Hedging weather risk and coordinating supply chains

Abstract

The sales of many products can be influenced by weather conditions, positively or negatively. For the manufacturers in question, one of their entrepreneurial risks is to incur lower than expected sales because of adverse weather conditions. The variability of weather conditions is expected to continue to rise because of climate change. Manufacturers can choose to do nothing and suffer the financial consequences, or transfer the weather risk partly or wholly to others. This paper presents an approach to transfer weather risks to risk takers and reduce sales volatility using weather index-based financial instruments. In our approach, the risk of adverse weather conditions is calculated on the basis of adverse conditions observed in the past. We do not use forecasts of weather conditions. We illustrate our action design with case studies of three companies: a company manufacturing automotive replacement parts, a clothing company and a company producing of sunscreen products. We demonstrate its efficiency in reducing cash-flow uncertainty and potential losses caused by adverse weather, and in influencing sales to the next tier.

Full text

Hedging weather risk and coordinating supply chains
Xavier Brusseta,∗, Jean-Louis Bertrandb
aSKEMA Business School, Université Côte d’Azur, Lille, France
bESSCA School of Management, Angers, France
Abstract
The sales of many products can be influenced by weather conditions, positively or negatively. For the manufacturers in
question, one of their entrepreneurial risks is to incur lower than expected sales because of adverse weather conditions.
The variability of weather conditions is expected to continue to rise because of climate change. Manufacturers can
choose to do nothing and suffer the financial consequences, or transfer the weather risk partly or wholly to others.
This paper presents an approach to transfer weather risks to risk takers and reduce sales volatility using weather index-
based financial instruments. In our approach, the risk of adverse weather conditions is calculated on the basis of adverse
conditions observed in the past. We do not use forecasts of weather conditions. We illustrate our action design with
case studies of three companies: a company manufacturing automotive replacement parts, a clothing company and
a company producing of sunscreen products. We demonstrate its efficiency in reducing cash-flow uncertainty and
potential losses caused by adverse weather, and in influencing sales to the next tier.
Keywords: design science, weather sensitivity, risk management, supply chain, coordination, weather hedge
1. Introduction
Weather plays an important role, directly or indirectly,
in determining both the demand and sales for many prod-
ucts (Chen and Yano,2010;Lazo et al.,2011). The fact that
temperature, rainfall and other weather variables have di-
rect effects on various economic series is not new (Steele,
1951;Granger,1978). In a context of rapid, chaotic and
threatening climate change, the issue of adaptation of
companies and the economy as a whole is of increasing
relevance. As a sign of this increasing importance to deci-
sion makers in all walks of life, the work of William Nord-
haus has just been rewarded with the 2018 Nobel Prize
for Economics for his work in the mid-1990s on an inte-
grated quantitative assessment model to describe the re-
lationship between the economy and climate.
In retail, weather determines the decisions of con-
sumers as to what they buy, in what quantity, when and
even where (Agnew and Thornes,1995). Weather condi-
tions do not only influence store traffic. They also change
consumers’ purchasing decisions about the product and
the amount of product they buy (Parsons,2001;Kirk,
2005). Even as retail is shifting from brick and mortar
stores to online shopping, the weather continues to exert
its influence on consumer behavior (Starr-McCluer,2000;
Lazo et al.,2011;Steinker et al.,2017;WeatherUnlocked,
2014).
∗Corresponding author
Email addresses: (Xavier Brusset),
(Jean-Louis Bertrand)
Although many studies have been done to analyze the
direct impact of weather on retailers’ sales (Starr-McCluer,
2000;Lazo et al.,2011), no similar research analyzes the in-
direct influence of the weather in the case of manufactur-
ers standing one or more echelons away from the end-user
market. Additionally, the risk of adverse weather condi-
tions has increased over time and is expected to continue
to do so because of climate change, not because of higher
temperatures or increased rainfall, but because of a higher
variability (WMO,2013;IPCC,2014).
Within the framework of Design Science Research (van
Aken et al.,2016), we apply the so-called CIMO-logic
(Denyer et al.,2008) to develop a new body of prescrip-
tive knowledge for management science. The problem-
in-Context (the C in CIMO) is the sensitivity of manu-
facturers’ sales to the erratic behavior of the weather. In
the event of adverse weather conditions over an extended
period of time, manufacturers may suffer a loss of sales
as replenishment orders drop in the current season, and
carried-over inventories and precautionary behaviors lead
to lower orders for the following season. Manufacturers
can choose not to do anything about this entrepreneurial
risk and bear the financial consequences, or transfer it
partly or wholly to others. This paper presents an ap-
proach to transfer this type of entrepreneurial risk to one
or more risk takers using weather index-based financial
instruments in the form of insurance or option contracts
that compensate for losses due to adverse weather.
This article proposes an Intervention based on the de-
sign of a bespoke financial instrument that allows each
manufacturer exposed to weather risks to transfer this
Preprint submitted to Journal of Operations Management Accepted October 17th, 2018, Online November 17th, 2018. DOI: 10.1016/j.jom.2018.10.002

risk and mitigate her1exposure to weather variability.
We develop a Design proposition which proceeds in sev-
eral steps: We identify the most significant weather vari-
ables, evaluate the sensitivity of a manufacturer’s sales to
a change in those significant weather variables, and es-
timate the potential losses based on historical observed
weather conditions. Knowledge of the significant weather
variables and their mathematical relationship to sales al-
lows the manufacturer and potential risk takers to have
a clear view of the weather risk in order to determine an
acceptable transfer price for each party. Through this Me-
chanism, the Outcome is a reduced exposure to adverse
weather conditions, which is measured both by the reduc-
tion of the maximum potential loss due to adverse wea-
ther, and the reduction in sales variability. In the event
of adverse weather conditions, the financial compensa-
tion to the manufacturer can be used simply to offset sales
losses or can be shared with her downstream partners to
improve supply chain coordination. The effectiveness of
the Intervention is demonstrated through its application
in three industrial cases in different sectors exposed to
weather variability.
The research questions addressed in this paper are the
following: (1) How can weather risks that affect man-
ufacturers be calculated so that all the parties involved
in the transfer of this risk can make an informed deci-
sion? (2) Can weather index-based financial instruments
reduce the financial consequences of adverse weather con-
ditions and improve coordination with downstream part-
ners in the supply chain?
In the paper, we provide a step-by-step detailed analy-
sis of the Intervention in the case of a manufacturer who
has limited access to data on sales to the end-consumer.
The effectiveness of the Intervention in the cases of the
other two manufacturers is provided in the Appendix.
The complete studies and data sets are available upon re-
quest from the authors. The main case study involves a
manufacturer of replacement parts in the automotive af-
termarket. The replacement part is a glow plug. It does
not sell if a winter is unusually warm. When this happens,
there is no replenishment order, and retailers have the
ability to carry the inventory of unsold products from one
winter to the next. Available retail sales data to end con-
sumers is fragmented and inconsistent, leading to further
material negative financial impacts (Meixell et al.,2008).
The manufacturer must understand her exposure in order
to protect herself. At the same time, she wishes to coax
her retailers into committing to order the same amount of
products year-on-year.
The second example concerns a manufacturer in the
cosmetics sector, and more specifically sunscreen prod-
ucts. The manufacturer places such products in drug-
stores, supermarkets, and pharmacies in spring, at the be-
ginning of the season. Additional products can be de-
1Following convention, we refer to the upper echelon in a supply
chain dyad as a ’she‘ and to the downstream partner as a ’he’.
livered during the season. Unsold products are returned
to the manufacturer by retailers at the end of the season.
Hence, the manufacturer suffers most of the financial con-
sequences of a bad summer. Due to the diversity of the re-
tail networks and the lack of consistency between the way
they report sales, we used weekly marketing panel data to
establish the weather-sensitivity model.
The third example is drawn from the clothing sec-
tor. Retailers order clothing from manufacturers several
months ahead of the selling season, and financial losses
caused by unsold products or discounted products are
mostly borne by the retailer. Hence, manufacturers are
looking for ways to encourage retailers to be less cautious
when ordering goods by offering protection on lost sales
caused by unfavorable weather. To this end, they need
to understand ex post how sales are affected by adverse
weather conditions defined as the difference between ob-
served weather and its normal value. In this third case,
we used a monthly sales survey to conduct the analysis to
address the lack of consistent retail sales data provided to
the manufacturer by a highly diversified distribution net-
work.
Whilst weather risk hedging was actually implemented
in all three examples, in the first two cases weather risk
hedging was done to reduce the own exposure of manu-
facturers to weather variability, whereas in the third one,
the manufacturer of glow plugs acted as the risk taker for
her retail network with the objective to increase sales.
In the next section, the literature is reviewed. After a
presentation of the data sets in §3, follows a description
of the methodology for building the weather-sensitivity
model in §4. In §5, we present the Design proposition in
the case of a manufacturer in the automotive sector. §5.1
shows how the weather risk is calculated so that an index-
based financial product can be constructed. §6illustrates
how the Intervention leads to reduced weather risk. Once
the manufacturer has hedged her exposure, in §6.2 we
present the example of a manufacturer who uses the Inter-
vention to coordinate her retailers, enhance their loyalty,
and more generally reinforce supply chain coordination.
In §6.3 the examples are extended to other coordination
mechanisms. The application of the general methodology
to the other two industrial cases is presented in Appendix
A. The conclusion is presented in §7.
2. Literature review
Weather observations have long been incorporated into
statistical models to explain various outputs. Notably, the
role of weather on mood has been extensively studied
(Goldstein,1972;Jorgenson,1981;Sanders and Brizzolara,
1982;Howarth and Hoffman,1984;Albert et al.,1991), as
has its effects on consumer behavior (Cunningham,1979;
Schneider et al.,1980;Parsons,2001).
Manufacturers are different from retailers because they
do not usually sell directly to the end-consumer but to dis-
tributors or wholesalers. Their ability to evaluate their
2

own sales exposure to weather is further complicated by
the paucity and inconsistency of information about retail-
ers’ sales and inventories.
Existing methodologies to determine the effects of wea-
ther consist of a direct cross-analysis between series of
historical sales and weather data (Agnew and Thornes,
1995;Starr-McCluer,2000;Lazo et al.,2011;Bertrand et al.,
2015). For most manufacturers, these are ineffective in iso-
lating weather as a distinct factor affecting sales, due to the
inventory and sales strategies of the various retailers in-
volved, as well as inconsistent granularity of data (Mason-
Jones and Towill,1999;Lee et al.,2000;Chen,2004).
The effects generated by the inability to monitor final
market demand, first ascribed to the dynamics of sup-
ply chains in Forrester (1965), lead to the characteriza-
tion of the Bullwhip Effect (Lee et al.,1997a,b). The Bull-
whip Effect has important implications on qualitative and
quantitative performance measures of customer satisfac-
tion (Harland,1996;Kouvelis et al.,2006) as well as mar-
ket share. The most common factors cited to explain this
effect are information delays (e.g., orders based on in-
ventory information which is several weeks old), inertia
(once orders are received, the production schedule is not
changed immediately), and lack of information about fi-
nal demand (Disney et al.,1997;Stone,2012). Whilst
the seasonality of retail sales is well-documented and ac-
counted for by supply chain managers and sales exec-
utives (Nieves-Rodriguez et al.,2015), the same is not
true for manufacturers exposed to the unexpected conse-
quences of weather, whether these are favorable or unfa-
vorable.
Several studies analyze how weather affects retail sales,
such as Agnew and Thornes (1995) for the UK food re-
tail industry, Starr-McCluer (2000) and Lazo et al. (2011)
for US retail sales. Other studies focus on estimating
how weather directly affects consumer decisions, sales vol-
umes, production costs, yields, prices, margins and prof-
its for small and large firms operating in a large variety of
sectors (Linden,1962;Gagne,1997;Connolly,2008;Mur-
ray et al.,2010;Lucas and Lawless,2013). They all confirm
weather is a significant explanatory factor.
Existing methodologies used to isolate the effect of wea-
ther on sales typically require the ability to directly com-
pare large amounts of business data with weather obser-
vations (Bahng and Kincade,2012;Steinker et al.,2017;
Bertrand et al.,2015;Starr-McCluer,2000;Lazo et al.,
2011). They also make assumptions about or control for
factors likely to cause a change in demand (e.g. market-
ing, advertising, inventory variations, lagged sales vari-
ables, retail store traffic, day of the week, holidays, price
markdowns) (Parsons,2001;Starr-McCluer,2000;Bahng
and Kincade,2012).
Further, when it comes to assessing demand, the liter-
ature generally considers that sales to end-consumers are
known to the upstream partners (Lee et al.,2000;Gavir-
neni et al.,1999) and information can be conveyed to
them in a timely fashion (Chen,2004) or that action can
be taken to overcome demand variability (Skjoett-Larsen
et al.,2003). However, in many sectors, none of these three
conditions are met and demand variability or uncertainty
still generates a severe Bullwhip Effect (Bray and Mendel-
son,2012). Because of this effect, the nature of manufac-
turers’ exposure to weather is different from that of retail-
ers: weather affects consumers, who in turn affect retail
sales, which in turn affect restocking and orders to manu-
facturers.
With climate change and increasing climate variability,
a new interest in understanding the relationship between
weather and economics has emerged (Dell et al.,2014;
Auffhammer et al.,2013, and references therein). Higher
weather volatility implies higher impact on manufactur-
ers. A different approach is needed to allow manufactur-
ers to assess their losses due to adverse weather conditions
when information provided by retail partners is scarce,
inconsistent or limited, and to protect themselves from
the financial consequences of such adverse weather con-
ditions (Deschênes and Greenstone,2011;Bertrand et al.,
2015;Bertrand and Brusset,2018).
One way to measure this risk is known as Weather
Value-at-Risk (Weather-VaR), which represents the max-
imum expected loss in sales caused by adverse weather
conditions over a given period of time for a given level
of confidence (Toeglhofer et al.,2012;Linsmeier and Pear-
son,2000). In finance, Value-at-Risk is a measure of losses
resulting from market volatility. By analogy, weather-VaR
measures losses resulting from weather variability. The
implementation of Weather-VaR requires information on
(a) the weather-dependent distribution of sales, which
is determined by both the sensitivity of sales to changes
in weather variables and their exposure to weather vari-
ability (Prettenthaler et al.,2016), and (b) the distribu-
tion of the historical weather observations. Knowledge of
Weather-VaR makes it possible to respond to weather risk
(Merna and Al-Thani,2011) through risk mitigation poli-
cies (Chen and Yano,2010;Tsay,2001;Padmanabhan and
Png,1995;Barrieu and Scaillet,2010).
Our article addresses this issue by providing manufac-
turers with a methodology to evaluate the weather effect
as distinct from the other effects affecting demand, and
design the relevant bespoke financial instrument. With
this methodology, manufacturers are now in a position to
mitigate the effects of adverse weather for themselves and
for their retailers.
3. Data
Two types of data are required to be able to model the
influence of weather on sales. For sales data, the more
data there is and the more local it is, the better. Weather
data must be measured as close as possible to where sales
take place.
3

3.1. Sales data
Ideally, a minimum history of two to three years of
monthly sales data is required to provide enough data to
allow for statistical analysis. Sales data must be associ-
ated with a geographical area (e.g., city, region or coun-
try), and the time step should be as short as possible (e.g.,
daily or weekly). Data can be the manufacturer’s sales to
the intermediate customers in the supply chain, such as
retailers and wholesalers, or better, the figures of sales to
final consumers, provided by retailers or other sources.
In our main example, an international automotive Orig-
inal Equipment Manufacturer (OEM) sells a replacement
part for vehicle repair through several retail channels in
the automotive aftermarket sector. The manufacturer has
five years ofmonthly sales to French retailers. In the trade,
sales between manufacturers and retailers are termed
“Sell-In” (SI). The second data set consists of monthly re-
tail sales to the final end-user. They are termed “Sell-Out”
(SO). Retailers are reluctant to provide SO data to the
manufacturer to prevent her from being able to figure out
their exact market share. The SO data set was bought by
the OEM from a third-party marketing consultancy firm.
It only covers four regions in France.
In the case of sunscreen products, SO data of sunscreen
come from marketing panels and span eleven years with
a time-step of rolling four week periods. In the clothing
sector, sales data come from a monthly survey published
by the Institut Français de la Mode (IFM), the most im-
portant French retail clothing organization. The data cov-
ers the period between 2000 and 2013. The survey reports
sales from all distribution channels representing approx-
imately 80% of total clothing sales in France to final con-
sumers (SO data).
3.2. Weather data
There are three categories of weather data: synoptic, cli-
mate and reconstructed (Dischel,2002). Synoptic data are
measured automatically at weather stations and are usu-
ally used for short-term forecasts. Climate data are also
measured at a station, but it is quality-controlled and cer-
tified by national weather services. In particular, climate
data are corrected and reconstructed because of missing
data due to causes such as failures in the measuring sys-
tem, discontinuities resulting from a change of measure-
ment technology, changes in the environment of a sta-
tion (a new building), or the displacement of a station
(Jewson and Whitehead,2001;Boissonade et al.,2002).
Climate data are the only data that should be used in
econometric analysis to avoid the common pitfalls that
empirical researchers face when using historical weather
data as explanatory variables in econometric applications
(Auffhammer et al.,2013). Climate data must then be
transformed into economic analysis variables, that is, they
must be consistent with the geographical mesh of sales
data. In most cases, demand is driven by the size of the
population. In all three case studies, we use the result in
§2.2.5 of Dell et al. (2014) and the methodology provided
in Quayle and Diaz (1980) to aggregate climate data into
weather variables for analysis using regional population
weights.
The meteorological data is provided by Meteo France,
the French national weather service. For research pur-
poses, data can also be extracted from the National
Oceanic and Atmospheric Administration GSOD data-
base. There are no discontinuities, missing data or change
in weather stations (Quayle and Diaz,1980;Dell et al.,
2014). In all three cases, we used average, minimum and
maximum temperatures, precipitation, humidity rate and
wind speed observations. A linear trend calculation for
each station was used to account for potential effects of
climate change.
4. Methodology
The methodology we develop aims at identifying and
quantifying the effects of weather on the sales of manufac-
turers to retailers in order to estimate the potential losses
caused by adverse weather conditions and design the rele-
vant weather index-based financial instrument to mitigate
these risks.
The straightforward cross-comparison between Sell-In
data and weather observations often proves unsuccess-
ful because, in most cases, various types of retailers, each
with different sales and inventory strategies, stand be-
tween manufacturers and the end-consumers who are the
ones that are directly affected by changes in the wea-
ther. Besides, information on retail sales to the end-
consumers, which we term Sell-Out, is frequently not uni-
formly and completely available to manufacturers. The
general methodology we expose is designed to overcome
these hurdles and allow manufacturers with limited infor-
mation from downstream partners to be in a position to
monitor and manage their own exposure to weather vari-
ability.
A weather station measures a vast quantity of observa-
tions on a continuous basis or at regular time intervals
(hour, day). Observations include temperatures, precipi-
tations, cloud cover, pressure, sun hours, wind speed and
so on. For each of these categories, minimum, maximum
and average observations are usually also measured. The
selection of weather stations depends on the geograph-
ical spread of the sales for which we test the weather-
sensitivity. To minimize the number of weather stations,
national weather services can be of valuable assistance in
pointing out which weather stations are most representa-
tive for a given geographical area.
The methodology aims to identify which of the avail-
able weather observations have the greatest influence on
sales. The potential effects of weather on sales are specific
to each product or product category, and may change with
the time of the week or the time of the year, the city, the
region or the country in which products are sold. Sales
can change linearly according to the evolution of weather
4

•Aggregation to match
business and geographical
activity
•Computation of continuous
and critical day variable
•Computation of averages
Building weather
variables for analysis
Building the weather-
sensitivity model
Managing Exposure
to Adverse Weather
•Selection of control variables
(promotion, calendar
variables, etc.)
•Determination of the
Weather Index and
Sensitivity Coefficient
•Contribution of weather to
sales
•Probability distributions of
expected losses caused by
adverse weather
•Hedging and Effectiveness
Figure 1: Methodology
variables. They can start or stop selling if weather condi-
tions are below or above specific thresholds, or sell only
if certain conditions occur. In general, the most influen-
tial weather conditions and their effects on sales are not
known a priori. As a result, an extensive range of weather
variables is required to identify the most influential one(s)
and determine the relationship between sales and wea-
ther (see Figure 1). These include “continuous” variables
that measure the accumulation of deviations from aver-
age conditions on various time periods and “critical day”
variables that measure the accumulation of days for which
specific thresholds are observed. Average conditions, also
called normal conditions, are computed using a history of
30 years (Arguez and Vose,2011). The current calculation
period for normal seasonal weather is 1981-2010 (World
Meteorological Organization). Due to climate change, cal-
culating the weather anomaly as the difference between
observation and average may require taking into account
a possible trend. Finally, weather variables are weighted
and aggregated to fit the geographical and temporal sales
spread of the manufacturer under consideration (Maun-
der,1973;Parsons,2001;Dell et al.,2014).
The next step in the methodology is to determine the
most relevant meteorological index (one or more influ-
ential weather variables) and the sensitivity coefficient(s)
that measure changes in sales based on the change in the
meteorological index (Pres,2009). Models to estimate
sales as a function of weather variables and other control
variables can then be built and tested for robustness.
Once the sales estimation model is selected, it is then
possible to simulate the historical distribution of the im-
pact of weather on sales by applying the value of the wea-
ther index corresponding to the weather conditions ob-
served over the past thirty years. The average weather
impact is obtained by applying the average value of the in-
dex. The model also makes it possible to estimate the level
of sales losses based on adverse weather conditions. Each
loss level corresponds to a value of the weather index. The
manufacturer can therefore decide the maximum loss she
tolerates and design the weather index-based instrument
that provides financial compensation for every change in
the index as weather conditions become more adverse.
The price of this instrument is measured by its expected
average payout (also called Burn method). A risk taker
who agrees to supply this instrument to the manufacturer
adds a charge factor to meet its return on equity require-
ments.
The risk taker’s role is to take on a risk on a weather
index, not a risk based on the manufacturer’s sales or pro-
cesses. In our examples, the risk taker will estimate the
probability distribution of the weather index, calculate the
expected payout to the manufacturer when the weather
conditions become adverse, and will provide a price to the
manufacturer for accepting that risk.
5. Intervention to reduce manufacturers’ exposure to
weather variability
In this section, we present in detail the Intervention at
the Original Equipement Manufacturer (OEM) who sells
a car part to the aftermarket. The other two industrial ap-
plications are presented in Appendix A. The manufac-
turer has limited Sell-Out (SO) data from different French
retailers in four sales regions (South West, West, Paris,
South East). We used weather observations from ground
stations and prepared weather variables for analysis. A
sales season extends from October to March. The analysis
of the regional SO data reveals that regions exhibit differ-
ent sales variances.
The application of the methodology leads to the selec-
tion of the Critical Day for temperature as the most influ-
ential variable on sales. The Critical Day (CD) for temper-
ature is either a binary measure that indicates whether a
temperature threshold has been crossed, or an intensity
measure that provides the extent to which the threshold
has been crossed. We obtained the best results with the
binary CD. Technical experts at the OEM confirmed that
what triggers the replacement of the component under
consideration is most likely the fact that the temperature
threshold is triggered. The extent to which the tempera-
ture falls below that threshold should not make a differ-
ence.
We then compute the regional weather variable using i
to index months, jto index the region and kto index the
weather station within each region. We have
CDi,j=X
k
wj,kcdi,j,k,∀j∈Region,∀i∈Month;(1)
where wj,kis the population weight in the area covered by
the weather station kwithin region j, constant over the
different months, cdi,j,kis the number of days in station k
of region jin month iduring which the temperature was
observed to be less than the threshold for that region j.
For every month i, and every region j, we now model
sales to end consumers provided by the marketing consul-
tancy firm as a linear function of the Critical Day index:
S Oi,j=ajCDi,j+bi,j+i,j,(2)
with ajand bi,jas the parameters for the model and i,jas
the residuals of the model. The choice of representing the
different months with different offset parameters but the
5

●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●●
●
●
●
0 2 4 6 8 10 12
500 1000 1500 2000 2500
AUTOBACS.JCRIT..4
Sales of AUTOBACS
October
November
December
January
February
March
Figure 2: Linear Regression line and confidence intervals (<5%) be-
tween sales and the cumulated number of critical days with tempera-
tures of less than -4◦Cfor the Paris region.
same slope is to address the question of the seasonality of
sales by month.
The data for the Paris region are presented in Figure 2.
The other regions exhibit a similar profile. We observe
that several months in which no critical days were regis-
tered stack up to the left of the graph, over the 0mark.
The trend line is clearly heavily influenced by the outliers
that represent the months with a large number of critical
days and high sales. It is this effect which we especially
want to capture. We observe that the slope of the regres-
sion lines fit the data of the different months, justifying
the choice of one single slope parameter for all months
of the season (October to March). Following Cleveland
et al. (1990) and Jiang et al. (2010), the goodness of fit is
assessed (see Table 1 for the parameter a, those of the pa-
rameters bare not presented to conserve space as there is
one for each month). The need for data transformation is
evaluated using quantile plots of the residuals to ensure
that the normal distribution is a good estimate of the ac-
tual residuals (Hafen et al.,2009). Additionally, marginal
residual plots confirm that there are no residual patterns.
Table 1: Parameter table for aof the linear regressions between monthly
sales and number of critical days from October to March for each region.
Region Estimate Standard t-Statistic P-Value
Error
West 964.217 111.937 8.61396 2.68 x10−12
Paris 40.766 4.73523 8.60909 2.73 x10−12
SWest 451.198 77.7822 5.80079 2.90 x10−7
SEast 26.2348 3.45027 7.6037 2.74 x10−9
Table 2 provides a validation of the model over the SI
data. As can be seen, the linear regression is a valid and
parsimonious usage of the model inferred from the sales
data to final users of the automotive part. The model’s
Adjusted R2(77%) is important in view of the structur-
ing of weather index-based financial instrument. In most
countries outside of the United States, climate data is not
free. The more variables in a model, the higher the cost. A
robust weather sensitivity model that requires as few wea-
ther variables as possible should therefore be constructed.
Table 2: Linear regression analysis of the weather model over SI
Estim. Std. Er. t-value p-val. R2Adj. R2
β537676 55834 9.63 .0106* .8446 .7669
α13592 4122 3.297 .081
Residual standard error: 26200 on 2 degrees of freedom
F-statistic: 10.87 on 1 and 2 DF, p-value: .08096
* signif p<0.05
The first case is an illustration of our methodology
when only sparse or incomplete retailer data are available.
The other two examples illustrate the use of our method-
ology when direct access to retailer data is not possible,
and manufacturers have to revert to surveys or panel data
to test their products’ weather-sensitivity. The weather-
sensitivity analysis of sunscreen product sales uses sales
data provided by a marketing panel data company. The
analysis of clothing sales is based on a monthly indus-
try survey conducted by the Institut Français de la Mode
(IFM), the French apparel retailers’ organization. The re-
sults of the application of our method are briefly high-
lighted and discussed in the appendix.
We now proceed to evaluate the exposure of the sales
of the OEM to adverse weather and estimate her Weather
Value-at-Risk.
5.1. Evaluating Weather Value-at-Risk
Estimating weather risk and structuring a weather
index-based financial instrument is essentially about
using historical weather data in the Sell-In weather-
sensitivity model.
In the case of automotive products, Weather-VaR rep-
resents the maximum expected loss caused by an unusu-
ally low number of Critical Days over the winter season
(October-March) with a level of confidence of 95% (Pafka
and Kondor,2001). We estimate Weather-VaR using the
historical value method (Linsmeier and Pearson,2000).
The frequency distribution diagram in Figure 3 indicates
that 95% of the observations are above 2 CDs. We find
that the corresponding maximum expected loss (sales
Weather-VaR, CD =2) is 16.1% of sales with a confidence
level of 95%. In other words, the OEM is exposed to losing
16% of normal sales once every 20 years. Weather-VaR in
the case of sunscreen products and clothing are provided
in Appendix A.
6. Outcome: Mitigating the cost of adverse weather
The whole purpose of our Intervention is to allow man-
ufacturers to control and mitigate the financial conse-
quences of weather risks that managers used to accept
as an ’Act of God’ against which there is nothing to be
done. Obviously, the weather cannot be changed, but
knowing the mathematical relationship that links sales
and weather, makes it possible to change the paradigm
6

5
10
15
20
Critical Days
0.01
0.02
0.03
0.04
0.05
0.06
0.07
Frequency
Weather VaR
Figure 3: Historical frequency distribution of Critical Days. The shaded
area is the risk of a winter having 2 CDs or less (5% cumulated proba-
bility).
from a “cope and avoid“ attitude to an “anticipate and exe-
cute“ mindset (IBM,2015). This is the purpose of weather
index-based financial protection.
In traditional insurance, an indemnity payout is set by
an expert who assesses the financial loss related to the loss
of sales actually recorded. Weather index-based insurance
products work differently. They trigger a payment linked
to a weather variable and not to actual losses incurred.
The advantage of index solutions over traditional insur-
ance is that the compensation is determined mathemati-
cally and is known in advance to all parties.
With index-based protection, covered periods range
from hours to days, seasons, or years. The index can be a
single weather variable or a combination of variables that
represents the weather risk the manufacturer faces. With
significant advances in data processing, modelling, and
forecasting, these products have evolved substantially and
are now cheaper, more effective, and more widely avail-
able to all companies in a wide range of activity sectors
(Hershey and Breslin,2015). Index-based instruments
work like other traditional financial instruments, except
that the index on which they trigger, and on which the
payout is calculated, is a weather index (Barrieu and Scail-
let,2010). They are mostly sold over-the-counter by rein-
surance companies.
Two strategies can be implemented: manufacturers may
purchase a simple hedge to reduce their exposure to wea-
ther risks by using an option that pays out if the weather
index crosses a threshold (strike price); or they may pur-
chase a more complex hedge called a collar, which fore-
goes potential additional gains from favorable weather to
fund a portion of the option price. In the following sec-
tion we present how such strategies work in practice. To
conserve space, we only address the case of the OEM. The
hedging and coordinating strategies for the clothing fol-
low the very same principle mutatis mutandis. The partic-
ular case of the sunscreen products has been dealt in detail
in Bertrand and Brusset (2018).
6.1. Effectiveness of hedging weather risk
The first hedging strategy consists in purchasing an op-
tion so that the manufacturer limits sales losses if the num-
ber of Critical Days (CD) over the season falls below a
certain threshold. On average over the last 30 years, the
average observed number of CDs is 10. The minimum ob-
served value of the index was 2, causing a maximum loss
in sales of 16.4% of average (or normal) sales.
Based on her margin levels, the manufacturer decides to
hedge her risk with a put option that pays out if the num-
ber of CDs over the season is less than 7 CDs. To lower the
cost of hedging, she is prepared to look at an alternative
hedging structure (collar) that still protects her if the in-
dex falls below 7 CDs, but implies that she pays the risk
taker if the index is above 13 CDs.
Table 3 is a reconstruction of the net cash-flows based
on the last 30 years of historical values of the index, under
three scenarios: unhedged (column "Sales"), hedged with
a option and hedged with a collar. The average expected
sales with no hedge is 672 681 units, the maximum loss is
110 078 units or 16.4% of normal sales and the standard
deviation, or sales cash-flow variability, is 11.2%. With an
option, the average expected sales rise to 686 460 units, the
maximum potential loss is reduced by 50% to 8.6% of nor-
mal sales, and the variability of sales cash-flows is reduced
by 21% to 8.9% from 11.2%. With a collar, the reduction
in sales variability is further improved. Variability drops
50% to 5.6% from 11.2% (see Std Dev in Table 3).
Since temperatures cannot be bought, sold, borrowed or
stored in a portfolio, traditional valuation methods do not
apply to calculate the price of weather index-based prod-
ucts (Davis,1998,2001;Dischel,1998;Geman,1999;Cao,
2000;Geman and Leonardi,2005). The value at expiry of
a weather index-based instrument is calculated based on
the estimated value of the weather index at maturity (Jew-
son and Zervos,2005). The most commonly used method
is the Burn method, which consists in calculating the av-
erage value of the financial instrument when all available
historical observations of the weather index are succes-
sively applied to it. The Burn Analysis is in fact the av-
erage payout over the considered historical period. The
Burn (average payout) is 13 779 units in the case of the op-
tion and 1 203 units in the case of the collar. The load fac-
tor added by a potential risk taker willing to supply such
a hedging instrument is either a required rate of return
applied to the maximum payout reduced to the duration
of the period covered or a percentage of the variability of
the payout (Jewson and Brix,2005). Using a required rate
of return of 10%, the cost of capping the loss at 8.6% of
average sales is 3.1% for an option, and 0.8% for a collar
(Table 4).
Weather-sensitivity models, reconstructed cash-flow ta-
bles and prices in the case of sunscreen products and
clothing are detailed in Appendix A. Table 5 is a sum-
mary table to measure hedging effectiveness. In all three
cases, expected average sales are marginally improved ex-
7

Table 3: Reconstructed net cash flows without and with hedging in place
Weather Index Sales Impact Option Net Cash-Flow Collar Net Cash-Flow
1985 23 851 094 178 413 - 851 094 - 125 712 725 382
1986 21 826 085 153 403 - 826 085 - 79 065 747 020
1987 20 804 868 132 186 - 804 868 - 67 715 737 152
1988 3 583 005 - 89 676 31 914 614 919 31 914 614 919
1989 4 593 716 - 78 965 61 069 654 785 61 069 654 785
1990 8 645 433 - 27 248 - 645 433 - 645 433
1991 13 713 910 41 229 - 713 910 - 11 648 702 262
1992 10 672 250 - 431 - 672 250 - 672 250
1993 8 642 905 - 29 776 - 642 905 - 642 905
1994 7 627 927 - 44 754 24 615 652 542 24 615 652 542
1995 3 573 926 - 98 755 63 638 637 564 63 638 637 564
1996 9 655 546 - 17 135 - 655 546 - 655 546
1997 14 728 494 55 813 - 728 494 - 728 494
1998 7 626 146 - 46 535 11 771 637 917 11 771 637 917
1999 9 657 938 - 14 743 - 657 938 - 657 938
2000 5 607 743 - 64 938 34 021 641 764 34 021 641 764
2001 2 562 604 - 110 078 75 803 638 406 75 803 638 406
2002 11 691 021 18 340 - 691 021 - 691 021
2003 12 696 227 23 545 - 696 227 - 696 227
2004 7 628 253 - 44 428 4 173 632 426 4 173 632 426
2005 11 683 708 11 027 - 683 708 - 683 708
2006 14 729 459 56 778 - 729 459 - 11 295 718 164
2007 3 582 067 - 90 614 50 481 632 548 50 481 632 548
2008 9 663 565 - 9 116 - 663 565 - 663 565
2009 12 696 811 24 130 - 696 811 - 696 811
2010 18 776 990 104 309 - 776 990 - 65 038 711 953
2011 13 710 933 38 252 - 710 933 - 3 412 707 522
2012 14 723 125 50 444 - 723 125 - 13 402 709 724
2013 9 654 812 - 17 869 - 654 812 - 654 812
2014 2 569 875 - 102 806 55 890 625 766 55 890 625 766
Average 10 672 681 - 13 779 686 460 1 203 673 884
Max Loss 2 -16.4% -8.6% -8.6%
Std Dev 56% 11.2% 8.9% 5.6%
Table 4: Hedging effectiveness in the case of the OEM
No Hedge Option Collar
Max. loss due to adverse weather (% of average sales) -16.4% -8.6% -8.6%
Standard Deviation (% of margin) 11.2% 8.9% 5.6%
Cost of hedging (% of sales) - -3.1% -0.8%
8

cluding hedging costs when hedged against weather vari-
ability (between +0.1% and +3%). In all three cases also,
cash-flow uncertainty drops when hedging against wea-
ther variability. With a cover that cuts the maximum loss
by approximately 50% versus the unhedged situation, the
variability drops by 16% to 21% with an option, and by
27% to 50% with a collar.
6.2. Protecting the downstream retailers
The financial compensation received by the manufac-
turer in case of adverse weather can be kept for her sole
benefit. It can also be used to coordinate downstream
partners.
Bosch, for instance, designed an offer for its retailers of
diesel engine heating plugs2. This offer was in two steps.
In the first step, prior to the beginning of the high sea-
son, the retailer was required to specify an annual order
commitment as a percentage of the previous year’s or-
dered volume. An increasing volume rebate scheme was
offered, proportional to this commitment. In the second
step, the retailer had the ability to reduce his commit-
ment by 25% and keep the initial rebate if the observed
average temperature over a 3-month period (from Jan-
uary to March of the year in the retailer’s region) was
higher than the 30-year average. The benefit to the retailer
was a smaller inventory in case of adverse weather con-
ditions. The manufacturer might appear initially to lose
some sales, but the benefit was an increased stock at the
retailer’s in the high selling season. In a mild winter, the
manufacturer thus also subsidized sales to the retailer by
providing unjustified volume discounts. Bosch did not
communicate on whether or not they hedged with wea-
ther index-based insurance products.
In our OEM case study however, weather index-based
cover was used. The OEM offered retailers an incentive
program against a commitment to pre-order a minimum
quantity of products (125% of the previous year). In case
of adverse weather in winter, retailers would receive a fi-
nancial compensation. The program was launched fol-
lowing an adverse winter season, as retailers were reluc-
tant to pre-order. As a result, the OEM was able to enroll
most of her existing retailers, gain new ones and increase
sales year-on-year by more than 20%. Besides, the OEM
was perceived as innovative and supportive of her retail
partners.
6.3. Other coordination mechanisms
Several other mechanisms could be chosen to help the
manufacturer and the first tier of downstream partners
overcome the negative impact of weather. Cachon (2004)
presents several newsvendor type mechanisms according
to the moment when orders are placed with regard to the
2The offer can still be seen as of March 10th, 2018 at Bosch weather
protection 2014.
selling season. The manufacturer decreases the whole-
sale price so that the retailer orders more and both the
manufacturer and retailer use the distribution function
of the relevant weather variable to assess the inventory
level. Tsay (2001) provides a comprehensive analysis of
contracts with markdown provisions. Another possible
mechanism involves returning excess inventory at the end
of the season (Pasternack,1985;Padmanabhan and Png,
1995).
Chen and Yano (2010) propose a weather-linked rebate
to improve the retailer’s expected profit in a newsvendor
context which falls into the end-of-season concession or
markdowns category. This coordination mechanism en-
hances sales in favorable conditions and protects the first
tier of downstream partners under unfavorable weather
conditions in such a way that they keep enough inventory
to avoid losing sales. It is shown that the manufacturer
can even completely hedge her risk. This mechanism is
different from the one in Taylor (2002) as it refers to pur-
chases by the retailer, not his sales, i.e., it does not depend
upon information about sales to the end consumer.
As noted in Chen and Yano (2010), this type of rebate
function is flexible. It ensures that the retailer orders the
centralized supply chain optimal quantity. Note that con-
trary to Cachon (2004), this mechanism can accommodate
any distribution function, so it does not require that the
weather variable distribution function have an Increas-
ing Generalized Failure Rate. Among the three exam-
ples presented here, this mechanism can accommodate
the Gamma distribution which models rainfall and tem-
perature anomalies for which the best fitting distribution
is a Normal Inverse Gaussian (NIG) distribution (Ahčan,
2012).
7. Conclusion
In the framework of Design Science Research, we
present new generic knowledge to support organizational
improvement actions (van Aken et al.,2016). We apply
CIMO-logic (Denyer et al.,2008). The problem-in-context
is the inability of manufacturers to mitigate the effects of
adverse weather on sales. Weather directly affects the ac-
tivity and performance of 70% of business sectors world-
wide (Hanley,1999;Dutton,2002;Larsen,2006). The con-
sequences of adverse weather go up the supply chain and
indirectly impact manufacturers.
Several such manufacturers called upon us to mitigate
sales losses caused by adverse weather (the Intervention).
The approach adopted goes from the particular to the gen-
eral, from solving each case individually to offering other
practitioners and scholars a general mechanism.
The Intervention provides weather sensitive manufac-
turers with an innovative and effective Mechanism to un-
derstand and mitigate the impact of adverse weather on
sales with bespoke weather index-based financial cover.
The cases analyzed are of three manufacturers who expe-
rienced losses caused by adverse weather. They represent
9

Table 5: Hedging effectiveness summary for all three cases under study
No Hedge Option % Change Collar % Change
Expected Sales
OEM 672 681 686 460 2.0% 673 884 0.2%
Sunscreen 3 185 431 3 202 960 0.6% 3 206 737 0.7%
Clothing 10 005 847 10 038 370 0.3% 10 018 141 0.1%
Maximum Loss
OEM -16.4% -8.6% -48% -8.6% -48%
Sunscreen -9.7% -3.9% -60% -3.9% -60%
Clothing -7.2% -2.5% -65% -2.5% -65%
Standard Deviation
OEM 11.2% 8.9% -21% 5.6% -50%
Sunscreen 4.5% 3.8% -16% 2.8% -39%
Clothing 2.7% 2.3% -16% 2.0% -27%
Cost of Hedging
OEM - 3.1% NA 0.8% NA
Sunscreen - 0.7% NA 0.1% NA
Clothing - 0.3% NA 0.1% NA
a cross-section of the types of weather risks to which man-
ufacturers are exposed. In each case, the manufacturer
sells through different channels and has limited access to
end-consumer sales data.
In all three cases, it was possible to identify the most
significant weather variables, determine how these vari-
ables affect manufacturers’ sales, and construct efficient
index-based protection that compensates manufacturers
for sales losses caused by adverse weather. With this De-
sign proposition, manufacturers were able to significantly
reduce their exposure to adverse weather, both in terms
of the size of the potential losses and the uncertainty of
sales cash-flows (the Outcome). Moreover, sharing the
benefits of the proposed Intervention with retailers leads
to enhanced coordination and customer service.
Our research breaks new grounds for supply chain and
operations managers to apply well-accepted instruments
such as futures, swaps, or options traditionally used by
financial managers to cover the effects of the climate in
the energy and agricultural sectors. On the other hand,
it also opens up new opportunities for traditional risk tak-
ers such as insurance companies as they can now diversify
their weather risk portfolios away from the energy and
agriculture sectors.
Understanding the link between weather and sales
yields considerable value to manufacturers. They can bet-
ter measure ex post the contribution of the weather to
sales. Marketing, financial or sales reports benefit from
increased objectivity, and managerial effectiveness clearly
emerges as a result.
From a more theoretical perspective, the discussion in
the literature therefore shifts from testing how weather
observations are correlated to sales (see Caro and de Al-
béniz,2009;Caliskan Demirag,2013;Bahng and Kincade,
2012, to name but the latest) to providing an actionable
weather-sensitivity model that allows scholars and man-
agers to understand the disruptive effects of weather vari-
ability.
Supply chain managers can use the result from these
hedging strategies to coordinate the downstream partners
into maintaining inventory at sufficient levels so as to be in
a position to take advantage of favorable weather too. In
this way inventory variations at two levels of the supply
chain might be lessened.Hedging weather risks can also
be used to enhance customer service levels by offering fi-
nancial compensation to retailers in exchange of increased
orders or pre-orders.
This research contributes to both the Science Design
and the Operations Management literature by providing
a generic mechanism to explain the indirect effects of wea-
ther on manufacturers’ sales and more generally through
supply chains. This is of utmost importance as weather
variability is expected to rise with climate change over the
coming decades (IPCC,2014). Businesses in competitive
environments will increasingly suffer, therefore, growing
interest and further application of this mechanism to other
academic and managerial areas is to be expected.
References
References
Agnew, M. D., Thornes, J. E., 1995. The weather sensitivity of the UK
food retail and distribution industry. Meteorological Applications 2,
137–147. 2 citations, pages 1and 3
Ahčan, A., 2012. Statistical analysis of model risk concerning tempera-
ture residuals and its impact on pricing weather derivatives. Insur-
ance: Mathematics and Economics 50. Cited on page 9
Albert, P. S., Rosen, L. N., Alexander, J. R., Rosenthal, N. E., 1991. Effect
of daily variation in weather and sleep on seasonal affective disorder.
Psychiatry Research 36, 51–63. Cited on page 2
Arguez, A., Vose, R., 2011. The definition of the standard WMO climate
normal the key to deriving alternative climate mormals. Bulletin of
American Meteorological Society 92, 699–704. Cited on page 5
Auffhammer, M., Hsiang, S., Schlenker, W., Sobel, A., 2013. Global cli-
mate models: A user guide for economists. Review of Environmental
Economics and Policy 7 (2), 181–198. 2 citations, pages 3and 4
10

Bahng, Y., Kincade, D., 2012. The relationship between temperature and
sales. International Journal of Retail and Distribution Management
40 (6), 410–426. 2 citations, pages 3and 10
Barrieu, P., Scaillet, O., 2010. Uncertainty and environmental decision
making. Springer, Heidelberg, Germany, Ch. A primer on weather
derivatives, pp. 155–175. 2 citations, pages 3and 7
Bertrand, J.L., Brusset, X., 2018. Managing the financial con-
sequences of weather variability. Journal of Asset Manage-
ment 19, 301–315. .
2 citations, pages 3and 7
Bertrand, J.-L., Brusset, X., Fortin, M., July 2015. Assessing and hedging
the cost of unseasonal weather: case of the apparel sector. European
Journal of Operational Research 244 (1), 261–276. Cited on page 3
Boissonade, A., Heitkemper, L., Whitehead, D., 2002. Climate risk
and the weather market: financial risk management with weather
hedges. Risk, Ch. Weather data: cleaning and enhancement, pp. 73–
93. Cited on page 4
Bray, R. L., Mendelson, H., 2012. Information transmission and the Bull-
whip Effect: An empirical investigation. Management Science 58 (5),
860–875. Cited on page 3
Cachon, G., 2004. The allocation of inventory risk in a supply chain:
Push, pull, and advance-purchase discount contracts. Management
Science 50 (2), 222–238. Cited on page 9
Caliskan Demirag, O., 2013. Performance of weather-conditional re-
bates under different risk preferences. Omega 41, 1053–1067.
Cited on page 10
Cao, M., May 2000. Pricing the weather. Risk, 67–70.
URL Cited on page 7
Caro, F., de Albéniz, V. M., 2009. Operations Management Models with
Consumer-Driven Demand. Springer US, Ch. The effect of assort-
ment rotation on consumer choice and its impact on competition.
Cited on page 10
Chen, F., 2004. Information sharing and supply chain coordination. In:
de Kok, T., Graves, S. (Eds.), Handbooks in Operations Research and
Management Science: Supply chain Management. Vol. 11. Elsevier,
Ch. 7, pp. 341–421. Cited on page 3
Chen, F. Y., Yano, C. A., 2010. Improving supply chain performance and
managing risk under weather-related demand uncertainty. Manage-
ment Science 56 (8), 1380–1397. 3 citations, pages 1,3, and 9
Cleveland, R. B., Cleveland, W. S., McRae, J., Terpenning, I., 1990. STL: a
seasonal-trend decomposition procedure based on Loess. Journal of
Official Statistics 6, 3–73. Cited on page 6
Connolly, M., 2008. Here comes the rain again: weather and the intertem-
poral substitution of leisure. Journal of Labor Economics 26 (1), 73–
100. Cited on page 3
Cunningham, M. R., 1979. Weather, mood, and helping behavior: Quasi
experiments with the sunshine samaritan. Journal of Personality and
Social Psychology 37, 1947–1956. Cited on page 2
Davis, M., 1998. Mathematics of Derivatives Securities. Cambridge
Univeristy Press, Ch. Option prices in incomplete markets.
Cited on page 7
Davis, M., 2001. Pricing weather derivatives by marginal value. Quanti-
tative Finance 1, 1–4, Review article. Cited on page 7
Dell, M., Jones, B. F., Olken, B. A., September 2014. What do we learn
from the weather? the new climate-economy literature. Journal of
Economic Literature 52 (3), 740–798.
URL
3 citations, pages 3,4, and 5
Denyer, D., Tranfield, D., van Aken, J., 2008. Developing design proposi-
tion through research synthesis. Organization Studies (29), 249–269.
2 citations, pages 1and 9
Deschênes, O., Greenstone, M., 2011. Climate change, mortality, and
adaptation: Evidence from annual fluctuations in weather in the
US. American Economic Journal: Applied Economics 3 (4), 152–185.
Cited on page 3
Dischel, R., 1998. Black-Scholes won’t do. Weather Risk Special Report,
energy and Power Risk Management, 8–9. Cited on page 7
Dischel, R., 2002. Climate Risk and the Weather Market: Financial Risk
Management with Weather Hedges. Risk Books, London, Ch. Intro-
duction to the weather market: Dawn to mid-morning, pp. 25–41.
Cited on page 4
Disney, S. M., Childerhouse, P., Naim, M. M., 1997. The development of
supply chain structures in the automotive aftermarket sector. In: Pro-
ceedings of the 1997 Logistics Research Network Conference. Institute
of Logistics. Cited on page 3
Dutton, J., 2002. Opportunities and priorities in a new era for weather
and climate services. Bulletin of American Meteorological Society
83 (9), 1303–1311. Cited on page 9
Forrester, J. W., 1965. Industrial Dynamics. MIT Press, MA, USA.
Cited on page 3
Gagne, J., May 1997. Fair-weather trends. American Demograph-
ics/Marketing Tools. Cited on page 3
Gavirneni, S., Kapuscinski, R., Tayur, S., 1999. Value of information
in capacitated supply chains. Management Science 45 (1), 16–24.
Cited on page 3
Geman, H., 1999. Insurance and weather derivatives: From exotic
options to exotic underlyings, Working Paper, Dauphine Recherche
en Management.
URL
Cited on page 7
Geman, H., Leonardi, M.-P., 2005. Alternative approaches to wea-
ther derivatives pricing. Managerial Finance 31 (6), 46–72.
Cited on page 7
Goldstein, K. M., 1972. Weather, mood, and internal-external control.
Perceptual and Motor Skills 35 (3), 786–786. Cited on page 2
Granger, C. W. J., 1978. Seasonality: causality, interpretation, and im-
plications, in seasonal analysis of economic time series. Economic re-
search report, U.S. Department ofCommerce, ER-1, Arnold Zeller, ed.
Cited on page 1
Hafen, R., Anderson, D., Cleveland, W., Maciejewski, R., Ebert, D.,
Abusalah, A., Yakout, M., Ouzzani, M., Grannis, S., 2009. Syn-
dromic surveillance: STL for modeling,visualizing, and monitoring
disease counts. BMC Medical Informatics and Decision Making 9, 21.
Cited on page 6
Hanley, H., 1999. Hedging the force of nature. Risk Professional, 21–25.
Cited on page 9
Harland, C. M., March 1996. Supply chain management: relationships,
chains and networks. British Journal of Management 7 (SI), 63–80.
Cited on page 3
Hershey, L., Breslin, E., 2015. Meteo protect: Empowering the bottom
line wth weather modeling. Report, SAP. Cited on page 7
Howarth, E., Hoffman, M. S., 1984. A multi-dimensional approach to the
relationship between mood and weather. British Journal of Psychol-
ogy 75 (1), 05–23. Cited on page 2
IBM, December 2015. Weather mmean business. White paper, IBM
Analytics.
URL
Cited on page 7
IPCC, 2014. Climate change 2014: Impacts, adaptation and vulnerability.
Report, Intergovernmental Panel on Climate Change.
URL 2 citations, pages 1and 10
Jewson, S., Brix, A., 2005. Weather derivative valuation: the meteorolog-
ical, statistical, financial and mathematical foundations. Cambridge
University Press, p. 370. Cited on page 7
Jewson, S., Whitehead, D., 2001. In praise of climate data. Environmental
Finance 3 (2), 22–23. Cited on page 4
Jewson, S., Zervos, M., May 2005. No-arbitrage pricing of weather
derivatives in the presence of a liquid swap market, Working Paper.
URL
Cited on page 7
Jiang, B., Liang, S., Wang, J., Xiao, Z., 2010. Modeling MODIS LAI time
series using three statistical methods. Remote Sensing of the Environ-
ment 114, 1432–1444. Cited on page 6
Jorgenson, D. O., 1981. Perceived causal influences of weather: Rating
the weather’s influence on affective states and behaviors. Environ-
ment and Behavior 13, 239–256. Cited on page 2
Kirk, B., 2005. Better business in any weather. ICSC Research Review
12 (2), 28–34. Cited on page 1
Kouvelis, P., Chambers, C., Wang, H., jan 2006. Supply chain manage-
ment research and production and operations management: Review,
trends, and opportunities. Production and Operations Management
15 (3), 449–469. Cited on page 3
11

Larsen, P. H., 2006. Estimating the sensitivity of U.S. economic sectors to
weather, Working Paper, Cornell University. Cited on page 9
Lazo, J. K., Lawson, M., Larsen, P. H., Waidmann, D. M., 2011. U.S. eco-
nomic sensitivity to weather variability. Bulletin of American Meteo-
rological Society 92, 709–720. 2 citations, pages 1and 3
Lee, H. L., Padmanabhan, V., Whang, S., 1997a. The bullwhip ef-
fect in supply chains. Sloane Management Review 39 (3), 93–102.
Cited on page 3
Lee, H. L., Padmanabhan, V., Whang, S., 1997b. Information distortion in
a supply chain: the bullwhip effect. Management Science 43, 543–558.
Cited on page 3
Lee, H. L., So, K. C., Tang, C. S., 2000. The value of information shar-
ing in a two-level supply chain. Management Science 46 (5), 626–643.
Cited on page 3
Linden, F., 1962. Consumer markets: merchandising weather. The Con-
ference Board Business Record 19 (6), 15–16. Cited on page 3
Linsmeier, T. J., Pearson, N. D., Mar-Apr 2000. Value at risk. Financial
Analysts Journal 56 (2), 47–67. 2 citations, pages 3and 6
Lucas, R. E., Lawless, N. M., 2013. Does life seem better on a sunny day?
examining the association between daily weather conditions and life
satisfaction judgments. Journal of Personality and Social Psychology.
Cited on page 3
Marteau, D., Carle, J., Fourneaux, S., Holz, R., Moreno, M., 2004. La
Gestion du Risque Climatique. Edition Economica, Ch. La mesure de
l’exposition au risque climatique, p. 35. Cited on page 15
Mason-Jones, R., Towill, D. R., 1999. Using the information decoupling
point to improve supply chain performance. The International Journal
of Logistics Management 10 (2), 13–26. Cited on page 3
Maunder, W. J., 1973. Weekly weather and economic activities on a na-
tional scale: an example using united states retail trade data. Weather
28 (1), 2–19. Cited on page 5
Meixell, M. J., Shaw, N. C., Tuggle, F. D., 2008. A methodology for as-
sessing the value of knowledge in a service parts supply chain. IEEE
Transactionson Systems, Man, and Cybernetics – Part C: Applications
and Reviews 38 (3), 446–460. Cited on page 2
Merna, T., Al-Thani, F. F., 2011. Corporate risk management. John Wiley
& Sons. Cited on page 3
Murray, K., Di Muro, F., Finn, A., Popkowski Leszcyc, P., 2010. The effect
of weather on consumer spending. Journal of Retailing and Consumer
Services 17, 512–520. Cited on page 3
Nieves-Rodriguez, E. B., Cao-Alvira, J., Pérez, M., 2015. Ideas in Market-
ing: Finding the New and Polishing the Old. Springer International
Publishing, Ch. The Influence of Special Occasions on the Retail Sales
of Women’s Apparel, pp. 213–221. Cited on page 3
Padmanabhan, V., Png, P. L., 1995. Returns policies: making
money by making good. Sloan Management Review 37 (1), 65–72.
2 citations, pages 3and 9
Pafka, S., Kondor, I., 2001. Evaluating the riskMetrics methodology in
measuring volatility and Value-at-Risk in financial markets. Phys-
ica A: Statistical Mechanics and its Applications 299 (1), 305–310.
Cited on page 6
Parsons, A. G., 2001. The association between daily weather and
daily shopping patterns. Australasian Marketing Journal 9, 78–84.
4 citations, pages 1,2,3, and 5
Pasternack, B., 1985. Optimal pricing and returns policies for perishable
commodities. Marketing Science 4 (2), 166–176. Cited on page 9
Pres, J., 2009. Measuring non-catastrophic weather risks for businesses.
The Geneva Papers on Risk and Insurance Issues and Practice 34 (3),
425–439. Cited on page 5
Prettenthaler, F., Köberl, J., Bird, D. N., 2016. Weather Value at Risk:
A uniform approach to describe and compare sectoral income risks
from climate change. Science of the Total Environment 543, 1010–1018.
Cited on page 3
Quayle, R. G., Diaz, H. F., 1980. Heating degree day data applied to resi-
dential heating energy consumption. Journal of Applied Meteorology
19, 241–246. Cited on page 4
Sanders, J. L., Brizzolara, M. S., 1982. Relationships between weather and
mood. Journal of General Psychology 107, 155–156. Cited on page 2
Schneider, F. W., Lesko, W. A., Garrett, W. A., 1980. Helping behaviour in
hot, comfortable, and cold temperatures: A field study. Environment
& Behaviour 12, 231–240. Cited on page 2
Skjoett-Larsen, T., Thernøe, C., Andersen, C., 2003. Supply chain collabo-
ration, theoretical perspectives and empirical evidence. International
Journal of Physical Distribution & Logistics Management 33 (6), 531–
549. Cited on page 3
Starr-McCluer, M., 2000. The effect of weather on retail sales. Tech. rep.,
Federal Reserve Board of Governors. 2 citations, pages 1and 3
Steele, A. T., 1951. Weather’s effect on sales of a department store. Journal
of Marketing 15, 436–443. Cited on page 1
Steinker, S., Hoberg, K., Thonemann, U. W., 2017. The value of weather
information for e-commerce operations. Production and Operations
Management (xx), n/a–n/a.
URL
2 citations, pages 1and 3
Stone, J., 2012. The impact of supply chain performance measurement
systems on dynamic behaviour in supply chains. Ph.D. thesis, Aston
University, Birmingham.
URL Cited on page 3
Taylor, T. A., 2002. Supply chain coordination under channel rebates
with sales effort effects. Management Science 48 (8), 992–1007.
Cited on page 9
Toeglhofer, C., Mestel, R., Prettenthaler, F., 2012. Weather value at risk:
on the measurment of non-catastrophic weather risk. Weather, Cli-
mate, and Society 4 (3), 190–199. Cited on page 3
Tsay, A., 2001. Managing retail channel overstock: Markdown
money and returns policies. Journal of Retailing 77 (4), 457–492.
2 citations, pages 3and 9
van Aken, J., Chandrasekaran, A., Halman, J., 2016. Conducting and
publishing design science research inaugural essay of the design sci-
ence department of the journal of operations management. Journal of
Operations Management, 1–8. 2 citations, pages 1and 9
WeatherUnlocked, August 21st 2014. Weather and ecommerce: How
weather impacts retail website traffic and online sales. Report.
URL
Cited on page 1
WMO, 2013. WMO statement on the status of the global climate in 2012.
Report 1108, World Meteorological Organization. Cited on page 1
12

Table A.6: Linear regression using both difference in temperature and
absolute rainfall to explain abnormal sales: R2=.3977, Adjusted R2=
.3785
Variable Estimate Stand Error t-Statistic P-Value
offset δ373 740 130 099 2.87 .0055
temp. β105 501 32 173 3.28 .0017
rain. α-7 658 2 426 -3.15 .0024
Appendix A. Interventions in the cases of a sunscreen
manufacturer and a clothing manufac-
turer
Sunscreen products
Sales data of sunscreen products are cumulated to in-
clude four weeks of sales for all retail stores in France . The
normal sales are then computed as the mean of historical
sales observed in each week of the year. Once abnormal
sales are computed by difference between average sales
and observed ones, correlations with weather categories
are used to sort weather variables for analysis. As ex-
pected, differences in temperatures (defined as the differ-
ence between the 30-year average and the observed value)
are the most influential variables. This effect is most im-
portant in periods with the highest sales results, that is,
between weeks 15 and 37 (summer weeks), representing
approximately 80% of annual sales over the eleven-year
period under review. The precipitation category also has
a material influence on sales: the larger the cumulative
rainfall observed during a four-week period, the lower the
sales. The weather index is therefore made up of both
temperature and precipitation variables. A parsimonious
model is
Sw−Nw=αRw+β∆θw+δ+, (A.1)
with the reported sales labeled Swin the 4-week period w.
For the same period, the normalized sales are labeled Nw.
Rwis the cumulated rainfall above a threshold; ∆θwis the
difference in temperature observed in the period w;α,β,
are the linear regression parameters, δis the offset in the
linear model, and is the error term (see Table A.6 for the
results).
Net cash-flows with and without hedging instruments
reconstructed using temperatures and precipitations ob-
served between 1984 and 2013 are displayed in Table A.7.
The observed temperatures and rainfall for four-week
rolling periods (566 tuples of data) yield a historical se-
ries of 566 theoretical differences from normalized sales as
registered over the period from 2003 to 2013 (evaluated as
mentioned above). We then evaluate the frequency with
which each difference occurs. When using the weather-
sensitivity model, this occurs if the temperature over a
four-week period is 2.58 ◦Cbelow average and rain at its
5% maximum level (i.e., over 65mm). In this case, sales
loss is 554 526 units in the four-week period. If this hap-
pens in week 27 (four-week period covering June), where
sales normally reach 2 661 million units, the loss reaches
20.8% of average sales. If we take the month of August,
given normal sales of 2 847 million units, the loss is 19.5%
of average sales. Over the entire season however, using
the historical distribution, Weather VaR is 9.7% of average
sales.
The hedging effectiveness of various types of options
are briefly presented in Table A.8. An option which caps
the losses to 3.9% of average sales (versus 9.7% for the un-
hedged scenario) reduces the average uncertainty by 16%
to 3.8 from 4.5%. A collar that also caps the loss at 3.9%
reduces the sales cash-flow variability by 39% to 2.8%.
Clothing sales
The IFM has been collecting volume sales figures from a
panel of thousands of textile and clothing retailers across
France since January 2000. Panel members range from in-
dependent multi-brand clothing stores to specialized sin-
gle brand chain stores and department stores. The data
are available by garment type for women, men and chil-
dren and by distribution channel (i.e. independent stores,
specialized chains, and department stores). We deter-
mine normal sales using a Loess Seasonal Trend Decom-
position (STL). This is required because clothing sales in
France show a general increase in volume between 2000
and 2007 followed by a decrease between 2007 and 2013.
STL is a filtering method for decomposing a time series Yt
into trend Tt, seasonal Stand remainder Rtcomponents.
Yt=Tt+St+Rt(A.2)
The STL method requires the user to evaluate the degrees
of variation in the trend and seasonal components of time-
series so as to produce robust estimates that are not dis-
torted by transient outliers. Once normal sales are esti-
mated, we can determine deviations from normal sales,
which we calculate as the difference between observed
and normal sales. We want to quantify the impact of
1◦Canomaly versus average temperature on clothing sales
anomalies. Since many time-series exhibit some trend, a
1◦Canomaly is unlikely to have the same impact on sales
in 2000 and in 2013. To circumvent this issue, we used a
“relative impact” Ztdefined as :
Zt=Rt
Tt+St
.(A.3)
Ztis the relative distance to the average sales index value
(seasonal + trend value).
We demonstrate that temperature anomalies are lin-
early correlated to clothing sales anomalies. For each sea-
son (spring, summer, fall, and winter), each product cat-
egory, and each distribution channel we construct a lin-
ear model to quantify the impact of monthly temperature
anomalies on monthly clothing sales anomalies. We as-
sume that within a season, the relative impact of 1◦Cfor
each month of the same season is the same. The linear
regression model is as follows :
Zm,y=a+bT0
m,y+, (A.4)
13

Table A.7: Reconstructed net cash flows without and with hedging in place
Weather No Hedge Option Net Cash-Flow Collar Net Cash-Flow
1984 3 092 013 852 3 092 865 852 3 093 717
1985 3 062 191 30 674 3 092 865 30 674 3 123 539
1986 3 106 555 - 3 106 555 - 3 106 555
1987 3 039 690 53 175 3 092 865 53 175 3 146 040
1988 3 075 816 17 049 3 092 865 17 049 3 109 914
1989 3 329 408 - 3 329 408 - 3 329 408
1990 3 352 006 - 3 352 006 - 9 141 3 342 865
1991 3 176 027 - 3 176 027 - 3 176 027
1992 3 260 303 - 3 260 303 - 3 260 303
1993 3 130 536 - 3 130 536 - 3 130 536
1994 3 439 377 - 3 439 377 - 96 512 3 342 865
1995 3 390 745 - 3 390 745 - 47 880 3 342 865
1996 3 166 717 - 3 166 717 - 3 166 717
1997 3 138 679 - 3 138 679 - 3 138 679
1998 3 154 021 - 3 154 021 - 3 154 021
1999 3 305 032 - 3 305 032 - 3 305 032
2000 3 093 835 - 3 093 835 - 3 093 835
2001 3 192 955 - 3 192 955 - 3 192 955
2002 3 133 501 - 3 133 501 - 3 133 501
2003 3 524 187 - 3 524 187 - 181 322 3 342 865
2004 3 156 017 - 3 156 017 - 3 156 017
2005 3 295 359 - 3 295 359 - 3 295 359
2006 3 420 583 - 3 420 583 - 77 718 3 342 865
2007 2 911 794 181 071 3 092 865 181 071 3 273 936
2008 3 074 985 17 880 3 092 865 17 880 3 110 745
2009 3 196 515 - 3 196 515 - 3 196 515
2010 3 205 692 - 3 205 692 - 3 205 692
2011 2 908 071 184 794 3 092 865 184 794 3 277 659
2012 3 052 485 40 380 3 092 865 40 380 3 133 245
2013 3 177 840 - 3 177 840 - 3 177 840
Average 3 185 431 3 202 960 3 206 737
Max Loss -9.7% -3.9% -3.9%
Std Dev (%) 4.5% 3.8% 2.8%
Table A.8: Hedging Effectiveness
No Hedge Option Collar
Max. loss due to adverse weather (% of average sales) 9.7% 3.9% 3.9%
Standard Deviation (% of margin) 4.5% 3.8% 2.8%
Cost of hedging (% of sales) - 0.7% 0.1%
14

where Zm,yrepresents the relative clothing sales anomaly
of the month mand the year y,T0the absolute tempera-
ture anomaly in the same period, aand bparameters to
be estimated, and a centered normally distributed vari-
able (error term).
In summer and winter, the correlation between clothing
sales and weather was not statistically significant. This is
consistent with Marteau et al. (2004) who found that the
most significant correlation factors between clothing and
temperature were observed in March, April, and May on
one side, and in September, October, and November, on
the other. In all distribution channels, a positive tempera-
ture anomaly has a positive impact on sales in spring and
a negative impact on sales in fall (see Table A.9).
When we analyze the results by market segment, sales
are always positively affected by warmer-than-usual tem-
peratures in spring and negatively affected in fall. Again,
all models are significant at the 1% level (Table A.10).
We consider the case of a manufacturer of children’s
clothing seeking to hedge the spring season sales. Ta-
ble A.11 and Table A.12 show that the no-hedge situation
leads to expected average sales of 10 005 847 units, with a
maximum loss caused by cold temperatures of 7.2%. An
option that reduces the loss to 2.5% reduces sales cash-
flow variability by 16% to 2.3%. A collar that also reduces
the loss to 2.5% reduces sales cash-flow variability by 27%
to 2.0%.
15

Table A.9: Temperature sensitivity per distribution channel
Spring Summer Fall Winter
%/◦CR2%/◦CR2%/◦CR2%/◦CR2
Independent stores 2.08*** .34 -1.14** .09 -1.84*** .32 - -
Department stores 1.49*** .16 -1.49* .06 -1.52*** .25 - -
Superstores 2.66*** .31 - - -3.51*** .58 - -
Catalogue - - -1.77** .1 -1.61*** .15 - -
Supermarkets 2.51*** .52 - - -2.03*** .48 - -
Retail networks 3.26*** 0.4 -1.15** .07 -2.59*** .45 - -
Specialized stores 2.41*** .48 - - -1.75*** .39 - -
clothing - all channels 2.21*** .52 - - -1.88*** .58 - -
*** significant at the 1% level (p<0.01) ; ** : at the 5% level (p<0.05);
* : significant at the 10% level (p<0.1); - : not significant
Table A.10: Temperature sensitivity per garment type
Spring Summer Fall Winter
%/◦CR2%/◦CR2%/◦CR2%/◦CR2
Men Ready to Wear 1.47*** .33 -0.92* .04 -2.3*** .57 - -
Men Small Garments 2.28*** .37 - - -2.02*** .44 - -
Men Underwear 1.15*** .25 - - -1.24*** .3 -.71** .07
Women Ready to Wear 2.54*** .45 - - -1.54*** .31 - -
Women Small Garments 2.44*** .45 - - -1.62*** .33 - -
Women Underwear 1.18*** .2 - - -1*** .22 - -
Kids 3.07*** .48 -2.06*** .23 -3.15*** .56 - -
*** significant at the 1% level (p<0.01) ; ** : at the 5% level (p<0.05);
* : significant at the 10% level (p<0.1); - : not significant
16

Table A.11: Reconstructed net cash flows without and with hedging in place
Weather No Hedge Option Net Cash-Flow Collar Net Cash-Flow
1983 9 790 486 - 9 790 486 - 9 790 486
1984 9 593 094 156 906 9 750 000 156 906 9 750 000
1985 9 813 280 - 9 813 280 - 9 813 280
1986 9 677 733 72 267 9 750 000 72 267 9 750 000
1987 9 759 832 - 9 759 832 - 9 759 832
1988 10 090 447 - 10 090 447 - 10 090 447
1989 10 374 907 - 10 374 907 - 124 907 10 250 000
1990 10 336 512 - 10 336 512 - 86 512 10 250 000
1991 9 968 816 - 9 968 816 - 9 968 816
1992 10 263 204 - 10 263 204 - 13 204 10 250 000
1993 10 183 306 - 10 183 306 - 10 183 306
1994 10 144 220 - 10 144 220 - 10 144 220
1995 9 937 931 - 9 937 931 - 9 937 931
1996 9 760 032 - 9 760 032 - 9 760 032
1997 10 326 782 - 10 326 782 - 76 782 10 250 000
1998 10 176 497 - 10 176 497 - 10 176 497
1999 10 252 770 - 10 252 770 - 2 770 10 250 000
2000 10 102 490 - 10 102 490 - 10 102 490
2001 10 112 573 - 10 112 573 - 10 112 573
2002 10 058 115 - 10 058 115 - 10 058 115
2003 10 278 641 - 10 278 641 - 28 641 10 250 000
2004 9 744 870 5 130 9 750 000 5 130 9 750 000
2005 9 917 698 - 9 917 698 - 9 917 698
2006 9 817 416 - 9 817 416 - 9 817 416
2007 10 335 201 - 10 335 201 - 85 201 10 250 000
2008 9 913 568 - 9 913 568 - 9 913 568
2009 10 071 534 - 10 071 534 - 10 071 534
2010 9 696 066 53 934 9 750 000 53 934 9 750 000
2011 10 438 831 - 10 438 831 - 188 831 10 250 000
2012 9 966 486 - 9 966 486 - 9 966 486
2013 9 277 933 472 067 9 750 000 472 067 9 750 000
Average 10 005 847 10 038 370 10 018 141
Max Loss -7,2% -2,5% -2,5%
Std Dev (%) 2,7% 2,3% 2,0%
Table A.12: Hedging Effectiveness
No Hedge Option Collar
Max. loss due to adverse weather (% of average sales) 7.2% 2.5% 2.5%
Standard Deviation (% of margin) 2.7% 2.3% 2.0%
Cost of hedging (% of sales) - 0.3% 0.1%
17