The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Federal Reserve Bank of Cleveland Economic Review July 1980 ISSN 0013-0281 ■illli.iJli.ilhi.illli.illh.illh.jlli. Ili.illli.illlu llli.illli.illli.irili.ill ■ilm.mli.immli.illi.mMm. Hi . i H i . i IIIi j IIIi . i Ii Ii .i IIIi . i BIi . i II July 1980 Economic Review Federal Reserve Bank of Cleveland Contents Interpreting Movements in Seasonally Adjusted Money-supply D a ta ........................................................1 Since policymakers need to evaluate money-growth rates between dates of different seasons, they must estimate what portion of money growth is seasonally transient and what portion is enduring. Unfortunately, the process o f identifying the seasonal component of a time series is not a perfect one, and it may introduce errors into the interpretation o f the raw data. Alternative seasonal adjust ment methods are applied to recent money supply data to illus trate this interpretation problem. Particular attention is paid to the influence of the income tax season on money supply changes. Monitoring the Economy: Survey of Fourth District Manufacturers.................................................................7 Monitoring changes in regional manufacturing activity is one means o f analyzing current business-cycle conditions. Since 1965, the Federal Reserve Bank of Cleveland has been surveying manu facturing firms within the Fourth Federal Reserve District on a monthly basis to ascertain basic conditions of production activity. This article describes the survey and the cyclical properties of the indexes derived from the survey and provides a statistical analysis of the performance o f the index over the last ten years. Interpreting Movements in Seasonally Adjusted Money-supply Data by John B. Carlson, Paulette M. Maclin, and Mark S. Snkjerman Why Seasonal Adjustment? Many economic time series commonly are adjusted for seasonal fluctuations. Seasonal adjustment is designed to eliminate movements in a time series that result only from seasonal events, such as holiday shopping and regularly occurring weather changes. Variations in the money supply are influenced strongly by income-tax filing and Treasury disburse ment patterns. Tax refunds are held temporarily in checking accounts, while those who owe taxes usually fund their accounts just before the payment date. Both actions increase the narrowly defined money supply. All seasonal-adjustment techniques are designed essentially to “ redistribute” the un adjusted (raw) data throughout a year in a pattern that differs from actual observance. The resulting seasonally adjusted data follow a pattern that should no longer be influenced by seasonal events. With seasonality thus neutralized, the basic movement in the data becomes more readily apparent. If policymakers were interested only in com paring the current money-supply level with the level o f the same season one year earlier, the raw figures frequently would be sufficient. Since policy makers need to evaluate money-growth rates between dates of different seasons, they must estimate the seasonal influences and take them into account when assessing the changes in the actual money-supply figures. The convenience o f employing seasonally adjusted data is not necessarily costless, however. Often, the seasonal influences cannot be accurately estimated, for a variety of reasons. As a practical m atter, the correct adjustment process for each data series can never be known with certainty in advance. An improper set of seasonal-adjustment factors introduces errors into the interpretation of the raw data, and the errors are likely to be inversely related to the length of the time interval being considered. For example, the seasonal component of weekly money-supply data is more difficult to ascertain than the seasonal component of quarterly money-supply data. In addition, the money supply is an aggregate of parts, such as currency and checking accounts, each of which has its own seasonal process. Failure to account adequately and separately for each process may introduce error. If seasonal processes are considered to be basically stable, the seasonal pattern would be ex tracted from a raw-data series by a technique that treats each year of the series with equal weight. But, if seasonal processes are expected to change frequently and along a trend, the seasonal pattern would be extracted by a technique that gives more weight to the most recent years in the data series. Some analysts argue that changes in seasonal processes cannot be determined until several years after the fact; these analysts would claim that stable seasonal ity ought to be presumed when adjusting current data, unless overwhelming evidence to the contrary exists. Other analysts are confident that unequal weighting of the years in a data series can be jus tified, based on information about changes in the underlying determinants of the seasonal process. One’s interpretation of a raw-data series there fore is conditioned by the “prescription o f the eyeglasses” that one wears. The raw data will appear differently when viewed through different seasonaladjustment lenses. The analyst’s problem stems from the fact that there is no single procedure known to be correct. Analysts who differ in the assumptions regarding the appropriate seasonal influences on the money supply consequently may differ in their in terpretations of current money-supply growth. More over, the differences introduced by seasonal-adjustment techniques can be sizable. An examination of the seasonal factors used to adjust M-1A for the 1972-1979 period appears in table 1. The data indicate that the February-July factors have become more volatile over time, while the August-January factors have become less volatile. The intra-period means have been very stable, suggesting that changes in m onthly factors within a period have been entirely offsetting. The consequence of these developments is that a constantly growing raw-money supply would appear, on a seasonally adjusted basis, to change first more sharply, then less sharply, in 1979 than in 1972. Table 1 Means and Standard Deviations in M-1A Seasonals: 1972-1979 February-July Standard Mean deviation 1979 1978 1977 1976 1975 1974 1973 1972 99.39 99.38 99.44 99.43 99.34 99.35 99.26 99.23 1.16 1.07 0.86 0.85 0.77 0.79 0.75 0.73 August-January Mean 100.76 100.80 100.72 100.71 100.71 100.70 100.74 100.78 Standard deviation 1.55 1.64 1.74 1.75 1.72 1.71 1.75 1.84 Seasonal-adjustment Techniques When approaching the problem of seasonality, the analyst should have some knowledge of the mech anism generating the data. This prior knowledge suggests the appropriate technique for adjustment. In some cases, it may not pay to attem pt seasonal adjustm ent—for example, when seasonal variation is dwarfed by irregular movements or for a new series offering little or no historical experience. Economic theory may suggest reasons for relating changes in the seasonality of the money supply to readily measured variables, such as the timing of concentrated volumes of tax receipts or refunds. In principle, a regression model could be used to estimate these relationships to forecast changes in seasonality. Unfortunately, the science of economics has yet to produce a model that explains enough about seasonality to be useful at this level. In the absence of a more complete theo retical structure, most seasonal-adjustment techniques estimate and remove seasonal variations in the data by a weighted-average smoothing process that is based on the past history of the variable being adjusted. The most widely used technique is the X -ll seasonaladjustment program of the Bureau of the Census. Because the X -ll program does not use infor mation exogenous to the data series being adjusted, it is not likely to anticipate and capture abrupt changes in seasonality.1 The X -ll is the method employed in seasonally adjusting the money supply. It is applied to each of the basic aggregate components, such as currency and demand deposits in M-1A. The X -ll is an iterative procedure involving two fundamental steps designed to separate any monthly time series into three dis tinct series, identified as the trend-cycle, seasonal, and irregular components. The first step seeks to iso late the trend-cycle component from the seasonal and irregular components by dividing the original series by an estimate of the trend-cycle component. The second step is designed to separate the seasonal and irregular components. The X -ll procedure is iterative in two senses: (1) it repeats the second step, using a revised seasonal component in which extreme irregular values are eliminated or replaced with damp 1. In recent years, a considerable amount of research has pro vided several alternatives to, and improvements in, X -ll. A review of this material is found in David A. Pierce, “A Survey of Recent Developments in Seasonal Adjustment,” American Statistician, vol. 34 (August 1980), pp. 125-34. For further information on the X -ll method, see Julius Shishkin, Allan H. Young, and John C. Musgrave, The X - ll Variant o f the Census M ethod II Seasonal A djust ment Program, Technical Paper No. 15 (U.S. Department of Commerce, February \96 1 ),X -1 1 : Information fo r the User, Papers prepared for the Seminar on Seasonal Adjust ments of the National Association of Business Economists, March 10, 1969 (U.S. Department of Commerce); and Thomas A. Lawler, “ Seasonal Adjustment of the Money Stock: Problems and Policy Implications,” Federal R e serve Bank o f Richm ond Economic Review, vol. 63 (No vember/December 1977), pp. 19-27. ened ones; (2) it repeats both steps by reestimating the trend-cycle component, using alternative averaging methods employed on a preliminary seasonally ad justed series obtained in subsequent rounds. Although the technique is considered me chanical, it permits the use of judgment to the extent that some parameters o f the X-l 1 program can be varied. Until recently, however, the staff o f the Board o f Governors o f the Federal Reserve System made additional judgmental modifications to the final estimates o f the seasonal-adjustment factors obtained from the X-l 1 procedures and used in the forthcoming year. For example, the X-l 1 is not designed to handle abrupt changes in seasonal patterns that may result from a change in a tax-filing date or from the im plementation o f new tax-processing methods. Given prior knowledge, such events can be anticipated by modifying the seasonal factors estimated by the X-l 1. Although this practice is often necessary, it is difficult to perform consistently over time. In recent years, the Board staff has limited itself to making adjust ments that are permitted within the structure of the X -ll procedure. The adjusted series that con forms most closely to prior expectations based on all available information is selected and used. The discretion allowed by the X -ll program is best exemplified by the user’s options for choosing both the length o f the period and the weighting structure of the moving average. The moving average options are available when estimating both the trendcycle and seasonal components of the series. Although the X -ll automatically selects default values for these options, the user has available alternatives that permit variations in the degrees o f smoothing.2 When estimating the final trend cycle, the degree of smoothing (length o f moving average) desired would depend on the relative importance (average percent change) o f the irregular variations to the trend-cycle movements. The greater the irregular movements relative to the trend cycle, the longer the moving average needed to smooth out the short-term 2. Experience has shown that the seasonal component of many economic time series can be adequately estimated by the same choices of X -ll options. Consequently, the X-l 1 program is preset to these default options, which can be changed as circumstances warrant. movements and reveal the trend. Conversely, if cyclical movements dominate, then a short moving average would better reveal the systematic move ments of the series. Similarly, when estimating the seasonal com ponents, the degree of smoothing desired would depend on the relative importance of the regular variation. If a seasonal for a given m onth is believed to be stationary, then all the movement in the seasonalirregular component for the m onth must result from irregular variation. Thus, the user would choose to average as many years of that m onth as possible in order to average out the noise. For this reason, the X -ll has an option that averages seasonal-irregular (S-I) values of the same m onth for all prior years available, giving equal weight to each year. On the other hand, if the seasonal factor is believed to be changing, then movements in the S-I component reflect movements of both individual components, and the default option may be de sirable. This option takes a five-year moving average that weights most heavily the S-I component in the year being estimated. The two years before and the two years after are weighted with lesser weights (declining away from the year). When the seasonal being estimated is for the most recently available year, only the two prior years are included. Although a short moving average may fail to average out irreg ular noise, it enhances the probability that a seasonal factor would correctly incorporate movements reflecting fundamental changes in the determinants of seasonality. It also enhances the probability of removing irregular variations under the guise of seasonal variations. The trade-off is clear. If a priori evidence exists that movements in the seasonal are large relative to irregular variation, then a short averaging period is desired. Finally, if seasonal factors are believed to be changing, and the character and extent of the changes are thought to be known, then it may be desirable to estimate this change and project it into the forthcom ing year. An option of the X -ll adds one-half the change from the previous year to the seasonals of the last available year to obtain the seasonals for the up coming year. While such a formula may be rigid, it may be preferable to using the last available seasonals when there is strong evidence that seasonals in fact are changing. Impact of the Tax-filing Season on Money The difficulties in estimating the seasonal factors during the personal tax-filing season (pri marily February-May) perhaps best exemplify one of the shortcomings of the X -ll method. While it is quite predictable that the nonseasonally adjusted money supply increases greatly around the tax-filing deadline of April 15, a number of variables affect the exact time pattern of this phenomenon. If these factors all changed regularly along trends, then the X -ll probably would adequately capture the move ment in the seasonal factors. Unfortunately, some clearly identifiable determinants of the timing of tax flows have been behaving in an unpredictable manner, causing abrupt changes in the time pattern of tax payments and refunds. Because the X -ll ignores economic structure, it offers no mechanism for prompt incorporation of such changes into the seasonal factors. Table 2 Federal Income-tax Refunds of Individuals Chart 1 M-1A Monthly Growth Rates: 1980 As a percent of nonseasonally adjusted demand deposits of previous m onth 1975 February March April May 2.0 4.0 3.1 6.1a Percent (saar) 1976 1977 1978 1979 1980 1.9 4.1 3.5 2.4 1.8 4.3 3.4 2.4 1.3 5.0 3.3 2.9 1.1 4.5 3.4 3.0 1.9 4.6 4.1 4.2 a. Includes income-tax rebate. Table 3 Non-withheld Income-tax Receipts of Individuals As a percent o f nonseasonally adjusted demand deposits of previous m onth February March April May period of March to May. Thus, refunds can produce a sizable bulge in the money supply during this period. Second, and quite aside from refunds, persons owing taxes fund their checking accounts on or before the date the tax is paid. The time required for mail service and actual processing of the tax payments tends to increase the money supply until the Treasury cashes the checks. The monthly patterns of tax refunds and payments relative to the level of demand deposits for recent years are given in tables 2 and 3. Variables that affect the timing of the tax-related impact on the money supply include the timing of tax filing, taxes and tax-withholding rates, the effi ciency of the tax-processing equipment, the number of people employed to process returns, and the mail service. In the past two years, for example, individuals filed returns later than in previous years. As the April 15 deadline approached, the Treasury faced a larger than usual bottleneck in tax processing. 1975 1976 1977 1978 1979 1980 0.5 1.3 6.2 0.4 0.4 1.2 6.0 0.3 0.5 1.2 6.5 0.9 0.4 1.1 5.7 2.7 0.4 1.3 7.3 2.2 0.5 1.2 9.3 0.8 Tax processing affects the money supply in two particular ways. The typical tax refund (averaging roughly $250) is frequently deposited in checking accounts, if only temporarily. Refunds in the past two years totaled $31 billion, or 8 percent of M-1A. Almost 90 percent of this amount is paid out in the Although new equipment was in operation, the Treasury’s budget did not permit the overtime required to process the tax returns (particularly those with tax payments) at the rate achieved in the previous years. As a result, a greater percentage of the tax payments were not actually collected until May. This series of events may explain some of the extraordinary M-1A growth in April of 1978 and 1979 (see table 4). Furthermore, because the X -ll often is used in a manner that weights the experience of recent years more heavily, the current year’s seasonal factors probably overstate the tax-processing impact in April 1980. This seems all the more probable since a greater dollar volume of taxes was processed in April 1980 than in April 1978 and 1979. One consequence o f in creased processing could be that the seasonal adjust ments in April are stronger than they should be, contributing to the negative growth of M-1A last April. In effect, tax-processing delays in 1978 and 1979 may have operated through the seasonal-adjust ment procedure to contribute to the sharp drop in the money supply in April. Without a well-estimated model, however, a quantitative assessment of this impact is extremely hazardous. Recent M-l A Growth By looking at the raw data through somewhat different sets of eyeglasses, one can obtain a rough idea of the extent to which the 1978-1979 taxprocessing problems may have contributed to the steep drop in M-l A this past April. As noted earlier, the official 1980 seasonal factors were constructed using the default option, in which greater weight was given to the data of the two most recent years. This m ethod is consistent with the assumption that the M-l A seasonals are changing. Alternatively, how ever, one can assume that the seasonals are relatively stable and use a longer averaging process in which each year receives equal weight. Charts 1 through 3 illustrate how this alternate weighing process can influence one’s perception of money growth, not only for April but for the first six months of the year as well. When using the con stant seasonals, M-l A still shows a considerable decline Chart 2 Official M-l A Series Billions of dollars (sa) 390 Table 4 April Growth Rates for M-l A Seasonally adjusted annual rates 1975 1976 1977 1978 1979 1980 1980 seasonal factors Stable seasonals M-l A growth for year ending in April -3.0 6.6 9.2 12.6 14.7 -18.5 -3.2 9.4 15.1 20.7 23.7 -9.6 3.4 6.2 6.9 7.3 5.8 3.7 While it is conceivable that the relationships between seasonal events and the money supply could be modeled and well estimated as a practical m atter, the problems involved are enormous. The seasonal factors used to adjust current data are calculated at the beginning o f the year. Consequently, optimal adjustment o f the data would be required to forecast the exact time pattern o f each seasonal event for the entire year ahead. It is not likely that any of these events could be forecast with the desired degree of accuracy. Thus, it may be of little consequence that the X-l 1 ignores this information. Billions of dollars (sa) in April but at only about half the rate obtained using the official factors. Thus, the April drop appears some what less alarming. At the same time, however, by redistributing money growth somewhat differently throughout the entire tax-filing period, the equal-weight factors ap pear to signal some cause for concern about money growth earlier in the year. As charts 2 and 3 show, money growth, officially adjusted, was close to or above the midpoint of its long-run growth range throughout the first quarter. In February, M-1A actually moved above the top of its growth range, con tributing to the adoption of more restrictive money and credit policies. By contrast, the February bulge in the stable seasonal money series merely carried M-1A to within the lower half of its long-run growth range. Otherwise, the equal-weight series has been below path throughout the period. This growth pattern would have been consistent with a less restrictive policy in the first quarter. The two different policy prescriptions indicated by applying alternate sets of seasonal factors to the same raw data highlight the problem faced by policy makers in trying to separate seasonal from trend-cycle movements. Moreover, it is quite probable that neither of the two sets of seasonal factors used here is entirely correct. The equal-weight factors do not capture known changes in the determinants of the seasonal process. The official factors, which are influenced heavily by the extraordinary events of 1978 and 1979, may fail to average out enough of the irregular com ponent of the data. April Growth and Moving-average Options The different growth patterns of M-1A implied by alternative smoothing assumptions are illustrated in table 4. In the years 1976 through 1979, seasonally adjusted growth rates for April (both official and those assuming stable seasonals) were greater than growth rates for the year ending in that month. Furthermore, the disparity increased over that period. In the absence of any explicit information, it seems unlikely that solely nonseasonal determinants could account for the four straight years of increasing disparity. It appears as if the seasonal relationship is changing. By comparison, it is evident that the official seasonals for March and April embody some change over time, which is verified in table 5. The trends that decrease the seasonal factors in March and increase them in April tend to dampen the growth of seasonally ad justed demand deposits in April. Thus, based on strictly empirical grounds, the official seasonal factors appear closer to the true seasonals than those that assume stability for this period. Table 5 Seasonal Factors for Demand Deposits March 1975 1976 1977 1978 1979 1980b April Actual3 Stable option Actual3 0.985 0.982 0.981 0.979 0.978 0.978 0.984 0.984 0.984 0.984 0.984 0.984 1.009 1.011 1.013 1.013 1.014 1.014 Stable option 1.010 1.010 1.010 1.010 1.010 1.010 a. Estimated in 1980. b. The 1980 seasonal. In 1980, however, the April growth trend was reversed, suggesting an abrupt change in seasonality. However, a number of cyclical factors (such as the declines in industrial production and business loans) also could have accounted for much of the sharp drop in M-1A in April. In the absence of a well-estimated model of seasonality, it is difficult to defend how much of this change, if any, should be embodied in the 1981 seasonals. As noted earlier, some analysts would argue that the Treasury’s expeditious performance in tax process ing during 1980 explains a significant portion of the M-1A decline in April. If the Treasury continues to perform in such a manner consistently over the next few years, perhaps the true April seasonal would sta bilize around a particular value. However, the April seasonal is probably related to other determinants that are changing both randomly and systematically. In either case, it could be argued that the five-year weighted-moving-average option is likely to be appro priate most of the time. In terms of time-series analy sis, seasonality is not likely to be stationary, whereas the stable option assumes it is. Anyone who has applied time-series methods to economic variables appreciates the rare occurrence of stationarity in the original series. Monitoring the Economy: Survey of Fourth District Manufacturers by Robert H. Schnorbus Monitoring changes in economic activity among manufacturing firms on a timely basis provides a valuable, but seldom possible, means o f analyzing current business-cycle conditions. In fulfilling its responsibilities toward formulating monetary policy, the Federal Reserve Bank of Cleveland monitors the current performance of the regional economy, especially regional changes in economic activity that may signal national cyclical developments. In response to the need for regional information, in 1965 the bank developed a monthly survey of manufacturing firms within the Fourth Federal Reserve District (4D) to ascertain basic conditions o f production activity. When converted into diffusion indexes that measure the direction o f change in eight key indicators of economic activity, the 4D survey findings can be used to interpret the timing, amplitude, and duration of the current phase o f the business cycle. As with any analytical tool, the value of the indexes obtained from the 4D survey depends on an understanding of both the strengths and weak nesses of the survey. Major weaknesses in businesscycle analysis are the limitations imposed by the quality, timeliness, and availability of data. Often, key indicator series for the national economy either are not published on a m onthly basis or are published with considerable time lags. Comparable regional data, where available, may involve even longer lags than national data. Although detailed monthly employment data for the United States, for example, are available within a week after the end of each m onth, state (payroll) employment data by industry are not released for another one to two months. Other monthly data, such as new orders, shipments, and inventories, take much longer to collect and process at the national level, and often are not available at the regional level. As a consequence, short-term economic forecasting and policymaking exclude key measures that may signal impending change in economic activity. The monthly survey of the 4D manufacturers collects a variety of reliable information about the district’s production base with a minimal time lag. The shorter time lag of the 4D survey, compared with alternative sources of information, is its most valuable contribution. However, it is not simply availability and timeliness of data that are important. A data series should have a record for accuracy in identifying important characteristics, such as turning points in economic activity. Ideally, the 4D indexes should closely correspond to the timing, amplitude, and duration of comparable national series published by other sources, except where distinct regional character istics are captured by the 4D survey. This article assesses the 4D survey indexes and their accuracy over the past ten years by statistically comparing them with rates of change in comparable aggregate series at the national level.1 1. Earlier evaluations of the 4D survey found the indexes to be consistent with behavior of their national counterparts. However, the survey sample has undergone changes in re cent years, and the popularity of surveys, in general, has declined with the growth of econometric forecasting mod els. For earlier studies, see “ Diffusion Indexes and Eco nomic Activity,” Economic Review, Federal Reserve Bank of Cleveland, January 1971, pp. 3-17; and Theodore S. Torda, “The Monthly Survey of Fourth District Manufac turers-A n Early Warning Signal,” Economic Commentary, Federal Reserve Bank of Cleveland, October 27,1969. Table 1 Survey Sam ple D istrib ution : Industry Food Furniture Paper Printing Petroleum Primary metals Fabricated metals Machinery Electrical equipment Transportation equip ment Sample total 1978 Total employment of sample3 Survey employment, percent Ohio employment, percent^5 U.S. employment, percent*5 1 3 2 2 1 6 4 5 4 2 27,100 16,300 37,000 13,698 12,927 227,505 16,105 93,894 205,000 14,500 4.1 2.5 5.6 2.1 1.9 34.3 2.4 14.1 30.9 2.2 7.2 1.7 3.8 6.3 1.5 14.9 16.5 20.9 11.8 15.4 12.7 3.6 5.2 8.8 1.5 8.9 12.4 17.5 14.7 14.6 32 664,029 100.0 100.0 100.0 Number of firms a. Employment figures are based on a survey sample taken in July 1978. b. Percentages are based on sum of employment for listed industries. SOURCES: Bureau of Labor Statistics and Federal Reserve Bank of Cleveland. Description of the 4D Survey The survey is derived from a sample of 4D manufacturing firms intended to reflect the regional manufacturing sector, especially the disproportionate share of durable-goods industries. To be sure, some of the participating firms in the current sample have only corporate headquarters remaining in the 4D, but most of the firms still have operating plant facilities within the district. To the extent that these firms also have nationwide operations, the sample captures national characteristics. (Table 1 compares a typical survey sample with the actual distribution of employ ment for Ohio and the nation among a limited set of ten industries.) Because of the predominance of durable-goods producers in the 4D, the sample has tended to overemphasize durable-goods industries, such as primary metals and electrical equipment. Other durable-goods industries, such as fabricated metals and transportation equipment, have been underestim ated.2 (Some im portant industries, such as rubber, are not represented.) While the sample size can vary from m onth to m onth, the number o f active participants currently averages 30 to 35. The monthly survey consists of eight indicators of economic activity and a composite index. One set of indicators tends to be production-oriented, includ ing employment, hours worked per week, prices paid for materials, and inventories (see chart 1). Another set tends to be sales-oriented and includes new orders, backlog of orders, delivery times, and shipments (see chart 2). For each of these indicators, participants are asked whether manufacturing activity increased, de creased, or remained unchanged in the previous m onth and what is expected for the m onth in progress. By adding the percentage of responses that report in2. For example, 34.3 percent of the total employment repre sented by the survey sample was contained in primary metals, while only 14.9 percent in Ohio (a proxy for the 4D) and 8.9 percent in the nation would have been re quired for an unbiased sample. Likewise, 2.4 percent of the sample’s employment was well below a required 16.5 percent in Ohio and 12.4 percent in the nation. In order to test the comparability of the sample employment dis tribution with the national distribution, a Spearman rank order correlation test was performed, using the percentage figures in table 1. The null hypothesis that the percentage distribution of employment is similar must be rejected on the basis of the computed correlation v a lu e (p s = 0.34, with nine degrees of freedom). Chart 1 Production-oriented Indexes (Nov.) P (Dec.) (Nov.) Percent p 70 __ Emp] Dyment 60 (Mar.) T \ 50 40 V' V / \ \ \ 30 — \I 20 .... 70 60 _ Hour 50 A V 40 V / l r -W v A i \ A A r 30 ; 7 20 70 60 — 50 Inver tories /" S /\ / V 40 A/A V \ / v K 30 20 90 / \\/n 1 tn il ' VV / / b u / * 7*^ 1\l1 1 Price 80 70 60 50 40 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 NOTES: Dotted lines indicate anticipated values for month in progress. Shaded areas indicate periods of contraction. creases and one-half the percentage o f responses that report no change, a numerical score—called a diffusion index—is com puted for each indicator.3 The score for the m onth in progress is computed as a forecast value, although it may be based on partial information avail able at the time of the survey. By a simple averaging technique, the eight index values are then combined into a composite diffusion index that serves as a measure of overall manufacturing activity. The 4D indexes suffer from statistical problems that are both inherent in the method of computing diffusion indexes and specific to the 4D sample. First, the sample distribution can be affected by the inclusion (or exclusion) of some of the largest firms in the 4D. For example, the overemphasis on primary The index values are published after a seasonal-weight fac tor is applied to produce raw scores, which are then aver aged with the raw scores of the previous month. The month-to-month fluctuations thus are smoothed out to reveal more clearly any underlying cyclical pattern. For alternative methods o f constructing diffusion indexes, see Arthur F. Bums, “ New Facts on Business Cycles,” in Geoffrey H. Moore, Ed., Business Cycle Indicators, vol. 1 (National Bureau of Economic Research/Princeton Univer sity Press, 1961), pp. 1344. Chart 2 Sales-oriented Indexes Percent 60 L 20 (Dec.) P (Nov.) T (Nov.) P (Mar.) T Delivery Time — B — I I ---------------------------------------------------------------------------------------L — 1— -- ---------------------------------------------------------------- Shipi tents A n 1969 1970 1971 1972 1973 1974 /V A _ V f 1975 1976 ' 1977 rAAa\ j ft'hL ' vvv%f\ 1978 1979 _ 1980 NOTES: Dotted lines indicate anticipated values for month in progress. Shaded areas indicate periods of contraction. metals was partly the result of including three of the largest steel-producing firms in the nation. The indexes are likely to show a bias toward events that are specific to a given industry, such as steel, and to the 4D economy, such as localized labor-management dis putes. A second related source of bias results from the treatm ent of each participating firm on a “ one firm-one vote” basis, regardless of the relative size of the firm ’s work force. Thus, a large firm is given as much weight in the com putation of the indexes as a small firm, even though the magnitude o f the change associated with the response of a large firm may represent a greater economic impact on the nation. Third, the eight individual indexes are not all independent. Delivery time, for example, is equiv alent to backlogs divided by shipments, and the new-orders index is equivalent to the change in backlogs plus shipments. The problem of overlapping indexes is particularly relevant to the composite index, which, as a result, may overemphasize certain aspects of manufacturing activity. Finally, diffusion indexes, in general, are valuable in terms of indicating the direction of change in activity, but are more limited in terms of measuring amplitude of change. Cyclical Properties of the 4D Indexes The rationale for the 4D indexes is derived from the tendency for economic expansions and contrac tions to spread by a cumulative process among firms, industries, and regions. The relationship between cyclical spreading, or diffusion, of expansions or con tractions and the timing, duration, and amplitude of business cycles reflects the interconnection among business activities of firms and the process by which adjustments to the cycle are made. A diffusion index can be related to the rate of change of a comparable aggregate series through the assumption that the deeper a contraction (and, therefore, the greater the rate of decline in the aggregate series), the more wide spread the contraction becomes among firms and industries (see Properties of a Diffusion Index, p. 12). The expected behavior o f the 4D survey indexes then is based on the performance of comparable aggregate series with which the indexes should have close his torical relationships. The mixture o f production-oriented and salesoriented indicators creates some unique properties for the behavior of the 4D composite index. Some of these properties tend to lead and others tend to lag the ref erence cycle as defined by the official dating o f peaks and troughs by the National Bureau o f Economic Re search (NBER). The combination of cyclical properties causes the composite index to differ from a diffusion series based on other aggregate series, such as the FRB Index of Production, which tends to coincide with cyclical peaks and troughs. Although the construction of the composite diffusion index as an average o f the eight individual indexes would suggest the likelihood of a coincident indicator, the expected behavior of the composite index depends on the predominance o f the lead-lag relationships among the eight individual sur vey indexes. Among the production-oriented indicators, only manufacturing employment qualifies as an aggregate series that is roughly coincident with the overall business cycle. The timing of the turning points of the employment series is expected to be coincident, because employment is a product of cyclical changes in demand. Thus, employment is at the center of cyclical movement, as it adjusts to changes in demand. The amplitude of employment swings may be mild, however, compared with changes in production, be cause employers often try to protect their skilled work force. Employers have the alternatives of not replacing workers who leave or retire and of slowing the rate of expansion of their work force. Production may expe rience wider fluctuations than other aggregate series, so that the relative adjustments of employment reflect the extent to which further adjustments in the other production-oriented aggregate series may be required. The flexibility of hours worked and, to a degree, of prices of materials over the cycle makes them potentially leading indicators. Shortening of the work week in response to an economic slowdown usually occurs before other employment adjustments. Changes in hours are easier to administer, easier to reverse when necessary, and just as likely to reduce costs per hour (for example, when overtime is reduced or when train ing costs for new employees are avoided). The prices of materials, too, generally change quickly as inven tories of raw materials accumulate or contract over the business cycle. When demand for a firm’s output is expanding, the firm’s demand for input expands. If the rise in input demand cannot be accommodated by existing inventories, suppliers are tempted to raise prices. If inventories accumulate during a slowdown, suppliers may offer price discounts to manufacturers. Whether adjustments are made by altering hours worked or prices, the failure to keep production in line with demand has a spillover effect on inventory levels of finished products. Manufacturers produce either to stock, as in products such as nuts and bolts, or to order, as in products with special requirements. Pro duction to order requires lower inventories of finished goods and greater control in managing inventory levels of materials. Production to stock depends on the accu racy of the firm’s economic forecasts of demand for its product. As a rule, inventory levels begin to accumu late beyond the cyclical peak in economic activity be cause of the lags inherent in detecting and confirming a peak. Once inventories are high relative to desired levels, firms will attem pt to liquidate their oversupply. The liquidation may continue beyond the cycle’s trough, not only because of the problem of recognizing a trough, but also because orders should pick up be fore production will be expanded. Yet, even though inventory levels may lag the cycle, additions to or re- Properties of a Diffusion Index Stage 1 2 3 4 a diffusion index that implies that the aggregate series rises (50% — 100%) falls (100% — 50%) falls (50% — 0) rises (0 — 50%) increases at an increasing rate increases at a decreasing rate declines at an increasing rate declines at a decreasing rate In computing a numerical value for a given diffusion index, one-half of the percentage of the partici pants reporting no change is added to the percentage of participants reporting increases. Thus, if 30 per cent o f the participants report “increasing,” 40 percent report “ unchanged,” and 30 percent report “ decreasing,” the value of the diffusion index would be 50 percent. While an aggregate series measures the actual level attained by a particular business-activity indicator, such as employment, the diffusion index measures the percentage of firms participating in the expansion (or contraction) in any given m onth. The proper interpretation, then, is not that employment is expanding, for example, but that one-half of the firms are experiencing expanding employment. The amplitude of the diffusion index above (or below) the 50 percent level measures the intensity of the expansion (or contraction), while the intersection of the 50 percent level signifies a peak (or trough). The distance between peaks measures the duration of the cycle. In the special case illustrated above, the diffusion index curve is equivalent to the rate of change in the aggregate series. 1 1. A discussion of the diffusion process as it relates to aggregate series appears in Geoffrey H. Moore, “The Diffusion of Business Cycles,” in Geoffrey H. Moore, Ed., Business Cycle Indicators, vol. 1 (National Bureau of Economic Research/ Princeton University Press, 1961), pp. 261-81. Although a diffusion index is similar to a rate of change, there are fun damental differences. As indicated earlier, the diffusion index takes into account only the direction, not the magnitude, of change. Thus, if a general expansion is under way in a specific aggregate series, a diffusion index would show whether it has been spreading among firms. While the scope of an expansion in its early stages appears to be roughly correlated with the magnitude of the expansion, the same is not necessarily true for the rates of change in most eco nomic aggregates during expansions. A diffusion index can do no more than measure the scope of period-to-period changes. For a discussion of the conditions required to make a diffusion index proportional to a rate of change of a comparable aggregate series, see Geoffrey H. Moore, “Diffusion Indexes, Rates of Change, and Forecasting,” in Geoffrey H. Moore, Ed., Business Cycle Indicators, vol. 1 (National Bureau of Economic Research/Princeton Univer sity Press, 1961), pp. 282-93. ductions in inventories (that is, the rate of change) may lead the cycle, especially near the peaks.4 The lead-lag relationship among the salesoriented series depends upon its placement within the production process, beginning with the receipt of a new order through final shipment. New orders for durable goods particularly show a tendency to lead the business cycle in general (as well as to lead the output of the industry receiving the orders) because of the causal connection between commitment to buy and production.5 The amplitude o f fluctuations in new orders is augmented by two factors. First, firms tend to place orders with more than one m anu facturer to assure delivery and an adequate supply around the peak of an expansion. Second, firms are able to cancel orders when they see the first signs of an economic slowdown, a problem further exacer bated by double ordering. Firms producing to stock have some advantage in using inventories to adjust to changes in new orders, while firms producing to order must rely on more complex adjustment mechanisms. Shipments, especially for firms producing to order, are expected to run parallel with new orders, separated only by production time, which may vary from industry to industry. In addition to production time, however, the span between receipt and shipment of a new order can be altered by the supply con straints that result in backlogs and by changes in delivery time, which serve a role similar to inventories as a “buffer” to unexpected changes. Whether pro ducing to stock or to order, firms may attem pt to stabilize their production by allowing backlogs to accumulate before capacity is reached (especially if employment expansion is required) and only run them down after a slowdown begins. Backlogs, then, are generally designated as leading a cycle peak, but lagging a cycle trough. Changes in backlogs and shipments also appear in the adjustments of delivery time. Indeed, the ratio of backlogs to shipments represents the number of months needed to dispose of existing backlogs at present rates of operation (average time interval between an order’s receipt and delivery). Delivery time includes both the time spent in production and the time spent in backlogs. In a seller’s market, where advanced orders are strong, manufacturing firms have a degree of discretion over delivery dates. (Even if firms cannot fill orders in excess of capacity constraints, they still can receive them.) When de mand is rising, buyers seek early delivery dates; yet once a slowdown begins, buyers, too, are often willing to wait. In general, delivery time leads the cycle at both the peak and trough. Behavior of the 4D Indexes 4. Because of the uncertainty in supply and demand fluctua tions and the time lags involved in the production process, inventories often serve as a buffer. Inventories, however, may accumulate for two different reasons. Businessmen may be building inventories voluntarily in anticipation of expanding sales, or businessmen involuntarily may be accumulating inventories from overestimating that expan sion. In the first case, the change in inventories would be a leading indicator; in the second, a lagging indicator. Over the past ten years, the 4D indexes have been effective in monitoring economic activity in the nation and in the district. In comparing the 4D indexes with the corresponding aggregate series (expressed as rates o f change), the 4D indexes generally have proved to be statistically significant in explaining current and future levels of the aggregate series (see Appendix). Except for the inventory series, the 4D indexes are able to provide meaningful information on the behavior of the aggregate series beyond what is known from exam ining past rates of change in the aggregate series. In deed, the shipments index has proved to be more successful than relying on past rates of change in the shipments series for indicating current rates of change. 5. Many firms experience large cyclical fluctuations in de mand and react to them strongly through output adjust ments. Industries in which there are many such firms con stitute an important part of durable-goods manufacturing, which, in turn, is a sector that carries much weight within the economy as a whole. See Victor Zarnowitz, “The Timing of Manufacturers’ Orders During Business Cycles,” in Geoffrey H. Moore, Ed., Business Cycle Indicators, vol. 1 (National Bureau of Economic Research/Princeton Uni versity Press, 1961), pp. 420-84. The 4D indexes, however, are less successful in forecasting rates of change in the next m onth than an aggregate series. Only half of the indexes (employment prices, backlogs, and delivery time) exhibited statistical significance. In three of the remaining indexes (new orders, shipments, and the composite index), even knowledge of the past behavior of the series has proved to be unsatisfactory for forecasting rates of change in the subsequent month. Although confidence in the Table 2 Comparison of Turning Points: 4D Index and Comparable Series 1969-70 recession Peak NBER reference points Deviation3 Trough 1973-75 recession Deviation3 Dec. 1969 Nov. 1970 Nov. 1969 Oct. 1969 Oct. 1971 Nov. 1971 +11 Dec. 1969 Oct. 1969 Mar. 1971 Nov. 1970 +4 Peak Deviation3 Trough Deviation3 Mar. 1975 Nov. 1973 Employment Index Series Oct. 1975 June 1975 +7 +3 July 1975 Mar. 1975 +4 +18 May 1976 June 1976 +15 +10 Sept. 1974 Sept. 1974 +10 +10 May 1975 Apr. 1975 +2 Oct. 1974 Dec. 1974 +12 +14 Apr. 1975 Apr. 1975 +1 +1 Mar. 1976 Mar. 1976 +13 Apr. 1976 Mar. 1975 +14 May 1975 Mar. 1975 +3 Oct. 1974 Aug. 1974 +12 Oct. 1974 Dec. 1973 +12 0 Dec. 1972 Jan. 1972 +25 +14 Dec. 1974 Mar. 1975 +14 Oct. 1969 Dec. 1969 Jan. 1971 Dec. 1970 +2 +1 Nov. 1969 Dec. 1969 Jan. 1971 Dec. 1970 +12 +10 Hours Index Series +1 0 Inventory Index Series May 1970 Dec. 1970 +6 +12 New orders Index Series +1 Shipments Index Series +2 +1 Backlogs Index Series Nov. 1969 Dec. 1969 -1 0 -1 Oct. 1970 Dec. 1970 +1 Nov. 1969 Oct. 1969 Oct. 1970 J a n .1971 +2 Dec. 1969 Dec. 1969 Jan. 1971 Dec. 1970 +2 Sept. 1974 Nov. 1974 +11 Oct. 1974 Mar. 1974 +12 +13 +13 Delivery time Index Series -1 +5 0 Composite Index Series +3 Oct. 1974 Oct. 1974 +12 +12 0 a. Numbers indicate difference in months between NBER reference cycle turning points and individual indicators. The specific NBER turning points are December 1969 (peak), November 1970 (trough), November 1973 (peak), and March 1975 (trough). NOTE: Prices were excluded because no peaks were registered by the 4D index over the period studied. SOURCES: Department of Commerce (Bureau of Economic Analysis) and Federal Reserve Bank of Cleveland. forecasting ability of the 4D indexes may be limited, the survey generally provides information that is con sistent with the expected cyclical behavior of the aggre gate series, at least in terms of duration and amplitude. The success of the 4D indexes in capturing the pattern of fluctuation in the comparable aggregate series does not guarantee that the timing of a 4D index peak or trough will closely correspond to the peak or trough of the aggregate series. In fact, almost one-third of the turning points among the 4D indexes followed their expected turning points (as determined by the rate o f change in the comparable aggregate series) by more than a one-quarter time period (see table 2). Almost half of these deviations were concentrated in the inventory index, which has a weak correlation with the aggregate series. Of the remaining deviations, most were contained in the 1973-75 recession, which sharply contrasted with the more accurate signaling of turning points in the 1969-70 recession. During the 1969-70 recession, the actual devia tion of the 4D index peak and trough from the NBER reference cycle peak and trough tended to be no more than one to two months. At the reference cycle peak (December 1969), for example, all of the 4D indexes except inventories closely corresponded to the refer ence cycle peak and to the specific aggregate series peak. At the reference cycle trough (November 1970), all of the indexes except employment, hours, and in ventories closely tracked both the reference and the specific cycles. Both the index and the specific cycle for employment reached a trough about one year after the reference cycle, so that the index was still accu rately representing the employment series. The hours index reached a trough four months after the reference cycle and the specific series in both the 1969-70 and the 1973-75 recessions. The inability of the inventories index in the 1969-70 recession to coincide with the turning points of the aggregate series was perhaps expected from the relative weak correlation with the specific aggregate series. The index lagged the reference cycle peak by 6 months and the trough by 25 months, while the spe cific cycle of the aggregate series indicated a 12-month to 14-month lag at each turning point. Still, the index may indicate a pattern of behavior peculiar to the re gion. Inventory adjustments in the 4D may have lagged the reference cycle at the peak because o f difficulties in making prom pt production adjustments or hesitancy among 4D manufacturers to acknowledge that a reces sion was developing. At the trough, 4D manufacturers may have been cautious about building inventories until sales were increasing faster than production. Although the number of 4D indexes that devi ated substantially from the turning points of their counterparts increased during the 1973-75 recession, the aggregate series themselves tended to lag the refer ence cycle by roughly one year, especially at the peak (November 1973). However, the recession did not actually affect the 4D economy until late 1974, not the peak of late 1973 determined by the NBER. At the reference trough (March 1975), new orders, ship ments, backlogs, and the composite index behaved satisfactorily with respect to specific turning points, but the other indicators signaled turning points much later than the specific aggregate series. In addition to the inventories index, which re peated its earlier pattern of reaching a peak sooner and a trough later than its specific cycle, hours and delivery time indexes had the largest deviation from their specific cycles. Because the 1973-75 recession was in many respects atypical, the discrepancies in the hours and delivery-time indexes may reflect regional charac teristics of the recession. If much of the miscalculation that led to the overextension of inventories was con centrated among 4D manufacturers, the expansion of hours worked might have lasted longer, and the sub sequent contraction may have led to a later trough in the 4D than elsewhere in the nation. With production expanding, delivery times may have continued to be extended in the 4D, while new orders and backlogs were tapering off in other regions. If so, the 4D in dexes may be more indicative of the regional pattern of activity than the specific aggregate series. Despite the shortcomings of individual 4D in dexes, the 4D composite diffusion index has proved to be an effective indication of overall manufacturing activity. When compared with the FRB index, the composite index revealed a close correspondence with manufacturing activity, but its forecast was no better than could be obtained with past knowledge of the FRB index (see chart 3). The turning points and, thus, durations of the two cycles were almost perfectly co incident with the reference cycle. But the amplitude of the two cycles may have been exaggerated in the composite index, as the 4D economy was dispropor tionately affected during both the expansion and re cession that followed.6 Particularly, expansion that occurred during 1974 appears to be much more in tense in the composite index than suggested by the rate of change in the FRB index. Expansion was strongest in inventory and business fixed investment, hence in primary metals and metal-working industries that held up well into the recession. Indeed, both in dicators revealed a spurt of activity in the middle of 1974 that appeared to be largely concentrated in the 4D economy. However, the subsequent sharp contrac 6. The regression residuals reflect the fact that most of the indexes overestimated the decline in November and De cember of 1974, presumably because of the intensity of the inventory liquidation and decline in goods-producing industries in the 4D. The plot of the two indicators also shows the effects of the automobile and expected steel strikes in 1970 on the FRB index, but not on the composite index (which was, of course, controlled for in the correla tion tests with the dummy variable, Z). The residuals of the regression revealed other exogenous shocks experienced by individual series, but few were widespread among the indexes. Chart 3 Overall Manufacturing Activity Percent 70 (Dec.) P (Nov.) T (Nov.) P Percent (rate o f change) NOTE: Dotted line indicates anticipated values for m onth in progress. SOURCES: Board of Governors o f the Federal Reserve System and Federal Reserve Bank of Cleveland. tion in 1975 is captured by both indicators, even though the composite index is slower to recover, as might be expected in a widespread expansion. Finally, the anticipated values for the composite index have proved to be about as accurate as the actual values. Although no test of significance could be used to check the accuracy of anticipated values, their performance between June 1978 and December 1979 has been reasonably consistent with the actual values. The anticipated values survey has a tendency to overstate amplitude, which is not unusual in fore casting. However, the apparent accuracy of the antic ipated values suggests that the forecast could be greatly improved by including the anticipated value in the forecast model. By treating the m onth in progress as an actual value, knowledge of manufacturing activity is extended one full m onth, thereby improving the accuracy of the forecast. Since all of the indexes indi cate reasonably accurate anticipated values, many of the problems with forecasting, especially with the composite index, may be overcome by the inclusion of anticipated values as a contemporary value in an analysis similar to confirming current levels. 1979 in Perspective Turning points in peaks and troughs of business activity are difficult to identify and generally have eluded forecasters. Since early 1975, manufacturing activity was in an expansionary phase that has been characterized by spurtsof growth between brief pauses, at least until late 1979. Each pause raised apprehension among manufacturers at the onset of another recession. Economic behavior is still sufficiently perplexing that forecasters who had expected a slowdown in early 1979 (based on available indicators) were required continually to revise those expectations. The credible performance of the 4D indexes over the past ten years provides a foundation for improving the month-tom onth analysis of manufacturing activity. While not designed specifically to determine turning points, the 4D indexes can be useful warning signals because of their currency and corroborative value. For example, both the composite index and the FRB index captured the spurts and pauses o f the re cent expansion. However, the composite index only once signaled a potential contraction (January 1977), compared with four by the FRB index (presumably related to strikes or other temporary disruptions) since 1975. In September 1979, the composite index clearly showed a peak and subsequent contraction in overall manufacturing activity. In contrast, the FRB index indicated only uncertainty, with alternating positive and negative rates of change each m onth throughout the year. Although one-half of the eight individual in dexes in 1979 deviated from the pattern of their cor responding series, the remaining indexes conformed to their expected patterns. For example, hours, ship ments, and backlogs also reached peaks by midyear and have continued to indicate decline into 1980, while their comparable series have fluctuated irregu larly. The inventory index has indicated relative sta bility, while its comparable series has shown moderate accumulation. In each of these cases, the indexes have underestimated the strength of their corresponding series. Among the conforming indexes, most of the corresponding series were also in a contraction phase. Manufacturing employment, delivery time, and neworders indexes indicated decline throughout the second half of 1979 and only temporarily showed signs of strengthening in early 1980. The price index, along with its corresponding series, again has been the ex ception in that the pace of inflation occasionally has slackened, but it has never declined. The discrepancies between the indexes and their corresponding series can be attributed to inherent shortcomings in the survey. The 4D survey sample currently contains a disproportionate number of steel producers that, along with automotive producers, have been the leading edge of the recent economic slow down. Steel inventories have been under extremely tight controls and appear to be preventing the inven tory index from exhibiting the expected trend toward accumulation. Similarly, hours, backlogs, and ship ments appear to be contracting in the 4D indexes rather than being stable as expected because of the steel industry’s experience. However, the correspond ing series may also be faulty because the length of the recent expansion and capacity constraints have pro duced sizable backlogs of orders. Until these backlogs are sufficiently reduced, production (and, as a result, hours and shipments) can be sustained at current levels, even though other areas of the economy are weakening. Therefore, the composite index, by being more broadly based than the FRB index and less dom inated by capacity constraints, may in fact be a more accurate depiction of the current state of manufactur ing activity. Concluding Remarks After nearly a year of unfulfilled recession fore casts, the outlook for manufacturing activity was un certain. The need for current and reliable indicators of manufacturing activity is acute. The results of this investigation indicate that the 4D survey is a useful tool for monitoring the economy. To be sure, the in dexes must be used with caution, incorporating knowl edge of the 4D economy with experience in businesscycle analysis. Nevertheless, the information conveyed by the indexes is a valuable piece in the puzzle of where the district and national economy is and where it is likely to go. Appendix Description of the Models and Empirical Results Two basic models were constructed to determine whether the amount of information contained in the 4D index series constituted a significant improvement over the body o f knowledge already avail able from the historical pattern o f the aggregate series. Since the most inclusive measure of that body of knowledge can be presented by the past performance of the aggregate series itself, the models took the following form: Coincident model X f - a + b ^ Z + b 2X f l + b 3X { 2+ b4D If + b5D It b6D I{ 2+ u Forecast model where X = rate of change in aggregate series Z = control variable for automobile and expected steel strikes in November-December 1970 D I = one of eight diffusion indexes for current time period, t, and for two previous time periods. Tests were conducted on all the indexes, using a Cochrane-Orcutt iterative technique to minimize autocorrelation problems. The models test the ability of the indexes to conform to the rate of change of an aggregate series (coincident model) and to predict the rate of change of the aggregate series in the next period (forecast model). The null hypothesis that the coefficients of the diffusion indexes are zero and, therefore, contribute no useful information is rejected if the coefficients are statistically sig nificant. (Although lags in data availability differ for each aggregate series, only one form of the model that gave the diffusion index a one-month availability advantage was tested for consistency.) The statis tical results for the period between January 1970 and January 1979 appear in table A. The comparable series selected for the comparison were as follows: Production-oriented Employment: Employment (U. S.)—manufacturing Hours: Weekly hours o f product workers—manufacturing Prices: Producer Price Index—durable-goods manufacturing Inventory: Inventory stock—durable-goods manufacturing Sales-oriented New orders:! Diffusion index for new orders of durable-goods manufacturing Backlogs: Unfilled orders—durable-goods manufacturing Delivery tim e:l Vendor performance—percentage of companies reporting slower deliveries Shipments: Shipments—durable-goods manufacturing 1. New orders and delivery time are not based on the aggregate series. Because the series are themselves diffusion in dexes, rates of change need not be computed. Table A Coefficients of Regression Equations Key: a = constant R2 = coefficient of determination DW = Durbin-Watson statistic Production-oriented indexes Employment • Coincident model {Xf) Sales-oriented indexes Hours Prices Inventories New orders Backlogs Delivery time Ship ments Composite index X t.\ 0.140 (1.87) 0.092 (0.96) 0.073 (0.81) 0.538 (5.72) -0.367 (-4.42) 0.773 (8.39) 0.562 (6.22) 0.053 (0.61) 0.291 (3.07) Xt-2 0.181 (2.32) -0.268 (-2.78) 0.230 (2.77) 0.189 (2.14) -0.114 (-1.46) -0.079 (-0.88) -0.171 (-1.99) 0.047 (0.54) 0.154 (1.78) D It 0.002 (2.29) 0.003 (2.42) 0.000 (1.99) 0.000 (1.41) 0.009 (4.87) 0.001 (5.58) 0.006 (2.10) 0.001 (3.57) 0.001 (5.01) D It -1 0.000 (0.72) -0.002 (-0.94) 0.000 (4.97) -0.000 (-0.00) -0.002 (-0.80) -0.001 (-3.68) 0.002 (0.38) -0.001 (-2.34) -0.001 (-3.22) DIt -2 -0.000 (-1.53) -0.000 (-0.32) -0.000 (-0.08) 0.000 (1.10) -0.000 (-0.16) 0.003 (2.46) -0.008 (-2.90) 0.001 (1.93) -0.000 (-0.08) Z -0.024 (-9.24) -0.018 (-0.43) 0.006 (3.15) 0.004 (1.52) 0.179 (2.96) -0.008 (-1.96) -0.094 (-1.60) -0.069 (-5.47) -0.022 (-3.93) a -0.010 (-2.36) -0.004 (-1.48) -0.018 (-3.97) -0.010 (-3.27) 0.453 (4.78) -0.013 (-4.03) 0.045 (1.13) -0.037 (-2.23) -0.002 (-0.39) 0.699 0.145 0.710 0.709 0.488 0.824 0.301 0.311 0.579 1.969 R2 2.020 DW Forecast model (X^+i) Xf. ] 0.101 (1.33) 2.001 1.991 2.028 2.019 2.028 2.150 1.993 -0.202 (-2.02) 0.234 (2.56) 0.279 (3.27) 0.071 (0.80) 0.108 (1.08) 0.046 (0-51) 0.054 (0.57) 0.059 (0.58) Xt-2 -0.131 (-1.75) -0.104 (-1.05) 0.030 (0.36) 0.262 (3.07) 0.067 (0.78) 0.110 (1.06) 0.000 (0.01) 0.008 (0.08) 0.089 (0.94) DIt 0.003 (2.83) 0.000 (1.45) 0.000 (5.41) 0.000 (1.78) 0.004 (1.72) 0.000 (1.94) 0.008 (2.48) -0.000 (-0.34) -0.000 (-0.16) DIt.x -0.001 (-0.86) -0.000 (-0.43) 0.000 (0.94) 0.000 (0.55) -0.003 (-0.94) -0.000 (-0.34) -0.000 (-0.13) 0.000 (1.10) 0.000 (0.33) DIt -2 0.000 (0.31) -0.000 (-0.50) -0.000 (-0.63) 0.000 (1.23) 0.002 (0.90) 0.000 (2.35) -0.009 (-3.01) 0.000 (0.03) -0.000 (-1.01) Z -0.023 (-9.88) -0.002 (-0.42) 0.006 (3.47) 0.003 (1.22) -0.029 (-2.71) -0.004 (-1.08) -0.070 (-1.05) -0.071 (-5.24) -0.022 (-4.06) a -0.012 (-2.09) -0.002 (-0.70) -0.017 (-3.31) -0.015 (-3.22) 0.331 (4.05) -0.021 (-2.65) 0.127 (2.05) -0.012 (-0.59) 0.013 (1.18) Rl 0.688 0.102 0.701 0.704 0.374 0.768 0.269 0.222 0.481 DW 2.090 2.009 1.962 1.975 2.002 1.824 1.978 1.988 2.021 NOTE: Values in parentheses are ^-statistics. The Economic Review is published quarterly by the Research Department of the Federal Reserve Bank of Cleveland. The views stated in the Economic Review are those of the authors and not necessarily those of the Federal Reserve Bank of Cleveland or of the Board of Governors of the Federal Reserve System. Material in the Economic Review may be reprinted provided the source is credited. Please provide the bank’s Research Depart ment with copies o f reprinted materials. For free copies of this and other Federal Reserve publications, contact the Public Information Center, Federal Reserve Bank of Cleveland, P. 0 . Box 6387, Cleveland, OH 44101. Telephone: (216) 241-2800.