51); 0j is the rank of the ith state using one measure; Aj is the rank of the ith state using the alternative measure; and N is the sample size. If the rank-orders of the two measures are identical, then Rs= l. For example, using one measure and assuming that three states—A, B FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis and C—had export values of 300, 200 and 100, the states would be ranked 1, 2 and 3. If, using the other measure, states A, B and C had export values of 5, 4 and 1, then an ordering of the states based on the two export measures would be identical and the Spearman rank correlation coefficient is one. A rank-order of zero means that the rank-orders of the two measures are not related. A rank-order of minus one means that the rank-orders are the reverse of each other. Thus, the closer Rs is to one, the more similar the rank-orders and the more inter changeable the measures for ranking purposes. As indicated in the text, the Spearman rank correlation coefficient was approximately one for the comparison of the 1987 export measures in EME and OMC (.96) and the 1986 measures in EME and our export value added (.98). Conse quently, the measures provide highly correlated rankings. These results suggest that, as a sum mary indicator of states' relative export perfor mance, the measures provide roughly identical information. Satisfying this criterion, however, does not preclude large differences in a specific state’s rank across the measures. For example, recall that the state of Washington dropped from fifth place using the shipments-based data in the EME for 1986 to 12th place using our ex port value added measure. A stronger condition than rank correlation in volves the simple correlation of the levels of the 77 alternative measures. The simple correlation coefficient provides information concerning the extent of a linear relationship between the alter native measures. The simple correlation coeffi cient is calculated as follows: (2) r = Z (x-x)(y-y)/V (2 (x -x )2 X(y-y)2), where x and y are the sample means of the alternative export series, x and y. For any given state, if the value of exports using one measure exceeds the mean of this measure based on all states by a certain amount and the value of exports using the other measure also exceeds (or falls below) its series mean by a set amount, then a perfect linear correlation exists. The value of the correlation coefficient will equal one (or minus one in the event of a negative relationship). For example, as in the numerical example above, assume states A, B and C had export values of 300, 200 and 100 using one measure. Using the other measure, assume states A, B and C had export values of 450, 250 and 50. Thus, A’s exports exceed the first series mean of 200 by 100 using one measure, and exceeds the second series mean of 250 by 200 using the other measure. For a per fect linear correlation, C's exports (which are 100 less than the first series mean) must be 200 less than the second series mean (that is, equal 50). Since this is the case, the correlation equals one. A correlation coefficient of zero means that no linear relationship between the measures exists. As indicated in the text, the linear relationship between both sets of measures is strong. For the 1987 measures in EME and OMC, the cor relation coefficient is .95. For the 1986 measures in EME and of export value added, the correla tion coefficient is also .95. Since these coeffi cients are virtually one, the measures can be viewed as interchangeable using this criterion. When using a more stringent criterion, how ever, this is not the case. This criterion for inter changeability requires that the measures are not only highly correlated, but that they consistently differ by a constant, possibly zero. Once again, 11n contrast to simple regression, the fitted line in or thogonal regression is the one that minimizes the mean square of the perpendicular (rather than the vertical) devia tion of the sample points from the fitted line. See Malinvaud (1980), for a thorough discussion of the differences between orthogonal and simple regression. assume states A, B and C had export values of 300, 200 and 100 using one measure. Using the other measure, assume state A had exports of 350. For “difference preservation,” states B and C's exports must be 250 and 150. In this case, the two measures differ by 50 for each state. This difference preservation is known as an or thogonal regression criterion.1 Jackson and Dunlevy (1982) illustrate this criterion, in a time-series context, by a simple example of estimating a consumption function with different perm anent income measures. Assume perfect correlation between two income measures yj and y2, so that: (3) y, = a + by2, where a is the intercept and b is the slope. Suppose the following consumption function is estimated: (4) c = d + eyi + z, where d is the intercept, e is the slope and z is the random disturbance term. The slope is called the marginal propensity to consume and is the change in consumption associated with each $1 change in income. If y2 is used rather than y,, however, the consumption function becomes: (5) c = (d + ea) + eby2 + z. The two measures of income will yield the same estimate of the marginal propensity to consume only if b equals one. Using orthogonal regression, we generated estimates of b for the alternative export mea sures discussed in the text. Specifically, we estimated two equations similar to equation 3. In one regression, the 1987 measures of state exports in OMC and EME were used as y and y , respectively. In the other regression, the 1986 measures of state exports based on our calculations of export value added and in EME were used as y, and y2, respectively. The orthogonal regression criterion is not satisfied by alternative export measures. For the 1987 measures in EME and OMC, the or- 78 thogonal least squares estimate of b equals 1.37. A t-statistic can be used to test the null hypoth esis that b equals one.2 The critical t-value for a 5 percent significance level with 49 degrees of freedom is 2.01, which is far below the actual t-value of 7.03. Consequently, the null hypothesis is rejected. Similarly, the 1986 measures in EME and of export value added produce an ortho gonal least squares estimate of b (1.36), which yields a rejection of the null hypothesis that b equals one; the critical t-value of 2.01 at the 2Because of random variation in the data it is unlikely that b exactly equals one. Therefore, the t-statistic is used to test whether we can reasonably infer that the estimated value of b equals one. See Jackson and Dunlevy (1981) for additional details on hypothesis tests involving the or thogonal least squares slope estimator. 3A related issue involves whether the two measures are consistently proportional to one another, that is, whether they tend to differ by a given percentage. This is investi gated by testing whether the orthogonal least square slope estimator between the logarithms of the two measures significantly differs from one. Using the logarithms of the 1987 measures in EME and OMC, the slope estimate equals 1.01. The associated t-statistic is 0.109, which is less than the critical t-value of 2.01 (5 percent significance level). Consequently, the null hypothesis that the slope 5 percent level of significance is far less than the actual t-value of 6.35.3 The implication of this analysis is that the levels of alternative export measures are not in terchangeable using this criterion and their use would generate different regression results in otherwise identical estimations. Specifically, the coefficient estimates for the impact of a change in state exports on some variable of interest, say state economic growth, would differ depen ding on the measure used.4 equals unity cannot be rejected. These results suggest that the logarithmic forms of the two 1987 export measures in EME and OMC are interchangeable. On the other hand, the logarithmic forms of the 1986 measures in EME and of export value added yield an orthogonal least squares slope estimate of 0.892. Because the associated tstatistic of 2.95 exceeds the 2.01 critical value, the null hypothesis is rejected, suggesting that the two measures of 1987 exports are not interchangeable. 4See Coughlin and Cartwright (1987) for an empirical ex amination of the effect of manufacturing exports on employment for individual states. This is an example of a study where the regression results could be altered by us ing different export series. Appendix B Estimating Value Added Related to Manufactured Exports by State In this appendix, we identify the data and methodology used to calculate the value added related to manufactured exports by state. We begin by identifying the variables used in the calculations and the data sources. Various employment, shipments and gross state product data are essential for our calculations. Manufacturing employment (ME), export-related manufacturing employment (XME) and exportrelated nonmanufacturing employment (XNME) for each state are published in Exports from Manufacturing Establishments: 1985 and 1986, 1For a detailed explanation of how these data were developed, see this publication’s Introduction and Appen dix C. FEDERAL http://fraser.stlouisfed.org/ RESERVE BANK OF ST. LOUIS Federal Reserve Bank of St. Louis U.S. Department of Commerce (1989).1 ME is reported by respondents in the U.S. Census Bureau’s Annual Survey of Manufactures, while XME and XNME are calculated by the Census Bureau. Three other series are used. The first is un published data from the U.S. Department of Commerce (1991) on nonmanufacturing employ ment (NME). The second data series, which is published in Exports from Manufacturing Establishments: 1985 and 1986, is total state employment. The third data series, which is published by the U.S. Department of Commerce (1989), is gross state product (GSP). GSP is the 79 market value of the goods and services pro duced within a state during a year and is the state analog of U.S. gross domestic product. GSP data for individual manufacturing and nonmanufacturing industries were used. M eth od olo gy: C alculating E x po rt Value A d d ed To calculate total value added related to manufactured exports in state s (XTVS), we summed estimates of value added within the state's manufacturing sector (XMVS) and value added in nonmanufacturing sectors related to the export of manufactured goods (XNMVJ. That is, (1) XTVS = XMVS + XNMVS. Because identical data were not available for each manufacturing sector, the components of XMVS were calculated in one of two ways.2 For industries in which export-related data are published, XMVS was estimated by applying the following equation: (2) XMVsi = (GSPsi/MEsi)(XMEsi). As defined above, GSP is gross state product, ME is manufacturing employment and XME is export-related manufacturing employment. The subscript i designates the different SIC manufac turing industry groups. In our calculations, we used the two-digit manufacturing industry group so i ranged from SIC 20 to SIC 39. This method implicitly assumes, for each industry, that output per worker in the production of ex port goods (XMVsi/XMEsi) is the same as output per worker in the production of all goods (GSPsi/MEsi). Equation 2 was applied using data for individual industries rather than for total manufacturing, because this assumption is more plausible for each industry than for manufac turing as a whole. For those manufacturing sectors with no pub lished export-related employment at the two 2ln some states, export-related manufacturing employment data were not published for certain industries either to avoid disclosing data for individual companies or because the estimate did not meet publication standards. Summing over all states, unpublished export-related manufacturing employment accounts for 1.8 percent of total 1986 exportrelated manufacturing employment. digit industry level, XMVS was estimated using the following equation: (3) XMVsm = (GSPsm/MEsm)(XMEsm), where the m subscript refers to the total of those sectors not reported. For example, to compute a state’s total unreported export-related manufacturing employment (XMEsm), we simply subtracted the amount reported from the total export-related manufacturing employ ment. Consequently, XMVS is the sum of the estimates for the reported industries (XMVsj) plus the single estimate for the missing in dustries (XMVsm). To compute a state’s value added in its non manufacturing sectors related to manufactured exports (XNMVS), we first estimated the follow ing measure for each of a state's four nonmanu facturing industries [where j = 1, trade; j = 2, business services; j = 3, transportation, com munication and utilities; and j = 4, other]; (4) XNMVsj = (GSPsj/NMEsj) (XNMEsj), where GSP for the "other” sector is calculated as total state GSP minus manufacturing and minus the three nonmanufacturing industries, (j = 1...3); NMEsj is nonmanufacturing employ ment in industry j; and XNMEsj is export-related nonmanufacturing employment in sector j.3 “Other” employment is total state employment minus employment in manufacturing, trade, business services and transportation, com munication and utilities. The state total for the value added in these nonmanufacturing sectors, XNMVS, is simply the sum of the value added in the four nonmanu facturing sectors. The accuracy of this calcula tion, similar to the calculation for the manufac turing sectors, rests on the degree to which the productivity of export-related workers in sector j (XNMVsj/XNMES is equal to the productivity of j) all workers in that sector (GSPsj/NMEsj). 3Export-related nonmanufacturing employment in the “ other” sector accounts for 31.8 percent of the 1986 national total for such employment, JULY/AUGUST 1991 F ederal R eserve Bank o f St. L ouis Post Office Box 442 St. Louis, Missouri 63166 The R ev iew is published six tim es p e r yea r b y the Research and Public Inform ation D epartm ent o f the Federal R eserve Bank o f St. Louis. Single-copy subscriptions are available to the public fr e e o f charge. Mail requests f o r subscriptions, back issues, o r address changes to: Research and Public Inform ation Departm ent, Federal R eserve Bank o f St. Louis, P.O. Box 442, St. Louis, M issouri 63166. The view s expressed are those o f the individual authors and do not necessarily reflect official position s o f the Federal R eserve Bank o f St. Louis o r the Federal R eserve System . Articles herein m ay be reprinted p ro vid ed the source is credited. Please provide the B ank’s R esearch and Public Inform ation D epartm ent with a co p y o f reprinted m aterial.