Time-Phased Safety Stocks Planning and its Financial Impacts:

Empirical Evidence based on European Econometric Data

Martin Stlein
Dept of Logistics and Supply Chain Management, University of Vienna
A-1210 Vienna, Austria
martin.stoesslein@univie.ac.at

Abstract
This paper explores the rationale for planning time-phased safety stocks. We assert that a single safety stock
vector for the entire planning horizon (typically based on stationary demand forecast errors and stationary
replenishment lead times) may be insufficient for hedging against uncertainties. We argue that planning timephased safety stocks is prudent when faced with non-stationary demand and/or non-stationary supply. We
scrutinize particularly whenever non-stationarity is due to heteroscedastic demand and resulting heteroscedastic
demand forecast errors. Consequently, empirical evidence on a wide basis is provided that such errors for
manufactured products are highly heteroscedastic. To test the phenomenon and to estimate its impact at stock
keeping unit level, we have conducted an econometric analysis using EUROSTAT data from 1985 onwards.
Specifically, we analyze new industrial orders across various industries and types of goods manufactured in the
five largest European economies by using EViews 7.0. To demonstrate which inventory savings can accrue when
safety stock levels are deliberately planned to vary in accordance with the observed heteroscedasticity, we
estimate potential safety stock savings reusing the same data sets. Our findings indicate that one realization of
non-stationarity, i.e., heteroscedastic demand, is indeed pervasive in European industry. Thus, recognition of this
demand nature may add to effective inventory management policies: reducing unnecessary safety stocks,
improving service, or both relative to a single-valued safety stock regimen.
Keywords: Time-phased safety stocks planning; econometric analysis; heteroscedasticity tests; empirical study.

1.
Introduction
European companies invest and hold inventory worth billions of Euros (e.g., in Germany on
average 14% of revenue; Harting, 2005), with around one fifth of inventory assets accounting
for safety stocks. Yet, despite this significance, it can be observed in inventory practice that
safety stock targets are typically set at the beginning of an operating year to a single
appropriate level to be applied for future time periods (after estimating risks).
We argue that a time-invariant safety stock level may not be sufficient to respond to all the
variabilities a company is exposed to nowadays. We propose that time-phased targets for each
planning period could enhance inventory performance, when deliberately considering nonstationary demand and supply processes. Indeed, various sources of systematically increasing
or decreasing variabilities can be identified. On the supply side, capacities change, e.g.,
factories schedule holidays during the summer, resulting to variabilities in lead times. On the
demand side, distribution of demand may change during a products life cycle - in the initial
stage demand is often more variable than later. In practice, however, safety stock levels are
commonly calculated on the assumption of stationary processes. It is observed that (time
invariant) safety stocks are significantly higher during the operating year than the singlevalued target level agreed on at the beginning; additionally, there is usually no system to alert
inventory managers to this situation (Mahadevan, 2009, p. 363). Certainly, when replenishment lead time and demand are stationary, constant safety stocks may then suffice.

In this article, we focus on the case of non-stationary demand. We explicitly aim to provide
evidence that demand forecast errors systematically vary over time (i.e., heteroscedastic),
adding to the motivation to implement time-phased safety stock policy. Three specific
questions are studied here:
(1) To what degree is non-stationarity in demand forecast errors due to heteroscedastic
demand variance in manufacturing industries in the chief European economies?
(2) What are the right doses of safety stocks reflecting heteroscedastic demand variance?
(3) Which European industries could have the largest cost-saving potentials switching to a
time-phased safety stock policy?
Our empirical study could bring us nearer to our vision of using econometrical findings for
inventory planning and control purposes. More specifically, policies for different inventory
types may be refined if econometric knowledge about the non-stationary nature of supply and
demand processes can be incorporated.
This article builds upon prior results produced by the research team (Kanet, Gorman, and
Stlein, 2010). Compared to previous contributions, we expanded the analysis to include
most recent sample data comprising several European countries, to forecast models for each
set of time series data, and to a rigorous approach to testing heteroscedasticity.
The rest of the article is organized as follows. Section 2 shortly outlines relevant literature.
Section 3 describes the selected sample data and adjustments. Next, we present our empirical
analysis and test results in section 4. Managerial implications are illustrated with potential
safety stock savings by both using case data and industry data (section 5). Finally, we
conclude with a brief discussion and promising directions for future research.

2.
Literature Review
Safety stock research has been an area of intensive inquiry since Graves (1988) identified a
huge research gap and been revisited since then, with considerable efforts invested into
inventory modeling (e.g., Silver and Pyke, 1998; Grubbstrm and Tang, 1999; Minner, 2000;
Zipkin, 2000). However, as we see it, the body of literature for treating demand and lead
times as random variables is small, although there are notable exceptions (Graves, 1999;
Graves and Willems, 2008; Neale and Willems, 2009).
As the area of supply chain management has evolved, researchers have incorporated central
multistage design issues (e.g., Lee and Billington, 1993). Nevertheless, a large stream of
research focuses on varying safety stocks by geographic position rather than by time - as
proposed by us.
Recently, inventory performance has been more and more empirically analyzed - with the
majority of research linked to financial indicators (e.g., Hendricks and Singhal, 2009).
Notably, econometric studies about inventory in Germany are rare, with exceptions such as
Obermaier and Donhauser (2009). Even rarer is empirical research into the inventory type of
safety stocks, in particular, especially dealing with non-stationary processes.
(An elaborated literature review is to be found in Kanet et al., 2010, and Stlein et al., 2012.)

3.
Sample Data Description
To explore the aforementioned research questions employing econometric analysis, this paper
uses publicly available data.
3.1
Data Selection
To estimate demand for products and services, shipment data drawn from national statistical
bureaus could have been a suitable indicator. However, whereas these data are published by
the U.S. Census Bureau, they are not available on the European level. Instead, we have
chosen monthly data from Industry New Orders to give us a close idea of demand. The
selected data are considered a leading indicator by investors for evaluating the overall future
economic condition and industrial production of a region, industry or company. The order
comprises the contract value between a manufacturer and customer which exists after there is
sufficient evidence for a valid agreement from the perspective of a manufacturer.
We obtained Industry New Orders time series from EUROSTAT, the statistical office of the
European Union situated in Luxemburg. The selection of our sample data is based on the
NACE Rev.2 classification (part of the short-term business statistics), which includes those
manufacturing industries that most usually work on orders. To achieve a high degree of data
precision, we have omitted data that describe less specific industries. This included three time
series, namely manufacture of other non-metallic mineral products, manufacture of other
transport equipment, and other manufacturing. Due to our focus on manufacturing in this
paper, we are disregarding construction, building and civil engineering (NACE F), and
trade and services (t_sts_ts).
The EUROSTAT time series seems advantageous because they enable us to evaluate data on
different aggregation levels. Time series represent five types of goods manufactured (i.e.,
intermediate goods, capital goods, consumer goods, durable consumer goods, non-durable
consumer goods), industries (e.g., manufacture of electrical equipment; two-digit NACE_R2
classification), and disaggregated sub-industries (e.g., manufacture of electric motors,
generators and transformers; four-digit NACE_R2 classification). Moreover, the data
distinguish between domestic and non-domestic markets; the origin is determined by the
residency of the third party that has ordered or purchased the product. In addition, nondomestic new orders are further sub-divided into orders received from euro-zone countries
and all other non-domestic new orders.
The body of available data allows manifold analysis on a number of national levels. Time
series are provided on a European Union level (up to 27 countries), Euro area level (up to 17
countries), and national level (e.g., France). As the length of this paper is limited, we focus
our report on aggregated level to the five largest economies in the European Union. We
selected the sample data according to the national gross domestic products list, calculated by
the International Monetary Fund (2011): Germany, UK, France, Italy, and Spain. A separate
analysis responds to the tremendous growth of new orders in other European countries, each
having more than 100% growth since 2005, among them Latvia, Estonia, Turkey, Croatia and
Poland. An elaborated study of all European countries can be found at Stlein and Minner
(2012).
Our final data set for this article contains monthly time series data across 22 aggregated
industries (using the two-digit NACE_R2 classification), five types of goods manufactured,
and up to 26 years of observations for each of the five sample countries.

3.2
Data Adjustments
To improve forecast accuracy, we considered refining raw data for the following two key
effects.
3.2.1 Seasonality
Adjusting for seasonality is often a procedure used to break down a time series into its
underlying effects. However, in order to capture full seasonality effects, we have based our
forecasts on monthly raw data that is not seasonally adjusted. Indeed, seasonal adjustments
would dampen the uncertainty and variability of demand and reduce the implied incentive for
safety stocks. It is the seasonal variability inherent in raw data that boosts firms safety stocks.
3.2.2 Inflation
In order to remove the effects of changes in price from the Industry New Orders, we
adjusted the data for inflation. We therefore used the gross domestic product (GDP) deflator
which is derived by dividing the GDP at current prices by the GDP at constant prices. The
result is the volume of orders in real terms. The national GDP deflators were calculated from
(mostly quarterly) raw data produced by national statistical bureaus. Data sources were the
Statistisches Bundesamt (Germany), the Institut National de la Statistique et des tudes
conomiques (France), the Office for National Statistics (UK), the Istituto Nazionale di
Statistica (Italy), and the Instituto Nacional de Estadstica (Spain).
There are alternative measures for inflation that could be considered. A widely known
measure of changes in price is the Consumer Price Index. However, this reflects only the price
of goods and services of a certain basket - purchased for the purpose of consumption by
households. Equally, the Producer Price Index refers to prices of a basket purchased from
producers. Indeed, the GDP deflator is a much broader price index because it reflects the
prices of all domestically produced (and even not-yet consumed) goods and services in an
economy beyond an often changing basket of goods.

4.
Statistical Analysis and Test Design
4.1
Regression Analysis: Determining Forecast Errors
4.1.1 Forecast Statistics
As a first step, we determined demand forecast errors. To ensure appropriate forecasting, we
evaluated state-of-the-art methods available in common software packages (e.g., SPSS, Stata,
Eviews). Among them have been exponential smoothing, Holt-Winters, autoregressive
integrated moving average (ARIMA), and autoregressive conditional heteroscedasticity
models.
4.1.2 Model Fit Statistics
We evaluated forecasting methods for each time series by criteria such as the highest
predictive accuracy. Results are illustrated in Table 1. We selected forecast models with the
lowest mean absolute percent errors which are as low as around 3% in our sample. Also,
depending on the nature of the data, adjusted R-squared values - mostly in the range between
90% and 95% - confirm that the forecast models were reasonably specified. To proceed
efficiently with the statistical tests as described below, we applied ARIMA (p,d,q)x(P,D,Q)s
forecast models.

4.2
Regression Diagnostics
4.2.1 Autocorrelation Tests
A key assumption in regression analysis is that residuals are serially independent, i.e., they do
not correlate over time. Such autocorrelation can be investigated by several tests. The first and
most commonly used test is the Durbin-Watson; our analysis shows values near 2, i.e.,
between the recommended limits of 1.5 to 2.5, indicating non-autocorrelation (see Table 1).
Since this test examines only for correlation between the current error and the immediately
preceding error (and not over more lagged error terms), we performed the Breusch-Godfrey
Lagrange multiplier test (Godfrey, 1978). Also these test results indicate that the assumption
of independence is not violated (see Table 1). Thus, the forecast models are well-fitted.
4.2.2 Heteroscedasticity Tests
The homogeneity of the residuals variance of is another vital assumption in regression
analysis. A well-fitted forecast model shows constant variance of prediction errors. Yet
heteroscedasticity could also be inherent in the data of a well-specified model.
To investigate the validity of homoscedasticity, we first evaluated obvious signs such as the
presence of outliners (see checklist in Gujarati and Porter, 2009, pp. 366-370). Second, we
performed several formal tests to detect heteroscedasticity. Basically, each test makes
assumptions about the structure of heteroscedasticity or the distribution of residuals, and
regresses on different numerical variants of residuals, such as absolute or logged values.
In line with our previous research, we started with the Whites test (White, 1980). In essence,
the Whites test is computed by an auxiliary regression; the squared residuals are regressed on
all possible (non-redundant) cross products of the regressors. Normally distributed errors are
not assumed. It tests the null hypothesis that the variance of the forecast errors is homogenous.
Overall, the White test rejects the presence of homoscedasticity in 85% of the times series at
the 5% significance level. For the German sample (as illustrated in Table 1) 12 out of 15 time
series behave heteroscedastically.
To confirm preliminary results, more restrictive tests than the White were additionally carried
out to see whether heteroscedasticity is present. Building on the methodology of BreuschPagan-Godfrey (Breusch and Pagan, 1979; Godfrey, 1988), the Harvey test regresses the logs
of the squared residuals on the original regressors (Harvey, 1976), whereas the Glejser test
uses the absolute residuals for the auxiliary regression (Glejser, 1969); the ARCH test
regresses the squared residuals on lagged squared residuals and a constant. Also, the more
restrictive tests provide strong evidence of heteroscedastic forecast errors. In terms of the
German sample - as highlighted in Table 1 -, all Breusch-Pagan-Godfrey tests and all Glejser
tests for each time series are found to be significant at the 5% confidence level. Harvey shows
66% and ARCH tests 73% of the statistical phenomenon to be significant at the same
confidence level. (The four more restrictive tests were performed with Eviews 7.0.)
To summarize, as is evident from the analysis, heteroscedasticity seems to be inherent in the
time series data across the five sample countries. The study leaves a low probability that
heteroscedasticity could be a consequence of misspecified forecast models.

Time Series
(Germany)
Industry
or
Type of Goods Manufactured

Model Fit Statistics

Forecast
Statistics
Forecast
Model

Mean
Absolute
Percent
Error

Root
Mean
Square
Error

Adjusted
R-Square

Autocorrelation
Tests
Durbin- BreuschWatson Godfrey

Heteroscedasticity
Tests
White

BreuschPaganGodfrey

Harvey

Glejser

ARCH

Manufacture of Food
Products

ARIMA
(1,1,7)(1,1,1)

3.8%

465

93%

1.99

0.0021

0.0025

0.0002

0.0657

0.0012

0.0001

Manufacture of Beverages

ARIMA
(2,1,0)(0,1,1)

3.3%

412

84%

2.09

0.0001

0.0001

0.0001

0.0446

0.0014

0.0113

Manufacture of Tobacco
Products

ARIMA
(2,1,0)(0,1,1)

3.8%

509

81%

2.05

0.0144

0.1143

0.0036

0.1378

0.0107

0.0955

Manufacture of Textiles

ARIMA
(2,1,0)(0,1,1)

3.9%

598

94%

1.94

0.0001

0.0001

0.0001

0.0001

0.0001

0.8606

Manufacture of Wearing
Apparel

ARIMA
(1,1,6)(0,1,0)

6.5%

939

98%

2.04

0.0396

0.0003

0.0001

0.0001

0.0001

0.0098

Manufacture of Paper and

Paper Products

ARIMA
(3,1,0)(1,0,1)

3.7%

434

89%

1.92

0.0363

0.0039

0.0042

0.0258

0.0027

0.5177

Manufacture of Chemicals and

Chemical Products

ARIMA
(2,1,12)(0,1,1)

3.1%

364

94%

1.98

0.0001

0.0011

0.0003

0.2149

0.0031

0.0014

Manufacture of Basic
Pharmaceutical Products and
Pharmaceutical Preparations

ARIMA
(2,1,0)(0,1,1)

4.1%

441

93%

2.09

0.0001

0.0101

0.0008

0.0111

0.0011

0.5998

Manufacture of Basic Metals

ARIMA
(2,1,2)(0,1,1)

4.4%

508

96%

1.86

0.0007

0.0001

0.0001

0.0003

0.0001

0.0001

Manufacture of Fabricated
Metal Products (except
Machinery and Equipment)

ARIMA
(3,1,10)(0,1,1)

3.6%

436

94%

2.08

0.0007

0.0001

0.0001

0.2019

0.0001

0.0001

Manufacture of Computer,
Electronic and Optical
Products

ARIMA
(3,1,0)(0,1,1)

5.7%

654

83%

2.03

0.0001

0.2437

0.0801

0.2710

0.0520

0.4502

Manufacture of Electrical
Equipment

ARIMA
(0,1,2)(0,1,1)

4.5%

574

85%

2.18

0.0345

0.0008

0.0171

0.2770

0.0708

0.0015

Manufacture of Machinery
and Equipment

ARIMA
(0,1,10)(0,1,1)

4.0%

470

95%

2.01

0.0005

0.0001

0.0001

0.0001

0.0001

0.0001

Manufacture of Motor
Vehicles, Trailers and SemiTrailers

ARIMA
(2,1,2)(0,1,1)

4.4%

431

96%

1.86

0.0001

0.1085

0.0023

0.0034

0.0016

0.0009

Capital Goods

ARIMA
(3,1,0)(0,1,1)

3.3%

376

95%

2.24

0.0294

0.0001

0.0065

0.0038

0.0020

0.0001

Table 1. Regression Analyses and Regression Diagnostics

(Example: Time-Series Data from Germany)

5.
Estimated Inventory Performance Improvements
Given the statistical phenomenon of heteroscedastic forecast errors, the natural question arises
how much safety stocks could be saved whilst ensuring that the customer service level
remains unchanged. Therefore, we aim to estimate cost savings by comparing three generic
safety stock policies:
a) Conventional Safety Stock Planning Policy: Safety stock targets are planned invariant to
time (production planner does not consider heteroscedastic forecast errors). A single-valued
standard deviation is applied to estimate expected shortages.

b) Adaptive Safety Stock Planning Policy: Single-valued safety stock levels are aimed at for
the planning horizon. Expected shortages take into account monthly standard deviations of
forecast errors.
c) Time-phased Safety Stock Planning Policy: Time-varying safety stock targets are devised
(the production planner is aware of heteroscedastic forecast errors) taking also the monthly
varying forecast errors into account.
5.1
Determining Time-Varying Safety Stock Levels based on Fill Rate Targets
Imagining a typical planning situation to determine monthly safety stock targets, we assume
that the production and inventory planner aims to keep the yearly amount of safety stock costs
(SSC) to a minimum (objective function), and provide a certain product availability to internal
or external customers (constraint). C denotes variable production or purchasing costs per unit
and h the inventory holding costs per unit.
12

min SSC =

(1)

SS (c + h ) 12
t

t =1

We assume a planning approach using a fill rate target. Although the measure is widely used
in practice, definitions of it can vary greatly. It may dependent on supplier contracts, on
industry practice, or on internal control requirements. At least three cases can be distinguished.
First, the goal could be to meet a specific customer service on average for the planning
horizon (total fill rate); second, to meet service for each planning period (marginal fill rate);
third, to meet service on average (simple yearly average). Since the first two cases are the
more likely ones, we set both values to 99% for our purposes:
; (ii) FRm arg inal 0.99

(i) FRtotal 0.99

(2)

The fill rate is typically measured as the ratio of the average number of units of demand filled
by the average demand (in a periodic review inventory system over an infinite horizon).
However, the fill rate underestimates the fill rate achieved over a finite horizon (Chen et al.,
2003). Here, we calculate the fill rate with expected shortage per replenishment cycle (ESC)
and the average monthly demand () in a replenishment cycle:
(3)

12

(i) FRtotal = 1

ESC

ESC

= 1

; (ii) FRm arg inal = FRt = 1

t =1
12

ESCt

t =1

The relation between fill rate and expected value of the number of units of shortage per
replenishment cycle can be expressed in a formal way (see Silver et al., 1998, p. 268; Chopra
and Meindl, 2012) as
ESCt =

( x ROP) f ( x)dx

(4)

x = ROP

with the density function, f(x), and reorder point (ROP). Given normally distributed demand
during lead time, (4) can be simplified to

 SS
ESCt = t f s t
t

SS
SSt 1 FS t

(5)

Notably, we assume the fill rate to be invariant to time. Certainly, fill rates may reflect time
varying service level agreements, different stream of demands etc.
5.2
Potential Inventory Savings - Illustrated by a Case
We illustrate the saving potential in the following fictional example. A manufacturer forecasts
the average demand as shown in Table 2 - obviously a seasonal pattern with a peak in April.
Forecast errors perceptibly vary. As demonstrated in section 4, regression diagnostic supports
the hypothesis of heteroscedastic errors. Would it be prudent for the manufacturer to
contemplate switching from a conventional via an adaptive to a time-phased safety stock
policy? Dependent on the three policy options, the planner uses different input values.
(1) Conventional
Safety Stock Planning

Input Data

Month

Forecasted
Demand

Forecast
Error

(2) Adaptive
Safety Stock Planning

(3) Time-phased
Safety Stock Planning

Single-valued Standard Deviation

Varying Standard Deviations

Varying Standard Deviations

Constant Safety Stock Targets

Constant Safety Stock Targets

Varying Safety Stock Targets

Safety
Stock
Target

Expected
Shortage

Total
Fill
Rate

Safety
Stock
Target

Expected
Shortage

Total
Fill
Rate

Safety
Stock
Target

Expected
Shortage

Total
Fill
Rate

(M ean)

(Standard
Deviation)

9500

400

397

34

100%

376

37

100%

251

64

99%

10000

450

397

47

100%

376

51

99%

279

73

99%

11000

450

397

47

100%

376

51

100%

288

71

99%

13000

500

397

61

100%

376

65

99%

309

82

99%

11000

750

397

142

99%

376

148

99%

470

121

99%

9500

800

397

159

98%

376

166

98%

507

128

99%

9500

400

397

34

100%

376

37

100%

248

65

99%

7500

350

397

22

100%

376

25

100%

218

57

99%

7500

400

397

34

100%

376

37

100%

250

65

99%

10

8000

800

397

159

98%

376

166

98%

521

124

98%

11

9000

890

397

191

98%

376

198

98%

565

142

98%

12

8500

650

397

108

99%

376

114

99%

415

103

99%

397

86

376

91

9500

600

(Average)

(RM S)

(Average) (Average)

99%
(Total)

(Average) (Average)

99%

360

91

99%

(Total)

(Average)

(Average)

(Total)

-4.3%
4.3%
- 9.3%
9.3%

Table 2. Potential Safety Stock Savings comparing three Generic Planning Approaches
Being unaware of heteroscedastic forecast errors - when following a conventional safety
stock planning policy, the manager calculates expected shortages with a single-valued
standard deviation that is a constant value throughout the planning period. The policy is
illustrated in (1) in Table 2. He might use the root mean square (RMS) of monthly standard
deviations of forecast errors of previous periods (here: 600). Such a value could be suggested
by his inventory system as an inexact, average, standard deviation. From these data, the
planner computes constant safety stocks of 397 per month for the entire planning horizon while minimizing yearly safety stocks under the constraint of the total fill rate of 99%. To

reflect the degree of monthly uncertainties leading to different monthly expected shortages,
the manager uses the monthly varying standard deviations of forecast errors. The result is
average expected shortages of 86 over the year.
If the planner takes heteroscedastic forecast errors into account (when applying a time-phased
safety stock policy), he will look for safety stock targets that could vary monthly - see (3) in
Table 2 for more details. The manager considers the monthly varying standard deviation of
the forecast errors rather than its aforementioned root mean square. Following this approach
with the total fill rate of 99%, the average safety stocks per year are 9.3% less, i.e., 360 (37
less than with the conventional policy). Similarly, remarkable savings could be also
achieved when a marginal fill rate of 99% is targeted (this case is not illustrated in Table 2):
the average safety stock target per year decreases to 363 (and equals safety stock savings of
8.5%).
Adaptive safety stock planning - see (2) in Table 2 - could be interpreted as an intermediate
step from a conventional approach to the proposed time-phased safety stock planning policy.
In adopting this approach, the planner recognizes the nature of varying demand forecast errors
but seeks to gauge a single-valued safety stock level for each planning period. The
intermediate planning approach leads to possible safety stock savings of 4.3%.
5.3
Potential Inventory Savings - Illustrated by Industry Data
For this subsection, we use the same aggregated EUROSTAT data as in section 4. By
analyzing data from different aggregation levels, we report on possible inventory savings if
the proposed time-phased safety stock planning policy is followed instead of a conventional
safety stock policy. The results can serve as an estimate for the impact on stock keeping units.
The estimate might be a conservative measure since company data typically show higher
variances than data on a wide economic basis.
5.3.1 Product-specific Safety Stock Savings
The following Table 3 illustrates the saving potential for the five types of goods manufactured
across the five European economies. The lower number (left) represents possible savings
following an adaptive safety stock policy - the upper number (right) corresponds to the timephased safety stock policy. For example, manufacturers offering non-durable consumer goods
in Spain could obviously benefit from time-varying safety stocks (15.7%), whilst those
offering intermediate goods in Italy seem to have less monetary advantages (2.0%).
Time Series

Potential Safety Stock Savings

Type of Goods Manufactured

Germany

France

UK

n/a

3.3% - 4.2%

1.2% - 2.5%

0.9% - 2.0%

5.7% - 6.6%

1.5% - 11.3%

1.2% - 5.5%

0.7% - 5.1%

0.9% - 2.3%

3.3% - 6.5%

Consumer Goods

n/a

6.0% - 7.2%

2.6% - 4.7%

2.4% - 3.6%

3.5% - 7.0%

Durable Consumer Goods

n/a

0.6% - 5.2%

2.0% - 5.8%

4.5% - 11.8%

2.7% - 7.6%

Non-durable Consumer Goods

n/a

1.0% - 9.9%

2.1% - 5.2%

2.8% - 7.7%

0.2% - 15.7%

Intermediate Goods
Capital Goods

Italy

Spain

Table 3. Product-specific Safety Stock Savings

(Example: Time-Series Data from Five Sample Countries)

5.3.2 Industry-specific Inventory Savings

As one might expect, industry savings across industries and countries vary widely - from 1.2
to 21.0% (Table 4). For example, the manufacture of wearing apparel demonstrates the
greatest possible performance effects in Germany (10.4%), whereas the manufacture of
computer, electronic and optical products shows the least (2.2%). Likewise, companies
producing fabricated metal products in France (9.2%), chemicals and chemical products in the
UK (6.5%), basic pharmaceutical products and pharmaceutical preparations in Italy (9.5%),
and wearing apparel in Spain (21.0%) may profit. Overall, we find that numerous
manufacturers from Spain would benefit greatly when time-phased safety stocks are applied
(on average 7.3%). Conversely, we discovered that a range of UK industries would
experience only marginal effects (on average 4.1%).
Potential Safety Stock Savings

Time Series
Germany

Industry

France

UK

Italy

Spain

Manufacture of Food Products

2.0% -

7.7%

1.5% - 2.7%

0.7% - 2.8%

0.9% - 2.9%

2.5% - 5.9%

Manufacture of Beverages

0.7% -

4.7%

n/a

n/a

n/a

1.6% - 9.6%

Manufacture of Tobacco Products

1.1% -

5.1%

n/a

n/a

n/a

0.8% - 2.2%

Manufacture of Textiles

2.7% -

9.1%

n/a

2.5% - 5.9%

n/a

7.0% - 7.5%

Manufacture of Wearing Apparel

8.0% - 10.4%

0.6% - 1.2%

1.5% - 2.5%

1.3% - 4.3%

18.4% - 21.0%

Manufacture of Leather and Related Products

n/a

n/a

n/a

n/a

14.3% - 16.0%

Manufacture of Wood and of Products of

Wood and Cork (except Furniture)

n/a

n/a

n/a

n/a

3.0% - 3.7%

Manufacture of Paper and Paper Products

1.2% -

0.6% - 4.9%

0.5% - 2.0%

3.2% - 5.1%

1.5% - 9.0%

n/a

n/a

n/a

n/a

4.9% - 9.2%

n/a

n/a

n/a

n/a

4.4% - 4.5%

Printing and Reproduction of Recorded

Media
Manufacture of Coke and Refined Petroleum
Products
Manufacture of Chemicals and Chemical
Products
Manufacture of Basic Pharmaceutical
Products and Pharmaceutical Preparations

5.7%

0.5% -

4.9%

1.7% - 4.3%

1.2% - 6.5%

0.6% - 2.9%

2.4% - 8.4%

0.8% -

3.0%

0.7% - 3.0%

5.0% - 6.4%

1.8% - 9.5%

1.6% - 4.2%

Manufacture of Rubber and Plastic Products

n/a

n/a

n/a

n/a

3.9% - 7.9%

Manufacture of other Non-Metallic Mineral

Products

n/a

n/a

n/a

n/a

8.3% - 10.1%

Manufacture of Basic Metals

2.2% -

6.5%

2.9% - 6.8%

0.9% - 2.3%

2.7% - 5.3%

2.5% - 2.6%

0.9% -

4.7%

4.9% - 9.2%

3.6% - 5.6%

1.3% - 2.0%

2.0% - 3.9%

1.5% -

2.2%

4.7% - 4.7%

3.5% - 6.1%

4.2% - 6.1%

6.7% - 5.7%

Manufacture of Electrical Equipment

2.1% -

5.6%

2.6% - 4.4%

1.0% - 2.1%

3.9% - 4.7%

2.7% - 4.8%

Manufacture of Machinery and Equipment

0.8% -

2.8%

2.8% - 5.9%

0.8% - 2.1%

4.1% - 4.3%

6.1% - 7.1%

Manufacture of Motor Vehicles, Trailers and

Semi-Trailers

1.2% -

4.6%

1.9% - 2.2%

3.6% - 4.8%

2.1% - 2.3%

2.2% - 4.9%

Manufacture of Fabricated Metal Products

(except Machinery and Equipment)
Manufacture of Computer, Electronic and
Optical Products

Manufacture of Furniture

n/a

n/a

n/a

n/a

2.6% - 5.5%

Repair and Installation of Machinery and

Equipment

n/a

n/a

n/a

n/a

5.0% - 7.1%

Table 4. Industry-specific Inventory Savings

(Example: Time Series Data from Five Sample Countries)

10

5.3.3 Market-specific Inventory Savings

The aforementioned data sets can be further broken down - whether new orders are received
from the domestic or non-domestic market. To highlight some results, we present selected
industries from France and UK, which were chosen due to its high level of exports. For
example, we found that a UK manufacturer of paper and paper products exposed to domestic
demand (26.2%) seems to profit more than those in machinery and equipment (2.4%) when
applying time-varying safety stocks (Table 5). French manufacturers facing demand from
abroad seem to benefit more than the UK counterparts.
Time Series

Potential Safety Stock Savings

France

Industry

Domestic
Demand

UK
Non-Domestic
Demand

Domestic
Demand

Non-Domestic
Demand

Manufacture of Wearing Apparel

2.5% -

3.2%

6.0% - 6.9%

2.7% - 3.6%

6.9% - 8.1%

Manufacture of Paper and Paper Products

1.0% -

4.2%

2.2% - 7.2%

6.8% - 26.2%

2.7% - 3.3%

1.6% -

3.2%

1.8% - 3.8%

1.6% - 4.7%

0.7% - 1.1%

3.5% -

4.4%

8.2% - 9.7%

4.1% - 5.0%

3.0% - 3.8%

6.6% -

8.7%

2.4% - 3.2%

1.5% - 2.4%

2.4% - 3.5%

Manufacture of Chemicals and Chemical

Products
Manufacture of Computer, Electronic and
Optical Products
Manufacture of Machinery and Equipment

Table 5. Market-specific Inventory Savings

(Example: Selected Time-Series Data from France and UK)
Furthermore, we analyzed non-domestic demand received from the euro-zone and beyond.

6.
Conclusion
This paper has aimed to explore the pervasiveness of time-phased safety stock planning in
Europe. Specifically, our main contributions in this paper are threefold.

First, we have provided evidence of heteroscedastic forecast errors with an econometric study
using a large sample of data across the five largest European economies. This insight seems
not fully utilized in safety stock calculations. To our best knowledge, this paper is possibly
among the very first contributions to document the statistical effect in Europe. Furthermore, it
contributes to the emerging stream of empirical literature studying inventory behavior.
Second, the paper has practical implications for European managers. It demonstrates that
time-varying safety stock targets can improve inventory performance. The illustrated safety
stock saving potential on various aggregation levels could be used as the starting point of an
inventory benchmark to re-assess safety stock models. To supplement these performance
effects, it might also be fruitful to focus on more granular company data and disaggregated
stock keeping unit data sets.
Finally, opportunities for future research arise for the modeling of time-phased safety stocks.
Given the importance and managerial benefits outlined here, we hope we provided a small
step in promoting this intriguing new path of research in Europe.

11

7.
References
Breusch, T. S., Pagan, A. R., 1979. A simple test for heteroskedasticity and random
coefficient variation. Econometrica, 48, 1287-94.
Chen, J., Lin, D. K., Thomas, D. J., 2003. On the single item fill rate for a finite horizon.
Operation Research Letters, 31, 119-23.
Chopra, S., Meindl. P., 2012. Supply chain management - strategy, planning, and operations.
Upper Saddle River, NJ: Pearson Prentice-Hall.
Glejser, H., 1969. A new test for heteroscedasticity. Journal of the American Statistical
Association, 64, 316-23.
Godfrey, L. G., 1978. Testing against general autoregressive and moving average error
models when the regressors include lagged dependent variables. Econometrica, 46, 1293302.
Godfrey, L. G., 1988. Specification tests in econometrics. Cambridge: Cambridge University
Press.
Graves, S. C., 1988. Safety stocks in manufacturing systems. Journal of Manufacturing and
Operations Management, 1 (1), 67-101.
Graves, S. C., 1999. A single-item inventory model for a nonstationary demand process.
Manufacturing & Service Operations Management, 1 (1), 50-61.
Graves, S. C., Willems, S. P., 2008. Strategic inventory placement in supply chains: nonstationary demand. Manufacturing & Service Operations Management, 10 (2), 278-87.
Grubbstrm, R. W., Tang, O., 1999. Further developments on safety stocks in an MRP system
applying Laplace transforms and input-output analysis. International Journal of
Production Economics, 60-61 (1), 381-7.
Gujarati, D. N., Porter, D. C., 2009. Basic econometrics. New York: McGraw-Hill/Irwin.
Harting, D., 2005. Vorrats- und Bestandsmanagement schlank und modern. BA Beschaffung
aktuell, 6, 26.
Harvey, A. C., 1976. Estimating regression models with multiplicative heteroscedasticity.
Econometrica, 44, 461-65.
Hendricks, K. B., Singhal, V. R., 2009. Demand-supply mismatch and stock market reaction:
evidence from excess inventory announcements. Manufacturing and Service Operations
Management, 11 (3), 509-24.
International Monetary Fund (Ed.), 2011. World economic outlook database - nominal GDP
list of countries. http://www.imf.org.
Kanet, J. J., Gorman, M. F., Stlein, M., 2010. Dynamic planned safety stocks for
production planning. International Journal of Production Research, 48 (22), 6859-80.
Lee, H. L., Billington, C., 1993. Material management in decentralized supply chains.
Operation Research, 41 (5), 835-47.
Mahadevan, B., 2009. Operations management: theory and practice. New Delhi: Prentice Hall.
Minner, S., 2000. Strategic safety stocks in supply chains. Berlin: Springer.
Neale, J. J., Willems, S. P., 2009. Managing inventory in supply chains with nonstationary
demand. Interfaces, 39 (5), 388-99.
Obermaier, R., Donhauser, A., 2009. Disaggregate and aggregate inventory to sales ratios
over time: the case of German corporations 1993-2005. Logistics Research, 1 (2), 95-111.
Silver, E. A., Pyke, D. F., Peterson, R., 1998. Inventory management and production planning
and scheduling. New York: John Wiley & Sons.
Stlein, M., Minner, S., 2012. Empirical evidence for time-phased safety stocks planning
based on European econometric data (working paper in progress).
White, H., 1980. A heteroskedasticity-consistent covariance matrix and a direct test for
heteroscedasticity. Econometrica, 48, 817-38.
Zipkin, P., 2000. Foundations of inventory management. New York: McGraw-Hill/Irwin.

12