The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
orKing raper series Inventories and output volatility Paula R. W orthing to n Working Papers Series Research Department Federal Reserve Bank of Chicago Decem ber 1998 (W P -98-21) FEDERAL RESERVE B A N K O F CHICAGO Inventories and output volatility Paula R. Worthington Economic Research Department Federal Reserve Bank o f Chicago 230 South LaSalle Chicago, Illinois 60604 (312) 322-5802 prw@fihchi.org This draft: December 1998 Abstract: Analyzing disaggregate data on inventories and sales from the U. S. manufacturing and trade sector between 1960 and 1997 yields four main findings. First, I find that IS ratios are somewhat lower after 1984:1 among durable goods manufacturers and durable goods retailers outside the motor vehicle industry. Second, I find that industries which have lowered their IS ratios tend to be those in which the variance o f output relative to sales has declined. Third, by decomposing the variance o f output into its components, I find that the variance o f sales is less important, and the variance o f inventory investment is more important, after 1984:1 than in earlier years for the overall manufacturing and trade sector. Finally, the evidence suggests that industries where IS ratios fell are those where inventory investment volatility played a smaller role in output volatility in the later period. The author would like to thank Thomas Klier, Helen Koshy, David Marshall, Dan Sullivan, and seminar participants at the Federal Reserve Bank of Chicago and the October 1998 meetings of the Illinois Economic Association for valuable discussions and comments and Kenneth Housinger for excellent research assistance. The views expressed in this paper are strictly those of the author, and they do not necessarily represent the position of the Federal Reserve Bank of Chicago or the Federal Reserve System Introduction Are business cycles less pronounced now than in earlier years? Several studies offer evidence o f decreased aggregate volatility in recent years, and business analysts, too, often claim that future business cycles are likely to feature shorter, less pronounced contractions than earlier cycles displayed. For example, McConnell and Quiros (1997) present informal evidence that post-war GDP volatility declined in the early 1980s, and they specifically find evidence o f a oneĀ time decline in the volatility o f post-war GDP in 1984:1. Suggested reasons for such changes in output volatility or cyclicality are many and varied, but one item on nearly every "short list" o f factors is the widespread embrace o f just-in-time inventory management techniques by U.S. firms. For example, the E conom ist (1998) writes: What is dear, however, is that the economic cycle has become less bumpy than it used to be...There are several possible explanations for the taming o f the business cycle....[including] better inventory control through just-in-time techniques and the use o f computers. Similarly, McConnell and Quiros (1997) point to a decline in the share o f inventory investment in GDP fluctuations as a possible source o f the output volatility decline. In this paper, I investigate the relationship between inventory holdings and output volatility at the industry level. I use detailed data from the U.S. manufacturing and trade (M&T) sector from 1960 to 1997, and I relate inventory-sales ratios to several measures o f output and inventory volatility. I focus on output volatility because swings in business inventory accumulation have historically accounted for large fractions o f GDP volatility. In particular, I take as given the breakpoint identified by McConnell and Quiros (1997) and compare inventory behavior before and after 1983:4. In briefj I find that IS ratios are somewhat lower in the later period among durable goods manufacturers and durable goods retailers outside the motor vehicle industry. I also find that in the manufacturing and retail sectors, the industries which have lowered their IS ratios tend to be the ones whose output variance (relative to the variance o f sales) has declined. Third, by decomposing the variance o f output into its components, I find that the variance o f sales is less important, and the variance o f inventory investment is more important, in the later period than in the earlier period for the overall manufacturing and trade sector. One prominent exception in the manufacturing sector is the motor vehicles industry, where inventory investment variance declined. In retailing, the contribution o f inventory investment variance nearly trebled, rising from 20.7% to 59.3% between the earlier and later periods, suggesting an increased role for inventory investment fluctuations in that sector. Overall, the evidence suggests that industries where IS ratios fell are those where inventory investment volatility played a smaller role in output volatility in the later period. Facts and theories about inventories Economists care about inventory behavior because, historically, swings in inventory investment have played a prominent role in cyclical fluctuations. In brieĀ£ inventory investment is highly volatile and contributes significantly to recessionary declines in GDP, and inventory-sales (IS) ratios are strongly countercyclical, rising during recessions and falling in expansions. In fact, inventory disinvestment is a central part o f cyclical contractions. Table 1 reports the average post-war contribution o f changes in inventory accumulation ("inventory disinvestment") to the peak-to-trough decline in GDP during contractions. The table shows that the decline in inventory investment accounted for 76 percent o f GDP's decline in the average post-war recession. The table reveals three features o f the data. First, the manufacturing and retail sectors dominate the wholesale trade sector, accounting for most o f the inventory effect, with retailers accounting for 2 about one- third (,26/.76) o f the total contribution.1 Second, firms in the durable goods sectors account for most o f the impact. Third, during the two most recent recessions, the role o f durable goods manufacturers was quite muted, as their inventory disinvestment during those episodes accounted for a below-average 11% o f the total contraction in GDP. In contrast, the retail sector was o f little consequence in the 1981-1982 recession, but key in the 1990-1991 recession. On balance, table l's evidence suggests that durables goods inventories held by manufacturers and retailers are key to any analysis o f the cyclical behavior o f inventories. The tw o major competing models o f inventory behavior, production smoothing models and S,s threshold-type models, offer competing predictions about tw o key aspects o f inventories, namely the variance o f output relative to sales and the correlation between sales and inventory investment; see Fitzgerald (1997) or Homstein (1998) for useful discussions.2 In production smoothing models, output is predicted to be less (more) variable than sales, when shocks are solely on the demand (cost) side. Such models also typically predict a negative covariance between sales and inventory investment. In contrast, generalized S,s models do not offer predictions on these points (Homstein (1998)), though with specific assumptions about aggregation and other model features, such models do offer specific predictions. For example, McCarthy and Zakrajsek (1997) develop an S,s model in which output is predicted to be more ^Although not shown in the table, the shares for the three categories o f inventories held by manufacturers, namely materials, goods in process, and finished goods, confirm Blinder and Maccini's (1991) finding that finished goods inventories account for little o f the total contribution (6% post-war average) despite being the focus o f much economic research. In contrast, goods in progress and, to a lesser extent, materials and supplies held by manufacturers are more important. ^ o r retailers and wholesalers, the terms "production" and "output" are taken to mean deliveries o f goods from their suppliers. In this paper, I will use these terms interchangeably to denote output in the manufacturing sector and deliveries in the trade sector. 3 variable than sales and in which sales and inventory investment should be uncorrelated. Data The data used inthis paper are quarterly inventories and sales (shipments) data, in chained 1992 billions ofdollars, from the U.S. Department ofCommerce, for the manufacturing and trade sectors. Manufacturing includes 21 separate industries (essentially2-digit SIC industries), the merchant wholesale sector includes 19 industries, and the retailtrade sector includes 13 sectors.3 I construct output as the sum of shipments and inventory investment, and the inventory-sales ratio is denominated in months. Because itisthe business cycle aspects ofinventory investment which are of most interest, the data are detrended using the Hodrick-Prescott filter.4 Have inventory-sales ratios fallen? Table 2 reports inventory-sales (IS) ratios by sector over several alternative time periods, with period means reported in columns 1 through 3 and cyclical highs reported in columns 4 through 6.s Turning firstto the means, we find that IS ratios are higher in durable goods industries than in nondurable goods industries and that ratios are highest among durable goods manufacturers. Further, comparing the earlyperiod (1960:1-1983:4) to the later period (1984:11997:4), we see that IS ratios have not fallen overall; in fact, they have risen. Only among durable goods manufacturers, on average, did IS ratios fall,with the greatest declines occurring in SIC 3Data on the detailed sectors isavailable in manufacturing and retail from 1959 onwards; detailed wholesale sectors have data only from 1967 forward. Unless otherwise noted, the paper's calculations will use the 1960-1997 period, thus will include wholesale trade only at the level of durable and nondurable goods. 4See Hodrick and Prescott (1997), and also see Homstein (1998) for a more general discussion of detrending, extracting the appropriate frequencies from the data, and so on. 5The ratios are constructed from the data prior to detrending. 4 industries 35 (industrial and commercial machinery and computer equipment), 371 (motor vehicles and equipment), and 38 (instruments). In the retail sector, motor vehicle IS ratios rose from 1.55 to 1.85 months, while allother durable goods retailers saw IS ratios fall. Columns 4 through 6 offer another perspective on whether IS ratios have dropped in recent years. The table shows that the overall M & T IS ratio peaked at about the same point in each ofthe three recessions reported inthe table: 1.48 in the 1973-1975 recession, and 1.53 in the 1981-1982 and 1990-1991 recessions. From that perspective, littlehas changed. However, durable goods firms saw the cyclical highs fallin the last recession relative to the preceding one, especially inthe manufacturing sector. Furthermore, the three broadest sectors exhibit different patterns: in manufacturing, the cyclical maximum fell; in wholesale, itwas basicallyunchanged; and in retail, itrose. Again, one interesting aspect ishow motor vehicle-related inventories behaved: in manufacturing (SIC 371), the cyclical high fellfrom 1.06 to 0.86, comparing the 1981-1982 and 1990-1991 recessions, while inthe retailtrade sector, motor vehicle inventories reached cyclical highs of 1.92 in 1981-82 and 2.08 in 1990-1991. On balance, then, the evidence points to declining IS ratios among durable goods manufacturers and durable goods retailers excluding motor vehicles. Outside ofthese groups, IS ratios were at best flat, at worst up somewhat. This compares to earlierwork by Ben Salem and Jacques (1996) and Hirsch (1996), who find that inventory-sales (IS) ratios have declined in the manufacturing sector, but that ratios have risen in the wholesale and retailtrade sectors. The variance of output relative to sales In this section ofthe paper, I examine the variance of output relative to the variance of sales, and I relate this relative output variance to IS ratios. In brief I find some evidence that 5 industries with high IS ratios are those whose output variance isrelativelyhigh. I also find that in the manufacturing and retail sectors, the industries which have lowered their IS ratios tend to be the ones whose relative output variance has declined; the opposite seems to be true in the wholesale sector. This establishes, in an unstructured way, a connection between lower IS ratios and decreased output volatility, at leastin the manufacturing and retail sectors. Table 3 reports the ratio ofthe variance of output to the variance of sales for the broad sectors studied here. The ratio exceeds 1 in allcases, as output ismore volatile than sales. This is especially true inthe durable goods sectors. Comparing the early and later periods, I find that output volatilityrelative to sales has risen in allcases.6 However, the disaggregate data indicate that eight ofthe 21 manufacturing industries experienced declines, most notably several durable goods industries, including SICs 32 (stone, clay and glass), 35 (industrial machinery and computer equipment), 36 (electronic equipment), 371 (motor vehicles), and 37-exduding 371 (all other transportation equipment). In the retail sector, although only one disaggregate industiy (other durable goods retailers) showed a decline in relative output variability, the overall retail sector excluding motor vehicles experienced a decline from 3.12 to 2.05. This highlights the importance of retail motor vehicles, inwhich output variabilityrose from 2.74 to 4.16. Note the overlap between the sectors where IS ratios have declined and sectors where output volatilityhas declined: for example, SIC 35 (industrial machinery) had itsmean IS ratios fellfrom 3.32 to 2.14, while SIC 371 (motor vehicles) had itsIS ratio fallfrom 0.98 to 0.63; both industries experienced declines in output volatility. In the retail sector, motor vehicle IS ratios rose from 1.55 to 1.85 as 6As we shall see intable 5 below, the variance of output has declined in absolute terms; table 3's relative variance measure has risenbecause the variance of sales has declined even further. 6 output volatilityrose considerably. For motor vehicles, itappears that IS ratios and output volatilityrelative to sales volatilityhave fallenin the manufacturing sector but risen in the retail trade sector. Table 4 relatesthe relativevariance of output to IS ratios in a more formal way. For each detailed sector, I compute the relativevariance of output, firstover the fullsample period and then separately for the early and laterperiods. I also compute the mean and maximum IS ratio for those time periods. Table 4'stop panel reports the cross-sectional correlation coefficients between the IS ratio and the relative variance measure, for allindustries together as well as separately for the three broad sectors. In manufacturing and retail, the correlation islarge and positive, suggesting that high IS ratios are associated with high output volatility; the correlation is weaker in the wholesale sector. The bottom panel addresses the issue ofwhether those sectors that lowered their IS ratios are those whose output volatility declined. For each industry, I calculate the ratio ofthe early to laterperiod output volatility, the ratio ofthe early to lata*mean IS ratios, and the ratio ofthe 1981-1982 cyclical high to the 1990-1991 cyclical high IS ratio. The table'sbottom panel reports the cross-sectional correlation between the early to later output volatilityratio and the earlyto laterperiod IS ratio. Again, manufacturing and retail show positive correlations, suggesting that industries whose IS ratios fellare indeed those industries whose output volatility (relativeto sales) declined. The wholesale sector issomewhat different, showing a negative correlation. Covariance of sales and inventory investment Table 5 contains the covariances between sales (S) and inventory investment (CB1) for the broad sectors studied here, again for the fullperiod as well as for the early and later periods. In 7 allcases, the covariance ispositive over the foil sample period, and the covariance declines between the early and laterperiods. In retail, the covariance actually becomes slightlynegative in the laterperiod, implying that inventory investment declines when sales are rising. Because we are ultimately interested in output volatility, decomposing the variance of output into itscomponents isuseful. Since output isthe sum of sales and inventory investment, the variance ofoutput equals the sum ofthe variance of sales, the variance of inventory investment, and twice the covariance between sales and inventory investment. Table 6 reports, in levels and in percent terms, the components for the overall manufacturing and trade sector for the different time periods studied. Several patterns emerge from the table. First, the variance ofoutput has declined; this is true for the disaggregate industries in the manufacturing and wholesale trade sectors, as well as for the retail sector excluding motor vehicles. Much ofthe decline isdue to a decline in the variance of sales, which occurred in allbut eight ofthe 34 manufacturing and retail industries. In percentage terms, the variance of sales isless important in the laterperiod, accounting for 50.9% of total output variance, compared to 66.2% of the total during the earlierperiod. Second, the variance ofinventory investment has risen for the M & T sector overall, as well as for the manufacturing, wholesale trade, and retailtrade sectors independently. However, nine of 21 manufacturing industries show a decline inthe variance of inventory investment, though as a share oftotal output variance, inventory investment variance has risen in nearly allindustries. The one prominent exception is, again, SIC 371, the motor vehicles industry, where the share oftotal output variance accounted for by inventory investment variance fellfrom 7.8% to 3.5% between the earlier and laterperiods. In retailing, the variance of inventory investment rose in allindustries 8 but one (lumber stores), and as a percentage of total output variance, the contribution of inventory investment variance rose from 20.7% to a whopping 59.3% between the earlier and laterperiods. In that sense, we can say that inventory investment volatility has become a more pronounced factor in the retail sector. Furthermore, this increase isnot solely due to motor vehicles; itisprominent throughout the sector. In fact, computing correlations between IS ratios and the shares of inventory investment variance in total output variance, similarto the exercise in table 4, shows that inthe manufacturing and retail sectors, the industrieswhere IS ratios fellthe most are those in which inventory investment variance accounted for smaller shares oftotal output variance; the correlation isespecially strong in the retail sector. In the wholesale trade sector, the correlation is negative. Discussion and conclusions In thispaper, I use detailed manufacturing and trade sector data to examine several measures ofinventory behavior before and after 1984:1, a point identified by previous researchers as the time of a one-time decline in G D P volatility. Because movements inwholesale trade inventory investment are, on average, less important in business cycle fluctuations than are movements in manufacturing and retailinventory accumulation, I emphasize the key results from the lattertwo sectors. First, I find some evidence that IS ratios were lower after 1984:1 than inthe earlier period among durable goods manufacturers and durable goods retailers excluding motor vehicles; outside ofthese groups, IS ratios were at best flat, at worst up somewhat. Second, output ismore variable than sales in allindustries over each time period examined, by and large consistent with 9 previous research. Comparing the early and later periods, I find that output volatilityrelative to sales has risen overall, but that several durable goods manufacturing industries show declines, noticeably several whose IS ratios have declined over time. In fact, simple correlations show that in the manufacturing and retail sectors, the industries which have lowered their IS ratios tend to be the ones whose relative output variance has declined; the opposite seems to be true in the wholesale sector. Third, I decompose the variance of output into itscomponents and find that much ofthe decline in output variability after 1984:1 isdue to declines in salesvariability, which, in percentage terms, isless important in the later period. The share oftotal output variance accounted for by inventory investment variance has risen for the M & T sector overall. One prominent exception in the manufacturing sector isthe motor vehicles industry, where inventory investment variance became less important inthe later period. In retailing, the contribution ofinventory investment variance nearly trebled, rising from 20.7% to 59.3% between the earlier and later periods, suggesting an increased role for inventory investment fluctuations in that sector. More formally, correlations between IS ratios and the shares accounted for by inventory investment variance are positive inthe manufacturing and retail sector, suggesting that industries where IS ratios fellare those where inventory investment volatility played a smaller role in output volatilityin the later period. Finally, the motor vehicle industry stands out as sector worth further study. In the manufacturing sector, motor vehicle IS ratios fell, output volatility fell, and inventory investment volatilitybecame less important a factor in overall output volatility. In the retail motor vehicle sector, the opposite was true on all counts. Ifinventories and volatility have just been pushed 10 "downstream", then itishard to argue that, for the economy as a whole, changes in inventory management in one sector ofthe economy imply smoother aggregate output paths in the years ahead. In conclusion, this paper has established a cross-sectional correlation between IS ratios, output volatility, and inventory volatility. This isa useful first step in addressing the extent to which recent changes in inventory management techniques may have "tamed" the business cycle. Of course, as Homstein (1998) notes, attributing overall inventory investment volatilityto individual sectors isdifficultbecause ofthe covariance across sectors, and I cannot conclude that changes in inventory management techniques, as revealed through lower IS ratios, are responsible for declines in output volatility. However, the cross-sectional evidence does point to a connection between lower inventory holdings and decreased output volatility. Future research must address the covariance issue to make more progress in understanding the implications of new inventory management techniques forthe business cycle. 11 References Ben Salem, Melika, and Jean-Francois Jacques, "About the stabilityof the inventory-sales ratio: an empirical study with U.S. sectoral data." A p p lie d E c o n o m ic s L e tte rs 3 (1996): 467469. Blinder, Alan S., and Lotus J.Maccini, "Taking Stock: A Critical Assessment ofRecent Research on Inventories." J o u r n a l o f E c o n o m ic P e rs p e c tiv e s 5,1 (Winter 1991): 73-96. T h e E c o n o m is t. "The business cycle: puncture ahead." December 5,1998, p. 90. Fitzgerald, Terry J., "Inventories and the Business Cycle: An Overview." Federal Reserve Bank of Cleveland, 33,3 (1997): 11-22. E c o n o m ic R e v ie w , Hirsch, Albert A., "Has inventory management in the U.S. become more efficient and flexible" A macroeconomic perspective." I n te r n a tio n a l J o u r n a l o f P r o d u c tio n E c o n o m ic s 45 (1996): 37-46. Hodrick, Robert J.,and Edward C. Prescott. "Postwar U.S. Business Cycles: An Empirical Investigation." J o u r n a l o f M o n e y , C r e d it , a n d B a n k in g 29,1 (February 1997): 1-16. Homstein, Andreas. "Inventory investment and the business cycle." Federal Reserve Bank of Richmond E c o n o m ic Q u a r te r ly 84,2 (Spring 1998): 49-71. McCarthy, Jonathan, and Egon Zakrajsek, "Microeconomic Inventory Adjustment and Aggregate Dynamics." Working paper, Federal Reserve Bank ofN e w York, November 1998. McCarthy, Jonathan, and Egon Zakrajsek, "Trade Inventories." Working paper, Federal Reserve Bank ofN e w York, December 1997. McConnell, Margaret M , and Gabriel Perez Quiros, "Output Fluctuations in the United States: What Has Changed Since the Early 1980s?" Working paper 9735, Federal Reserve Bank ofN e w York, November 1997. 12 Table 1 Inventory investment's share of recessionary declines in GDP Percent mean Total change in business inventories manufacturing durable goods nondurable goods merchant wholesale durable goods nondurable goods retail durable goods nondurable goods 76 35 28 7 5 5 0 26 23 4 1981:1-1982:4 31 12 12 -0 3 2 1 -1 -7 7 1990:3-1991:1 49 6 10 -3 6 1 4 28 28 0 Notes: raw data are in billions of chained 1992 dollars. Shares are computed by sector for each postwar recession; the mean over all recessions is reported in column 1, and shares for the most recent two recessions are reported in columns 2 and 3. Table 2 Inventories-Sales Ratios Number of months Ma&mum ratio Meaq,ratjQ 1960:11997:4 1960:11231:4 1984:11997:4 1973:4197&1 1981:31982:4 1990:31991:1 1.36 1.33 1.42 1.48 1.53 1.53 1.51 1.52 1.50 1.78 1.78 1.66 manufacturing-durable gds 1.93 1.96 1.86 2.39 2.47 2.15 manufacturing-nondurable gds 1.13 1.13 1.13 1.22 1.20 1.18 1.15 1.07 1.30 1.14 1.36 1.38 wholesale trade-durable gds 1.56 1.50 1.67 1.77 2.09 1.80 wholesale trade-nondurable gds 0.79 0.71 0.94 0.70 0.82 0.98 1.24 1.14 1.41 1.31 1.31 1.47 retail trade-durable gds 1.94 1.91 2.00 2.32 2.22 2.20 retail trade-nondurable gds 0.91 0.82 1.08 0.90 0.93 1.07 manufacturing and trade manufacturing wholesale trade retail trade Columns 1-3 report the mean inventories-sales (IS) ratio, in monhts, for the time periods listed in the column headings. Columns 4-6 report the maximum IS ratio reached in the three contractions listed in the column headings. Table 3 Ratio of Variance o f Output to Variance o f Sales 1960:1- 1984:1- 1S32A 1S2L4 1.59 1.51 1:96 1.63 1.55 2.13 manufacturing-durable gds 1.88 1.79 2.24 manufacturing-nondurable gds 1.34 1.24 2.07 1.48 1.38 1.95 wholesale trade-durable gds 1.65 1.50 2.23 wholesale trade-nondurable gds 1.50 1.36 1.79 1.94 1.81 2.24 retail trade-durable gds 2.48 2.20 3.02 retail trade-nondurable gds 1.76 1.75 1.78 1960:11997:4 manufacturing and trade manufacturing wholesale trade retail trade Table 4 Correlation between inventories-sales ratio and relative variance o f output Correlation between Var(Q)/Var(S) and: full sample manufacturing wholesale* retail Mean IS ratio .585 .682 .274 .688 Maximum IS ratio .556 .742 .246 .693 Correlation between (Va^QyVar^X^Va^QyVar^S))*, and: full sample manufacturing wholesale* retail Mean IS ratio^^/ Mean IS ratio^ .306 .253 -.178 .436 Max IS rattOiMcj.iMj.j/ Max IS ratiotM03.IMi.i .175 .288 -.200 .178 The IS ratios and and relative variance measures are computed over the 1967:1-1997:1 period for the wholesale trade sector, all others use data from 1959:1-1997:4. Notes: Var(QyVar(S) and IS ratios are computed separately by industry. The top panel reports the cross-sectional correlation between the mean (maximum) IS ratio and the relative output variance (Var(Q)/Var(S)). For the bottom panel, the ratio of the early period (1960:1-1983:4) relative variance to the later period (1984:1-1997:4) relative variance is computed; similarly, the ratio of the early to later period IS ratio is computed. The table reports the correlation between these two ratios (line 3), as well as the correlation between the relative variance ratio and the ratio ofthe cyclical maxima (1981-1982 vs. 1990-1991). Table 5 Covariance o f sales and inventory investment 1960:11997:4 1960:1- 19?3;4 1984:11997:4 27.75 32.48 18.27 9.30 11.35 5.41 manufacturing-durable gds 5.82 7.14 3.30 manufacturing-nondurable gds 0.28 0.33 0.18 1.26 1.55 0.66 wholesale trade-durable gds 0.80 1.01 0.38 wholesale trade-nondurable gds 0.11 0.06 0.19 1.14 1.91 -0.32 retail trade-durable gds 0.37 0.79 -0.41 retail trade-nondurable gds 0.13 0.27 -0.14 manufacturing and trade tTiflUiifflcfiipng wholesale trade retail trade Note: the covariance of sales and inventory investment is computed separately for each sector over the time periods indicated.