Professional Documents
Culture Documents
Gaur Et Al 2005 An Econometric Analysis of Inventory Turnover Performance
Gaur Et Al 2005 An Econometric Analysis of Inventory Turnover Performance
Management Science
Publication details, including instructions for authors and subscription information:
http://pubsonline.informs.org
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@informs.org.
The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.
© 2005 INFORMS
With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org
MANAGEMENT SCIENCE informs ®
Vishal Gaur
Leonard N. Stern School of Business, New York University, 44 West 4th Street, New York, New York 10012,
vgaur@stern.nyu.edu
Marshall L. Fisher
The Wharton School, University of Pennsylvania, Jon M. Huntsman Hall, 3730 Walnut Street,
Philadelphia, Pennsylvania 19104-6366, fisher@wharton.upenn.edu
Ananth Raman
Harvard Business School, Morgan Hall, Soldiers Field, Boston, Massachusetts 02163, araman@hbs.edu
I nventory turnover varies widely across retailers and over time. This variation undermines the usefulness of
inventory turnover in performance analysis, benchmarking, and working capital management. We develop
an empirical model using financial data for 311 publicly listed retail firms for the years 1987–2000 to investigate
the correlation of inventory turnover with gross margin, capital intensity, and sales surprise (the ratio of actual
sales to expected sales for the year). The model explains 66.7% of the within-firm variation and 97.2% of the
total variation (across and within firms) in inventory turnover. It yields an alternative metric of inventory
productivity, adjusted inventory turnover, which empirically adjusts inventory turnover for changes in gross
margin, capital intensity, and sales surprise, and can be applied in performance analysis and managerial decision
making. We also compute time trends in inventory turnover and adjusted inventory turnover, and find that
both have declined in retailing during the 1987–2000 period.
Key words: benchmarking; inventory turnover; retail operations; performance measures
History: Accepted by Wallace J. Hopp, design and operations management; received November 21, 2003. This
paper was with the authors 4 months for 3 revisions.
Figure 1 Plot of Annual Inventory Turns vs. Annual Gross Margin account of the correlations with the explanatory vari-
for Four Consumer Electronics Retailers for the Years ables. We find that, on average, inventory turnover
1987–2000
in retailing has declined during the 1987–2000 period,
10 even though it is positively correlated with capital
intensity, and capital intensity has increased dur-
9
ing this period. However, there are marked differ-
8 ences in the time trends in inventory turnover across
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
6
inventory turnover.
5
Third, using the estimates from our model, we con-
4 struct an alternative metric of inventory productivity,
adjusted inventory turns, which empirically models
3
the trade-off between inventory turnover and the
2 explanatory variables. This metric can be applied in
performance analysis, benchmarking, and manage-
1
rial decision making because it enables comparison
0 of inventory productivity across firms and years. We
0 0.1 0.2 0.3 0.4 0.5 0.6 illustrate its interpretation with examples.
Gross Margin There is considerable interest in the operations
Best Buy Co. Inc. Circuit City Stores CompUSA Radio Shack
management community in evaluating time trends in
inventory turnover and assessing the impact of opera-
plots the annual inventory turnover of the above four tional improvements on operational and financial per-
consumer electronics retailers against their gross mar- formance. However, there are few empirical studies
gins (the ratio of gross profit net of markdowns to on these topics. Balakrishnan et al. (1996) compare
net sales) for the period 1987–2000. The figure shows the performance of a sample of 46 firms that adopted
a strong correlation between inventory turnover and just-in-time processes (JIT) during 1985–1989 with a
gross margin. Such correlation could possibly be matched sample of 46 control firms, and find that JIT
caused by many factors studied in the operations liter- firms achieved larger improvements in their inven-
ature, such as differences in variety and price. It raises tory turns. Billesbach and Hayen (1994), Chang and
the question of whether inventory turnover should be Lee (1995), and Huson and Nanda (1995) also study
used, per se, in performance analysis. the impact of JIT on inventory turns for different sam-
This paper uses public financial data to conduct a ples of firms. Hopp and Spearman (1996, Chapter 5)
descriptive investigation of inventory turnover per- summarize the findings of several survey-based stud-
formance in retail services. We identify the follow- ies on whether U.S. manufacturing firms that imple-
ing variables that should be correlated with inventory mented MRP systems achieved better inventory turns
turnover and can be measured from public financial as a result. Hendricks and Singhal (1997, 2001) exam-
data: gross margin, capital intensity (the ratio of aver- ine the impact of implementation of total quality
age fixed assets to average total assets), and sales sur- management programs on the operating incomes and
prise (the ratio of actual sales to expected sales for shareholder values of firms.
the year). Using results from the existing literature, In contrast to the above research, Rajagopalan and
we formulate hypotheses to relate these variables to Malhotra (2001) use aggregate industry-level data
inventory turnover. We then propose an empirical from the U.S. Census Bureau for 20 industrial sec-
model to represent these relationships and apply it to tors for the period 1961–1994 to determine whether
a panel of retailing data. the inventory turns for U.S. manufacturers have
Our paper reports three main findings. First, we decreased with time for each of raw material inven-
find that the explanatory variables explain a signif- tory, work-in-process inventory, and finished-goods
icant 66.7% of the within-firm variation and 97.2% inventory. They find that six sectors show increasing
of the total variation (i.e., within and across firms) trends in inventory turns for finished goods, and four
in inventory turnover. Annual inventory turnover is sectors show larger trends in the years 1980–1994,
found to be negatively correlated with gross margin when JIT became popular compared to the previous
and positively correlated with capital intensity and period (1961–1979). The results for raw material and
sales surprise. work-in-process inventories are marginally better.
Second, we estimate time trends in inventory This paper contributes to the operations literature
turnover in retailing, both with and without taking as it studies inventory productivity in retail services,
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
Management Science 51(2), pp. 181–194, © 2005 INFORMS 183
and utilizes firm-level panel data for all publicly segments except apparel and accessories and food
listed firms in this sector. Panel data are advanta- stores correspond to unique four-digit SIC codes. In
geous because they enable us to control for the effects apparel and accessories, we group together all firms
of unobserved firm-specific or time-specific factors that have SIC codes between 5600 and 5699 because
in measuring the trade-offs between the variables of there is substantial overlap among their products.
interest. We exploit the flexibility of panel data to This grouping enables us to increase the number of
compare alternative model specifications and assess degrees of freedom by estimating one set of coeffi-
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
their suitability for modeling inventory turnover cients for all apparel firms instead of estimating sep-
empirically. We also introduce the variable sales sur- arate coefficients for each SIC code. Likewise, in food
prise to control for the effect of unexpectedly high stores, we group together supermarket chains (SIC
sales on inventory turnover. Our model can be used code 5400) and convenience stores (SIC code 5411)
by managers to assess inventory turnover perfor- because of the overlap among their products.
mance, benchmark it against competing firms, and Let Ssit denote the sales, net of markdowns, of firm i
manage working capital requirements. The model can in segment s in year t, and CGSsit denote the cor-
also be applied in future research to assess the impact responding cost of goods sold. Both sales and cost
of improvements in operations on the inventory pro- of goods sold are obtained from the annual income
ductivity of firms. statements of the firms. Let GFAsitq denote the gross
The rest of this paper is organized as follows. Sec- fixed assets, comprised of land, property, and equip-
tion 2 summarizes the data and defines the perfor- ment, of firm i in segment s at the end of quarter q
mance variables used. Section 3 develops hypotheses in year t; NFAsitq denote the net fixed assets, com-
to relate inventory turnover with gross margin, cap- prised of gross fixed assets less accumulated depreci-
ital intensity, and sales surprise using results from ation; Invsitq denote the inventory, valued at cost; and
the existing literature. In §4, we discuss the empiri- TAsitq denote the total assets. These four data items are
cal model used in estimation. We present our empiri- obtained from the quarterly closing balance sheets of
cal results in §5, discuss their managerial implications the firms. From these data, we compute the following
in §6, and conclude in §7 with a discussion of the limi- performance variables for our study:
tations of our study and directions for future research. inventory turnover (also called inventory turns):
CGSsit
ITsit = 4
2. Data Description and Definition of 1
4 q=1 Invsitq
Variables gross margin:
We use financial data for all publicly listed U.S. retail-
Ssit − CGSsit
ers for the 16-year period 1985–2000, drawn from GMsit =
their annual income statements and quarterly and Ssit
annual balance sheets. These data are obtained from capital intensity:
Standard & Poor’s Compustat database using the 4
q=1 GFAsitq
Wharton Research Data Services (WRDS). CIsit = 4 4 and
The selection of firms is based on a four-digit q=1 Invsitq + q=1 GFAsitq
Standard Industry Classification (SIC) code assigned sales surprise:
to each firm by the U.S. Department of Commerce
Ssit
according to its primary industry segment. Our data SSsit =
set includes 10 segments in the retailing industry. sales forecastsit
Table 1 lists the segments, the corresponding SIC Here, average inventory and average gross fixed
codes, and examples of firms in each segment. All assets are computed using quarterly closing values to
Apparel and accessory stores 5600–5699 Ann Taylor, Filenes Basement, Gap, Limited
Catalog, mail-order houses 5961 Amazon.com, Lands End, QVC, Spiegel
Department stores 5311 Dillard’s, Federated, J. C. Penney, Macy’s, Sears
Drug and proprietary stores 5912 CVS, Eckerd, Rite Aid, Walgreen
Food stores 5400, 5411 Albertsons, Hannaford Brothers, Kroger, Safeway
Hobby, toy, and game shops 5945 Toys R Us
Home furniture and equip. stores 5700 Bed Bath & Beyond, Linens N’ Things
Jewelry stores 5944 Tiffany, Zale
Radio, TV, consumer electronics stores 5731 Best Buy, Circuit City, Radio Shack, CompUSA
Variety stores 5331 K-Mart, Target, Wal-Mart, Warehouse Club
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
184 Management Science 51(2), pp. 181–194, © 2005 INFORMS
control for systematic seasonal changes in these vari- firms that had missing data or accounting changes
ables during the year. The method for obtaining the other than at the beginning or the end of the mea-
sales forecast will be described in §3.3. surement period. These missing data are caused by
We also note that there are alternative measures bankruptcy filings and subsequent emergence from
of capital intensity that could be considered instead bankruptcy, leading to fresh-start accounting. There
of that defined above. In particular, NFAsitq could be were also accounting changes related to the inven-
used in place of GFAsitq to measure capital invest- tory valuation method. Out of 10 inventory valuation
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
ment, or TAsit could be used in the denominator as methods identified by the Compustat database, four
the scaling variable instead of the sum of the aver- are commonly used by retailers: FIFO (first in first
age inventory and the average gross fixed assets. We out), LIFO (last in first out), average cost method,
tested our hypotheses with these alternative mea- and retail method (see the Compustat data manual
sures and found that the results are consistent with for the definitions of these methods). Most retailers
those reported in this paper. The reader is referred to in our data set use a combination of these methods.
Stickney and Weil (1999) for detailed descriptions of After removing the firms with missing observations,
the income statement and balance-sheet variables. there were only three firms that switched between
Our original data set contains 5,088 observations exclusively FIFO and exclusively LIFO inventory val-
across 576 firms. After computing all the variables, uations during the subject time period.
the first two years of data for each firm are omitted. Our final data set contains 3,407 observations across
They could not be used in the analysis because the 311 firms, an average of 10.95 years of data per
computation of sales forecast required two years of firm. Table 2 presents summary statistics by retailing
sales data at the beginning of each time series. We segment for the performance variables used in our
also omit from our data set those firms that have less study. It lists the mean and standard deviation for
than five consecutive years of data available for any each variable within each segment. For example, the
subperiod during 1985–2000; there are too few obser- mean of inventory turnover for apparel and accessory
vations for these firms to conduct time-series analysis. stores is 4.57 and the standard deviation is 2.13. The
These missing data are caused by new firms enter- coefficient of variation of inventory turnover ranges
ing the industry during the period of the data set, from 0.36 for hobby, toy, and game shops to 1.92 for
and by existing firms getting delisted due to merg- home furniture and equipment stores. Thus, the vari-
ers, acquisitions, liquidations, etc. Further, we omit ability of inventory turnover is not limited to a few
Table 2 Summary Statistics of the Variables for Each Retail Segment: 1985–2000
Average Median
Number of annual Average Average Average annual Median Median Median
Number annual sales inventory gross capital sales inventory gross capital
Retail industry segment of firms observations ($ million) turnover margin intensity ($ million) turnover margin intensity
Apparel and accessory stores 72 786 9791 457 0.37 0.59 3019 4.22 0.35 0.62
213 0.08 0.14
Catalog, mail-order houses 45 441 4399 860 0.39 0.50 1420 5.38 0.40 0.51
911 0.17 0.18
Department stores 23 309 60586 387 0.34 0.63 13648 3.55 0.35 0.65
145 0.08 0.10
Drug and proprietary stores 23 256 23095 526 0.28 0.48 6676 4.38 0.29 0.51
290 0.07 0.12
Food stores 57 650 45736 1078 0.26 0.75 13001 9.79 0.26 0.77
458 0.06 0.08
Hobby, toy, and game shops 10 98 14555 299 0.35 0.46 2205 2.73 0.36 0.44
108 0.07 0.14
Home furniture and equip stores 13 125 3912 544 0.40 0.55 2240 2.90 0.41 0.54
1043 0.07 0.16
Jewelry stores 15 156 4752 168 0.42 0.36 2232 1.48 0.47 0.35
058 0.13 0.11
Radio, TV, consumer electronics stores 17 200 15850 410 0.31 0.44 4602 3.93 0.29 0.45
154 0.11 0.09
Variety stores 36 386 65487 445 0.29 0.51 7819 3.71 0.29 0.51
292 0.09 0.15
Aggregate statistics 311 3407 27914 608 0.33 0.57 5081 4.36 0.31 0.58
541 0.11 0.17
Note. The values for each variable are its mean and standard deviation across all observations in the respective segments.
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
Management Science 51(2), pp. 181–194, © 2005 INFORMS 185
firms, but is widely prevalent. Hereafter, we use IT, We explain below that the existing literature, even
GM, CI, and SS without the subscripts s, i, and t as though it is largely based on item-level models,
abbreviations for the respective variable names. We supports a negative correlation between inventory
collectively call GM, CI, and SS “explanatory vari- turnover and gross margin, and more importantly,
ables;” and IT, GM, CI, and SS “performance vari- identifies several factors that could explain this cor-
ables” or “firm-level performance variables.” relation. In particular, gross margin can be related
to inventory turnover directly because it determines
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
the impact of product variety on inventory turnover. retailer to reduce safety stock over the supplier lead
Lower variety through delayed differentiation is used time by postponing the decision to allocate inventory
to increase inventory turnover in the Benetton case across stores (“the joint ordering effect;” see Eppen
study (Heskett and Signorelli 1989), at Hewlett- and Schrage 1981). Second, the warehouse enables
Packard (Feitzinger and Lee 1997), and in research the retailer to centralize safety stock and rebalance
articles that explore these relationships (Lee and Tang store inventories between shipments from the sup-
1997, Swaminathan and Tayur 1998). Zipkin (2000, plier (“the depot effect;” see Jackson 1988).
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
Chapter 5) constructs an index of product variety, and We also expect inventory turns to increase with
observes from experience with a large firm that an investment in information technology. According to
increase in variety is associated with a decrease in Cachon and Fisher (2000), the benefits of imple-
inventory turnover. Van Ryzin and Mahajan (1999) menting information systems for the management of
also analyze the effects of variety on price and inven- inventory include better allocation of the inventory to
tory using a model of assortment choice. While they the stores, shorter ordering lead times, smaller batch
do not explicitly consider inventory turnover, they sizes, and a lower cost of processing orders. Clark
show that total inventory increases with variety, and and Hammond (1997), in a cross-sectional study, show
that variety under market equilibrium is increasing in that food retailers who adopt a continuous replen-
price given fixed procurement costs. ishment process (CRP) enabled by the adoption of
electronic data interchange (EDI) achieve 50%–100%
Length of the Product Life Cycle. The length of higher inventory turns than traditional ordering pro-
product life cycle has a similar effect on gross margin cesses. For further documentation of the benefits
and inventory turnover as product variety. A shorter of information technology, see Kurt Salmon Asso-
product life cycle implies rapid changes to products ciates (1993), Campbell Soup Company (Cachon and
to better match consumer requirements, and thus, Fisher 1997), Barilla SpA (Hammond 1994), H. E. Butt
increased consumer utility (Pashigian 1988). As dis- Grocery Co. (McFarlan 1997), and Wal-Mart Stores,
cussed above, higher consumer utility implies higher Inc. (Bradley et al. 1996).
prices and higher gross margin. A shorter product Because capital intensity is measured from gross
life cycle also implies less availability of historical fixed assets, it does not isolate the effects of differ-
data for forecasting. Because the accuracy of demand ent kinds of capital investments made by a firm. Fur-
forecasts increases with the availability of historical ther, it includes other capital investments of a retailer
data, products with longer life cycle and greater avail- as well—for example, investments in stores. These
ability of historical data should have lower demand could dilute the effect of capital intensity on inventory
uncertainty, less safety stock requirement, and higher turnover.
inventory turnover than products with shorter prod-
uct life cycle and less availability of historical data. 3.3. Sales Surprise
Because service level, price, variety, and life-cycle Inventory turnover can be affected by unexpectedly
length are not measured in our model, separate tests high sales. If the sales realized by a retailer in a given
of the linkages of these factors with inventory turns period are higher than its forecast, then the aver-
and gross margin are beyond the scope of our study. age inventory level for the period will be lower than
Instead, Hypothesis 1 is limited to estimating the cor- expected, and realized inventory turnover, which is a
relation of inventory turnover with gross margin. ratio of realized unit sales to the average inventory
for the period, will be higher than expected. The out-
3.2. Capital Intensity come will be reversed if realized sales are lower than
Investments in warehouses, information technology, expected sales.
and inventory and logistics management systems Therefore, we use sales surprise as defined in §2 to
involve capital investment by a firm, which is measure unexpectedly high sales, and formulate the
accounted as fixed assets, and is therefore measured following hypothesis.
by an increase in CI. Thus, we formulate the follow-
ing hypothesis: Hypothesis 3. Inventory turnover is positively corre-
lated with sales surprise.
Hypothesis 2. Higher capital intensity increases in-
ventory turnover. Sales surprise should, ideally, be measured with
respect to the management’s forecast of sales because
We expect that the addition of a new warehouse inventory decisions taken by the management are
should result in a decrease in total inventory at the based on these forecasts. Because managements’ sales
retailer, and thus an increase in its inventory turnover forecasts are not publicly reported, we estimate sales
for two reasons. First, the warehouse enables the forecasts from historical data using Holt’s linear
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
Management Science 51(2), pp. 181–194, © 2005 INFORMS 187
exponential smoothing method.3 The sales forecast for The following aspects of the model need elabora-
period t is tion:
sales forecastsit = Lsi t−1 + Tsi t−1 Firm-Specific Fixed Effects, Fi . Inventory turnover
can be correlated with factors that are omitted in
where Lsi t−1 and Tsi t−1 are smoothed series defined as our data set, such as managerial efficiency, marketing,
Lsit = Ssit + 1 − Lsi t−1 + Tsi t−1 location strategy, accounting policy, etc. These factors
can result in biased and inconsistent estimates of the
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
Tsit = Lsit − Lsi t−1 + 1 − Tsi t−1 parameters (see Hausman and Taylor 1981). There-
and 0 < < 1 and 0 < < 1 are weight- fore, we minimize their effect by using firm-specific
ing constants. We compared the forecast errors for control variables, Fi . These control variables can be
several values of and . The best forecasts were modeled either as fixed effects or as random effects.
obtained for = = 075. Thus, these values are We model them as fixed effects because they can be
used to compute all the results reported in this paper. used to compare average inventory turnover perfor-
We also computed sales forecasts using simple expo- mance across firms over the period of analysis.
nential smoothing and double exponential smooth- Omitted variables also imply that cross-sectional
ing. These forecasts had higher forecast errors and data for a single year or longitudinal data for a single
were biased compared to Holt’s linear exponential firm are unsuitable for estimating the model because
smoothing. they cannot distinguish the effects of the explanatory
variables from differences in Fi (see Hoch 1962).
4. Model Specification and Analysis Time-Specific Fixed Effects, ct . These variables
We propose the following log-linear model to test the control for changes in secular characteristics over
hypotheses: time, such as in economic conditions, in the inter-
est rates, in price level, etc., and thus enable us to
log ITsit = Fi + ct + bs1 log GMsit + bs2 log CIsit
compare inventory turnover across years. They also
+ bs3 log SSsit + sit (1) enable us to measure trends in average inventory
turnover in the retailing industry over time after
Here, Fi is the time-invariant firm-specific fixed effect
controlling for the effects of the other explanatory
for firm i; ct is the year-specific fixed effect for year t;
variables.
bs1 , bs2 , and bs3 are the coefficients of log GMsit , log CIsit ,
and log SSsit , respectively, for segment s; and sit Segment-Specific Coefficient Estimates, bs1 , bs2 , bs3 .
denotes the error term for the observation for year t The coefficients of the explanatory variables might
for firm i in segment s. Hypotheses 1, 2, and 3 imply differ across retailing segments. Thus, we estimate
that, for each segment s, bs1 must be less than zero, segment-specific coefficients to test for heterogeneity
and bs2 and bs3 must be greater than zero. across segments.
We use a log-linear specification for three rea- It is useful to estimate other model specifications
sons: (1) A log-linear relationship between the vari- with different combinations of the control variables
ables is suggested by plotting IT against GM, CI, to ascertain the correctness of the model and draw
and SS. (2) Retail industry reports (see, for exam- further insights. First, to test whether the coefficients
ple, Sack 2000) and surveys of retailers that we have of the explanatory variables differ across segments,
conducted show that multiplicative measures such we compare (1) with the following specification (with
as GMROI and return on assets are widely used to pooled coefficients of explanatory variables instead of
measure and reward the performance of inventory segmentwise coefficients):
planners and merchants. (3) We simulated a periodic-
review inventory model with stationary demand for log ITsit = Fi + ct + b 1 log GMsit + b 2 log CIsit
different values of gross margin, lead time and vari- + b 3 log SSsit + sit (2)
ance of demand, and collected data on the variables of
interest to compare log-linear and linear model spec- Second, to test whether the firmwise fixed effects Fi
ifications. We found that the log-linear specification are statistically significant, we compare (1) with
had significantly lower prediction errors than the lin- the following specification (with segmentwise fixed
ear specification. effects instead of firmwise fixed effects):
3
Alternatively, one could use the I/B/E/S data set to obtain sales log ITsit = Fs + ct + bs1 log GMsit + bs2 log CIsit
forecasts. This data set provides analysts’ forecasts of sales for a
subset of the publicly listed firms for the period 1997 onwards. + bs3 log SSsit + sit (3)
For the time period used in this paper, I/B/E/S provides forecasts
only for a few firms, giving a total of less than 300 observations. Here, Fs is the segmentwise fixed effect, and the other
However, this data set could be useful in future research. terms have their usual meaning. Both (2) and (3)
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
188 Management Science 51(2), pp. 181–194, © 2005 INFORMS
are useful because they have fewer parameters, and Table 3 Fit Statistics for the Maximum Likelihood Estimates of
hence can allow more precise estimation. However, Models (1) and (2)
(3) should not be used if firmwise fixed effects are Model (1) Model (2)
statistically significant.
Third, because IT and GM are both functions of cost −2 · log-likelihood ratio −42831 −38944
(chi-sq = 2,332.51) (chi-sq = 2,221.97)
of goods sold, it is possible that we observe a nega- AIC −35371 −32024
tive correlation between IT and GM even if inventory AICC −34209 −31035
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
levels are independent of the gross margin realized BIC −21434 −19095
by a firm. A similar argument could be applied to Tests of significance of variables (F -tests)
IT and CI because they are both functions of aver- Firm 1224 2013
age annual inventory. To test for this problem, we Year 609 418
estimate an alternative model log GM 11928 29099
using average annual log CI 12914 13905
inventory level, Invsit = 41 q Invsitq , as the depen- log SS 44812 40316
dent variable instead of ITsit . We add CGSsit to the list
Differences in coefficient estimates across segments (F -tests)
of explanatory variables in this model to control for log GM 419
scale: log CI 3026
log SS 1220
log Invsit = Fs + ct + bs1 log GMsit + bs2 log CIsit
Note. All the statistics are significant with p < 00001.
+ bs3 log SSsit + b 4 log CGSsit + sit (4)
whether each coefficient differs across segments are
For this model, we measure capital intensity as the also significant (p < 00001 for each of log GMsit ,
ratio of GFA to TA, rather than as defined in §2, to log CIsit , and log SSsit . Therefore, the coefficients’ esti-
avoid having a term in the explanatory variables that mates differ significantly across segments.
is a function of Invsit . Further, we also test for this We determine the fraction of variation in log ITsit
problem using Models (1) and (2). We estimate these explained by each model by computing the over-
models with values of gross margin and capital inten- all prediction accuracy and the within-firm predic-
sity lagged by one year, log GMsi t−1 and log CIsi t−1 , tion accuracy of each model using the usual formula
as the explanatory variables instead of log GMsit and for R2 :
log CIsit .
Other model specifications can be constructed to overall prediction accuracy
test if bs1 , bs2 , and bs3 change with time or if the time-
s i t log ITsit −
log ITsit 2
specific fixed effect ct differs across segments. We =1−
s i t log ITsit − log IT
2
compare the results of different specifications in §5.
The main results of the paper are based on (1) and (2). within-firm prediction accuracy
We estimate all models assuming that the error term,
sit , has first-order autocorrelation, and is segment- s i t log ITsit − log ITsit 2
=1−
s i t log ITsit − log ITsi
2
wise heteroscedastic; i.e., the variance of sit varies by
segment. The estimation process and statistical tests
of assumptions are presented in the appendix. The where log ITsit is the predicted value of log ITsit
reader is referred to Greene (1997, Chapter 14), Hsiao obtained from (1) or (2), log IT is the overall mean
(1986), and Judge et al. (1985, Chapter 13) for further of log ITsit , and log ITsi is the within-firm mean
discussion of the specification and estimation of panel of log ITsit .4 The overall prediction accuracy for
data models such as ours. Model (1) is 97.16% and for Model (2) is 96.83%.
The within-firm prediction accuracy for Model (1) is
66.7% and for Model (2) is 62.8%. The within-firm
5. Results accuracy is remarkable because it shows that year-to-
5.1. Basic Results year changes in the IT of a firm are highly correlated
Table 3 shows the fit statistics for Models (1) and (2), with simultaneous changes in GM, CI, and SS. The
estimated using MLE. The overall fit of Model (1) is overall prediction accuracy is higher than the within-
statistically significant p < 00001 . The coefficients firm accuracy because the between-firm variation in
of all the explanatory variables, log GMsit , log CIsit ,
4
and log SSsit , are also significantly different from zero There are several measures of R-square for the generalized regres-
sion model. One alternative would be to apply the formula for
p < 00001 . Comparing the results for Models (1)
R-square to the transformed model obtained in FGLS estimation.
and (2), we find that the likelihood ratio test that Because the R-square thus determined need not lie between 0
Model (1) is preferred to Model (2) is statistically sig- and 1, we do not use this procedure. See Kmenta (1996, Chapter 12)
nificant p < 00001 . Separate F -tests to determine for details.
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
Management Science 51(2), pp. 181–194, © 2005 INFORMS 189
Table 4 Coefficients’ Estimates for Models (1) and (2) Obtained from MLE
Note. Pooled coefficients are for Model (2) and segmentwise coefficients are for Model (1).
a b
Statistically significant at p < 0001 and p < 002, respectively, for two-tailed tests.
inventory turnover is larger than the within-firm Hypothesis 3. The value of the pooled coefficient is
variation, and is fully explained by the firm-specific 0.143, and the segmentwise coefficients range between
fixed effects. 0.053 and 0.279. These estimates are useful because
Table 4 shows the coefficients’ estimates for Mod- they enable us to control for the effect of surprisingly
els (1) and (2). The pooled coefficient for log GMsit is high sales on inventory turnover.
−0285 p < 00001 , and strongly supports Hypoth-
esis 1. The segmentwise coefficients also support 5.2. Time Trends in Inventory Productivity
Hypothesis 1 for nine of the ten segments. Thus, The year-specific fixed effects, ct , in our model can
inventory turns are negatively correlated with gross be used to estimate the time trend in inventory pro-
margin. Because we have a log-linear model, the ductivity after adjusting for the correlation with gross
coefficient of log GMsit gives the elasticity of inven- margin, capital intensity, and sales surprise. Table 5
tory turns with respect to gross margin. Thus, a shows the estimates of ct obtained from Models (1)
1% increase in gross margin (for example, from 0.5 and (2), and Figure 2 shows a time-series plot of ct
to 0.505) is associated with an estimated −0285% for Model (2). Note that these estimates are decreas-
change in inventory turns. ing with time. Taking the standard errors of the
Across segments, the coefficient estimate for estimates into account, we find that the estimates
log GMsit varies from −0153 for apparel and acces- of ct for the initial years, t = 1987 1993, are
sories retailers to −0571 for hobby, toy, and game
shops. From the discussion in §3, there can be sev-
eral reasons for this variation. For example, for the Table 5 Estimates of Time-Specific Fixed Effects ct for Models (1)
and (2)
same value of GM, product variety, life-cycle length
and demand uncertainty may vary across segments, Model (1) Model (2)
resulting in different coefficients. Because price, vari-
Year Estimate Std. error Estimate Std. error
ety, and life-cycle length are not included in our data
set, we cannot ascertain the causes of the differences 1987 01297 a
0.0171 01009a
0.0181
in coefficients’ estimates across segments. These could 1988 00890a 0.0168 00611a 0.0177
1989 00774a 0.0164 00423a 0.0173
be investigated in a further study. 1990 00697a 0.0160 00375a 0.0169
The pooled coefficient for log CIsit is 0.252 p < 1991 00681a 0.0155 00460a 0.0164
00001 , and strongly supports Hypothesis 2. Seg- 1992 00586a 0.0150 00388a 0.0159
mentwise estimates of the coefficient of log CIsit also 1993 00517a 0.0146 00363a 0.0155
strongly support Hypothesis 2 for seven of the ten seg- 1994 00380a 0.0141 00276b 0.0150
1995 00287b 0.0136 00174 0.0145
ments. The value of the coefficient ranges from 0.106
1996 00109 0.0129 00008 0.0139
to 1.085 across segments where it is statistically signif- 1997 00119 0.0121 −00009 0.0129
icant p < 002 . Thus, investment in capital assets is 1998 00044 0.0108 −00024 0.0116
positively correlated with inventory turnover. 1999 00000 0.0086 −00024 0.0093
The pooled and segmentwise estimates of the coef- 2000 — —
ficient of log SSsit are all positive and statistically a b
Statistically significant at p < 001 and p < 010, respectively, for two-
significant at p < 0001. They strongly support tailed tests.
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
190 Management Science 51(2), pp. 181–194, © 2005 INFORMS
Figure 2 Plot of Time-Specific Fixed Effects ct for Model (2) Table 6 Time Trends in IT, CI, and GM Estimated Using Equation (5)
0.16 Variable Coefficient Std. error t-statistic p-value
0.14
IT −005460 0.01354 −403 <00001
0.12 log IT −000454 0.00110 −411 <00001
0.1 CI 000568 0.00030 1900 <00001
log CI 001250 0.00077 1623 <00001
0.08
GM −000018 0.00031 −059 05568
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
c (t )
have similar inventory turnover and gross margin ple, in year 1996, the inventory turns of these firms
values, but different capital intensities, then the firm are 5.9 (Best Buy), 4.1 (Circuit City), 8.2 (CompUSA),
with the lower capital intensity could possibly have a and 3.0 (Radio Shack). The corresponding values of
better capability to manage inventory than the other gross margin are 0.15, 0.26, 0.14, and 0.32, respectively,
firm. Finally, if a firm realizes an increase in inventory and the corresponding values of capital intensity are
turnover with an unexpected increase in sales, then 31%, 47%, 33%, and 42%, respectively. After apply-
the increased inventory turnover may not indicate ing these values and sales surprise, we find that the
an improved capability to manage inventory. Thus, adjusted inventory turns of the four firms are 3.1,
changes in gross margin, capital intensity, and sales 2.6, 4.1, and 2.2, respectively. Thus, we observe that
surprise should be incorporated in the evaluation of the differences between the adjusted inventory turns
inventory productivity of a firm. turns of Best Buy, Circuit City, CompUSA, and Radio
Our results give a trade-off curve that computes Shack are smaller than the differences between their
the expected inventory turnover of a firm for given inventory turns because the differences in inventory
values of sales surprise, gross margin, and capital turns are partly compensated for by the differences
intensity. We term the distance of the firm from its in gross margin. For example, the inventory turns of
trade-off curve as its adjusted inventory turnover (AIT). CompUSA are 2.8 times those of Radio Shack, but the
The value of AIT for firm i in segment s in year t is adjusted inventory turns of CompUSA are 1.9 times
computed as those of Radio Shack.
The gross margins of these firms might differ due
log AITsit = log ITsit − b 1 log GMsit to any one or more of the factors listed in §3. Indeed,
the annual reports to shareholders of these firms
− b 2 log CIsit − b 3 log SSsit show that their product mixes differ from each other.
CompUSA has a higher proportion of personal com-
or, equivalently, as puters in its sales (high turns, low margins), Best Buy
and Circuit City have higher proportions of home
−b 1 −b 2 −b 3
AITsit = ITsit GMsit CIsit SSsit (8) electronics and appliances (midrange turns and mar-
gins), and Radio Shack has a higher proportion of
AIT is an empirical measure to compare inventory spare parts and accessories in its sales (low turns,
productivity across firms and across years by control- high margins). Because the adjusted inventory turns
ling for differences in gross margin, capital intensity, of these firms are less disparate than their inven-
and sales surprise. tory turns, it suggests that the differences in inven-
According to our results, managers of firms with tory turns between these firms may not indicate better
low AIT should investigate whether their firms are or worse inventory productivity. Figure 3 shows the
less efficient than their peers, and identify steps they trade-off between inventory turns and gross margin
might take to improve their inventory productivity. for the four firms obtained using (8) (assuming con-
Firms with higher inventory turnover may differ sys- stant capital intensity and no sales surprise to avoid
tematically from firms with lower inventory turnover. year-to-year variations in these variables).
On the one hand, such differences may be attributed Example 2. Ruddick Corporation is a holding com-
to differences in accounting policies and may not have pany that owns Harris Teeter, a regional supermar-
operational implications. On the other hand, they may ket chain in the southeastern United States with
indicate differences in efficiency that cannot be recti- 137 stores and sales of $2.7 billion in the year 2000.
fied simply by increased spending. This possibility is Figure 4 shows time-series plots and linear trends of
supported in our discussions with managers. Fisher IT and AIT for Harris Teeter for the years 1987–2000.
et al. (2000) discuss differences between retailers and The inventory turns of Harris Teeter do not show any
identify best practices in retail operations through on- significant time trend. During the same period, the
site interviews, surveys, and workshops in a four-year gross margin of Harris Teeter has increased steadily
study of 32 retailers. from 23.4% to 30.9%, and its capital intensity has
We present two examples to illustrate the interpre- increased marginally. After applying (8), we find that
tation of adjusted inventory turns. The first example the adjusted inventory turns of Harris Teeter have
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
192 Management Science 51(2), pp. 181–194, © 2005 INFORMS
Figure 3 Trade-off Curves Between Inventory Turnover and Gross 7. Limitations and Directions for
Margin Estimated for the Four Consumer Electronics
Retailers in Figure 1 for the Years 1987–2000 Future Research
We have shown that inventory turns in retail ser-
14 Best Buy Co. Inc. vices have a high correlation with gross margin, cap-
Circuit City Stores ital intensity, and sales surprise. Therefore, inventory
12 CompUSA
turns should not be used, per se, in performance
Radio Shack
analysis. Instead, we have proposed an empirical
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
in our study may change even when accounting poli- use maximum likelihood estimation (MLE) to estimate the
cies remain unchanged. Therefore, our results need to parameters of our model.5 See Greene (1997, Chapter 13)
be interpreted with caution. for a description of the estimation methodology, and Judge
Our paper identifies opportunities for future et al. (1985) for a survey of the research on the asymptotic
research on inventory productivity. One valuable area properties of various estimation methods.
We obtain the following results regarding the spec-
of research is to investigate why some firms realize
ification of the error term. We find that segmentwise
higher inventory productivity than others even after heteroscedasticity and first-order autocorrelation are also
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
controlling for differences in capital investment, gross statistically significant in our data set.6 The standard error
margin, and sales surprise. This question could be ranges from 0.011 for department stores to 0.145 for home
of relevance to both the operations and the business furnishings and equipment stores. The autocorrelation coef-
strategy literature. In the latter, it would contribute ficient ranges from 0.29 for hobby toys and games stores to
to research on the role of industry-level, corporate- 0.92 for home furnishings and equipment stores. Thus, the
level, and firm-level strategy on the operating per- use of MLE with segmentwise heteroscedastic and AR(1)
formance of firms (see, for example, Beard and Dess autocorrelated errors is suitable for our analysis.
1981, McGahan and Porter 1997, Rumelt 1991, and Further, we find that the coefficients’ estimates for the
Schmalensee 1985). A related research question is to explanatory variables are robust with respect to changes
in model specification if the assumptions of heteroscedas-
understand why the elasticities of inventory turns
tic and AR(1) autocorrelated errors are applied. For exam-
with respect to the explanatory variables differ across ple, even though firmwise fixed effects Fi are statistically
segments. We have hypothesized some plausible rea- significant, Model 3 with segmentwise fixed effects gives
sons for these differences that could be tested in similar coefficients’ estimates as Models (1) and (2). (The
future research. Finally, we have tested our model autocorrelation coefficient becomes very large and captures
for many alternative specifications, data sets, and some of the differences across firms.) However, Model 3
time periods. We have found the estimates to be does not give the same results when errors are assumed
very robust with respect to these variations. This to be homoscedastic and independent. We think that this
model could be applied to control for correlations indicates the presence of a correlation between the firm-
between performance variables in future research as wise fixed effects and the explanatory variables as discussed
in §4. Other changes, such as segment-specific year effects
to how operational improvements affect operating
or linear time trends also do not have any significant effect
performance and financial performance of firms.
on the coefficients of the explanatory variables.
Acknowledgments
The authors thank the referees and the editors for their valu- References
able feedback on the previous versions of this work.
Balakrishnan, R., T. J. Linsmeier, M. Venkatachalam. 1996. Financial
benefits from JIT adoption: Effects of consumer concentration
Appendix and cost structure. Accounting Rev. 71 183–205.
Beard, D. W., G. G. Dess. 1981. Corporate-level strategy, business-
Estimation and Econometric Issues level strategy, and firm performance. Acad. Management J. 24
Because our data contain observations across firms and 663–688.
years, it is likely that the variance of sit varies across Billesbach, T. J., R. Hayen. 1994. Long-term impact of JIT on inven-
firms, and that sit is correlated across years. Further, we tory performance measures. Production Inventory Management J.
find that the standard deviations of all the variables dif- 35(1) 62–67.
fer substantially across segments. For example, as shown Bradley, S. P., P. Ghemawat, S. Foley. 1996. Wal-Mart Stores, Inc.
in Table 2, the standard deviation of inventory turnover Harvard Business School Case 9-794-024, Harvard University,
ranges from a low of 0.58 for jewelry stores to a high of Cambridge, MA.
10.42 for home furniture and equipment stores. Likewise,
the standard deviation of gross margin ranges from a low 5
Other estimation techniques that may be used include feasi-
of 0.12 for food stores to a high of 1.02 for catalog and mail- ble generalized least squares (FGLS) estimation omitting the first
order houses. Therefore, we consider a flexible structure observation for each firm (the Cochrane-Orcutt procedure or the
of the variance-covariance matrix of sit with segmentwise Hildreth-Liu procedure) or full FGLS estimation using the Prais-
heteroscedasticity and first-order autocorrelation. Segmen- Winsten transform. We do not omit the first observation in the
time series of each firm because our data have relatively short time
twise heteroscedasticity implies that the variance of sit is
series, and it has been shown that discarding the first observation
identical across firms within a retailing segment, but differs
can adversely affect the efficiency of the estimates when the length
across segments. Autocorrelation is also a common charac- of the time
teristic of financial time-series data. Because our data are series is short. Instead, we apply the transformation
si1 = usi1 / 1 − rs2 to the first observation.
annual time series, we use a first-order autoregressive pro- 6
Upon estimating Models 1–3 with and without the assumptions of
cess; higher-order autoregressive processes would be suit- segmentwise heteroscedasticity and AR(1) autocorrelation, we find
able for monthly or quarterly data. that the value of the log-likelihood function improves significantly
For this variance structure, ordinary least squares (OLS) (likelihood ratio test p < 00001) after these assumptions. White’s
estimators are not efficient and the tests of significance general test for heteroscedasticity and the Durbin-Watson test for
performed on OLS estimators are not valid. Therefore, we AR(1) autocorrelation are also statistically significant.
Gaur et al.: Econometric Analysis of Inventory Turnover Performance in Retail Services
194 Management Science 51(2), pp. 181–194, © 2005 INFORMS
Cachon, G. P., M. L. Fisher. 1997. Campbell Soup’s continuous Jackson, P. 1988. Stock allocation in a two-echelon distribution sys-
replenishment program: Evaluation and enhanced decision tem or “What to do until your ship comes in.” Management Sci.
rules. Production Oper. Management 6 266–276. 34 880–895.
Cachon, G. P., M. L. Fisher. 2000. Supply chain inventory manage- Judge, G., C. Hill, W. Griffiths, T. Lee. 1985. The Theory and Practice
ment and the value of shared information. Management Sci. 46 of Econometrics. John Wiley and Sons, New York.
1032–1048. Kekre, S., K. Srinivasan. 1990. Broader product line: A necessity to
Chamberlin, E. 1950. Product heterogeneity and public policy. Amer. achieve success. Management Sci. 36 1216–1231.
Econom. Rev. 40 85–92. Kmenta, J. 1997. Elements of Econometrics, 2nd ed. University of
Downloaded from informs.org by [129.100.255.37] on 19 January 2024, at 09:23 . For personal use only, all rights reserved.
Chang, D., S. M. Lee. 1995. Impact of JIT on organizational Michigan Press, Ann Arbor, MI.
performance of U.S. firms. Internat. J. Production Res. 33 Kotler, P. 1986. Principles of Marketing. Prentice-Hall, Englewood
3053–3068. Cliffs, NJ.
Clark, T., J. Hammond. 1997. Reengineering channel reordering Kurt Salmon Associates Inc. 1993. Efficient Consumer Response:
processes to improve total supply chain performance. Produc- Enhancing Consumer Value in the Grocery Industry. Food
tion Oper. Management 6 248–265. Marketing Institute, Washington, D.C.
Dixit, A., J. Stiglitz. 1977. Monopolistic competition and optimum Lancaster, K. 1990. The economics of product variety: A survey.
product variety. Amer. Econom. Rev. 67 297–308. Marketing Sci. 9 189–206.
Eppen, G., L. Schrage. 1981. Centralized ordering policies in Lazear, E. P. 1986. Retail pricing and clearance sales. Amer. Econom.
a multi-warehouse system with lead times and random Rev. 76 14–32.
demand. L. Schwarz, ed. Multi-Level Production/Inventory Con- Lee, H., C. Tang. 1997. Modelling the costs and benefits of delayed
trol Systems: Theory and Practice, TIMS Studies in the Management production differentiation. Management Sci. 43 40–53.
Sciences, Vol. 16. North-Holland, Amsterdam, The Netherlands. Levy, M., B. Weitz. 2001. Retailing Management, 4th ed. McGraw-
Feitzinger, E., H. L. Lee. 1997. Mass-customization at Hewlett- Hill/Irwin Publishers, New York.
Packard: The power of postponement. Harvard Bus. Rev. 75 McFarlan, F. W. 1997. H. E. Butt Grocery Co.: A Leader in ECR
116–121. Implementation (A). Harvard Business School case 9-196-061,
Fisher, M. L., A. Raman, A. McClelland. 2000. Rocket science retail- Harvard University, Cambridge, MA.
ing is almost here—Are you ready? Harvard Bus. Rev. 78(4) McGahan, A. M., M. E. Porter. 1997. How much does industry
115–124. matter, really? Strategic Management J. 18 15–30.
Greene, W. H. 1997. Econometric Analysis, 3rd ed. Prentice-Hall, Inc., Nagle, T. T. 1987. Strategy and Tactics of Pricing. Prentice-Hall,
Englewood Cliffs, NJ. Englewood Cliffs, NJ.
Hammond, Janice. 1994. Barilla SpA. Harvard Business School case Pashigian, B. P. 1988. Demand uncertainty and sales: A study of
9-694-046, Harvard University, Cambridge, MA. fashion and markdown pricing. Amer. Econom. Rev. 78 936–953.
Hausman, J., W. Taylor. 1981. Panel data and unobservable individ- Rajagopalan, S., A. Malhotra. 2001. Have U.S. manufacturing inven-
tories really decreased? An empirical study. Manufacturing
ual effects. Econometrica 49 1377–1398.
Service Oper. Management 3 14–24.
Hendricks, K. B., V. R. Singhal. 1997. Does implementing an effec-
Rumelt, R. P. 1991. How much does industry matter? Strategic
tive TQM program actually improve operating performance:
Management J. 12 167–185.
Empirical evidence from firms that have won quality awards.
Management Sci. 43 1258–1274. Sack, K. 2000. Retailing: General Industry Survey. Standard & Poor,
New York.
Hendricks, K. B., V. R. Singhal. 2001. The long-run stock price per-
formance of firms with effective TQM programs as proxied by Schmalensee, R. 1985. Do markets differ much? Amer. Econom. Rev.
75 341–351.
quality award winners. Management Sci. 47 359–368.
Stickney, C. P., R. L. Weil. 1999. Financial Accounting: An Introduction
Heskett, J. L., S. Signorelli. 1989. Benetton (A). Harvard Business
to Concepts, Methods, and Uses, 9th ed. Dryden Press, Orlando,
School case 9-685-014, Harvard University, Cambridge, MA.
FL.
Hoch, I. 1962. Estimation of production frontier parameters combin-
Swaminathan, J., S. Tayur. 1998. Managing broader product lines
ing time-series and cross-section data. Econometrica 30 34–53. through delayed differentiation using vanilla boxes. Manage-
Hopp, W., M. Spearman. 1996. Factory Physics. McGraw-Hill/Irwin ment Sci. 44 S161–S172.
Publishers, New York. van Ryzin, G., S. Mahajan. 1999. On the relationship between inven-
Hsiao, C. 1986. Analysis of Panel Data. Cambridge University Press, tory costs and variety benefits in retail assortments. Manage-
New York. ment Sci. 45 1496–1509.
Huson, M., D. Nanda. 1995. The impact of just-in-time manufactur- Zipkin, P. H. 2000. Foundations of Inventory Management. McGraw-
ing on firm performance. J. Oper. Management 12 297–310. Hill/Irwin Publishers, New York.