The price elasticity for alcohol in Sweden 1984-2003

Research
report
T H O R N O R S T R Ö M
The price elasticity
A B S T R A C T
for alcohol in Sweden
T. Norström: The price elasticity for
alcohol in Sweden 1984–2003
1984–2003
AIMS
The article addresses the following
research questions: (i) How strong
is the price elasticity for beer, wine
and spirits? (ii) How rapid is the effect
of a price change? (iii) Is the price
elasticity stable across time and space?
(iv) Does an increase in price give a
Introduction
corresponding effect as a decrease?
The price of alcohol is considered to be one of
METHODS & DATA
the most important instruments for regulating
The sales data cover Systembolaget’s
overall consumption. This is because the de-
retail sales of beer, wine and spirits
mand for alcohol is sensitive to price changes,
for the period from January 1984 to
and the government has good control over March 2004. The price indexes are
prices through excise taxes. However, the pos-
based on weighted baskets defl ated
sibilities for a single country like Sweden to
by a consumer price index. Most of the
pursue a sovereign price policy has decreased
analyses were performed on quarterly
as a result of increasing economic integra-
data. The data were analysed using the
tion. Particularly the combination of practi-
Box-Jenkins technique for time series
cally abolished import quotas and low alcohol
analysis.
prices in Denmark and Germany has created
RESULTS
a troublesome situation. This has spurred a
The price elasticities—as estimated
marked increase in the private import of al-
from quarterly data—were statistically
cohol which undermines the legitimacy of signifi cant for all beverages; -0.8 for
Swedish alcohol policy. If the Swedish gov-
beer, -0.6 for wine and and -1 for
ernment chooses to adjust excise taxes to the
spirits. Similar estimates were obtained
lower levels of other countries it is important
from monthly data, suggesting a fast
to have a solid basis of knowledge for deter-
consumer response to price changes.
mining the effects on consumption and harm.
The elasticity for beer was weaker
The aim of this study is to analyse some is-
during the period 1995–2004 (-0.6) than
sues that concern the relationship between the
during the period 1984–1994 (-1.4), but
price and demand for alcohol. The next sec-
it was no different in southern Sweden
tion presents these issues.
than in the remainder of the country. An
A Swedish version of this article was published as: Nor-
increase in the price of spirits seems
ström, T. (2005) Priselasticiteten för alkohol 1984–2003. In:
to affect sales as much as a price
Gränslös utmaning – alkoholpolitik i ny tid. SOU 2005:25,
409–429.
decrease, that is, the price effect seems
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T
87

---

The price elasticity for alcohol in Sweden 1984–2003
to be symmetric. Finally, the
▀ What are the price elasticities for beer, wine and spirits?
results indicated that since
Existing reviews concerning price elasticities for alcohol
1995 sales of beer and wine
show a large range in the estimates (Ornstein 1980; Ornstein
increased more, and spirits
& Levy 1983; Österberg 1995; Österberg 2001). For instance,
sales less, than predicted
the estimated elasticities for beer vary between 0 and -3
from the development in
(Österberg 1995). Also after the elimination of extreme val-
prices.
ues, large differences remain. These may refl ect actual vari-
CONCLUSIONS
ations across time and space in sensitivity to price, but in
The study confi rms previous
addition differences in data quality and model specifi cation
fi ndings that the demand
have probably much impact on the outcome. When estimat-
of alcoholic beverages is
ing price elasticities one is in many countries confronted
responsive to changes in
with two fundamental methodological problems that hardly
price; however, price is not
apply to Sweden; one concerns the issue of exogenity, the
the sole factor that drives
other relates to data quality.
the trends in sales. The
Even though one should expect a relationship between
reduced elasticity for beer
price and the demand for alcohol the direction of the link is
may be due to the marked
not evident when a negative correlation is observed. A price
drop in beer prices.
increase should decrease demand, but decreasing demand
may also induce sellers on the market to reduce their prices.
In a free market price cannot be expected to be exogenous
(i.e. independent of demand) which complicates estimations
of price elasticities. An advantage of the Swedish alcohol
market in this context is that prices are not adjusted accord-
ing to demand but indeed are exogenous. Another advantage
is that the data on sales as well as prices are detailed and of
high quality. In the US, for instance, price data are based on
direct observation in certain stores of the prices for selected
brands. Since these brands are not always representative, and
only account for a fraction of total sales, these price series
have considerable measurement errors that create problems
in statistical analyses of the relationship between prices and
sales (Young & Bielinska-Kwapisz 2003).
Swedish data thus provide a sound methodological basis
for estimating the price elasticity for alcohol.
▀ How quickly is the impact of a price change realized?
It is common to distinguish between short-term and long
term elasticity, where the latter is usually the stronger of the
two. A motorist who faces an increased petrol price when
getting into the station hardly refrains from fi lling up because
of that. On the other hand, it is likely that s/he will reduce
driving for pleasure, and, when it is time to replace the car
88
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T

---

The price elasticity for alcohol in Sweden 1984–2003
s/he might switch to one that requires less
Data and methods
petrol. In this example there is a long-term
Sales data refer to Systembolaget’s (The
adjustment to the increased price, so that
State Alcohol Monopoly’s) retail sales of
it takes some time before the price change
beer, wine and spirits expressed in litres
has reached its full effect. How quickly do
of 100% alcohol. Price indexes are based
drinkers respond to a price change? Accord-
on weighted baskets that are defl ated by
ing to several studies the long-term elastic-
the cost of living index. Changes in price
ity exceeds the short-term elasticity in the
depend partly on actual price changes,
context of alcohol (Edwards et al. 1994). partly on changes in drinkers’ preferences
However, on the basis of these fi ndings it
towards cheaper or more expensive alco-
is diffi cult to infer how the time horizon hol. The latter source of change implies
looks, that is how long time it takes before
that the price is not entirely exogenous;
the impact of a price change is realized.
however, this drawback has to be weighed
against the drawbacks that a fi xed basket
▀ Is the price elasticity stable across
implies. Original data are on a monthly
time and space?
basis with a regional division (counties)
The price elasticity for a product is usually
and cover the period January 1984 – March
higher if there exists an alternative. Against
2004. For most of the analyses the data
this background we should expect that the
have been aggregated into quarterly data
elasticity would become stronger when the
for the whole country.
travellers’ allowances increased in January
One of the complications that are often
1995, which made cheaper alcohol more encountered in statistical analyses of time
available. One should also expect geo-
series data is that the series are trending,
graphical differences in the elasticity. The
which is also the case here (see fi gures 1–
increased travellers’ allowances spurred a
3). This may give rise to spurious relation-
marked increase in the private import of ships since two series may evolve in the
beer, particularly in southern Sweden. A same (or opposite) direction without be-
study by Norström (2000) shows a clear ing causally related to each other. Another
geographical gradient in this effect: it de-
complication is the structure of the error
creases proportionally to the square of the
term; the error term includes among other
distance to Helsingborg (where it is strong-
things causal factors that are not included
est). Can a corresponding pattern be found
in the analysis. One of the prerequisites
in the sensitivity to price changes?
in ordinary regression analysis is that the
error term does not have any structure. In
▀ Does a price increase yield a
time series analysis this assumption is not
corresponding effect as a decrease in
realistic since explanatory variables that
price?
are left out can be expected to be auto-
Due to the addictive character of alcohol it
correlated, that is to have a structure. In
is conceivable that the price effect is asym-
the present case there is the additional
metrical in the sense that a reduction in complication of seasonal variation that is
price yields a stronger impact than a cor-
found in monthly and quarterly data.
responding price increase.
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T
89

---

The price elasticity for alcohol in Sweden 1984–2003
The complications that have been de-
To illuminate the issue of the temporal
scribed here are taken into consideration stability of the price elasticity separate
in the technique for time series analy-
analyses are performed for the periods
sis that has been developed by Box and 1985:1-1994:4 and 1995:1-2004:1. County
Jenkins (1976), often referred to as ARI-
specifi c analyses elucidate the issue of re-
MA-modelling. By means of differencing gional differences in the price elasticity.
the series are made stationary. This means
By comparing elasticity estimates based
that rather than analysing the relationship
on monthly data with those based on quar-
between the raw series Y and X we ana-
terly data we get an indication of how fast
t
t
lyse the relationship between the chang-
a price change is realized. The topic of a
es, that is between ∇Y and ∇X , where possible asymmetry in the price effect is
t
t
∇Y =Y -Y . The differencing reduces the handled through the inclusion of a dum-
t
t
t-1
risk for spurious relationships, even if it my variable with feasible coding.
is not eliminated. Another feature of ARI-
MA-modelling is that the error term struc-
Results
ture is estimated and incorporated into the
The sales of beer, wine and spirits depict
model. This increases the reliability of the
fairly dissimilar trends during the study
model estimates.
period (fi gures 1–3). Sales of beer and
A log-log model of the following specifi -
wine were fairly stable until 1998, when
cation was used:
a strongly increasing trend started. There
is a decreasing trend for spirits sales dur-
lnS = elnP + N
ing the entire period. The prices have been
t
t
t
fairly stable for all beverages, with no
S is sales, P real price, and e denotes the
marked trends. The largest price change is
elasticity coeffi cient that is to be estimated.
noted for beer which decreased by about
N (noise) is the noise term that includes 15% in January 1997 due to a tax cut. The
other causal factors. The noise structure is
correlations between the trends in prices
estimated in terms of autoregressive and and sales are generally negative (table 1),
moving average-parameters. These are of which though should be interpreted with
two kinds; regular: AR(n) and MA(n), re-
great caution. However, the negative corre-
spectively (where n denotes the order of lations remain after seasonal differencing
the parameter), and seasonal: SAR(n), and
(table 1, fi gures 4–6).
SMA(n), respectively. An important crite-
rion of model fi t is that the residuals are
white noise. This is determined by means
of the Box-Ljung test.
90
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2. 2 0 0 5 . E N G L I S H
S U P P L E M E N T

---

The price elasticity for alcohol in Sweden 1984–2003
4.0
3.5
3.0
2.5
2.0
1.5
1.0
.5 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1
1984 -85 -86 -87 -88 -89 -90 -91 -92 -93 -94 -95 -96 -97 -98 -99 2000 -01 -02 -03 -04
Figure 1. Sales of beer (solid line) and real price of beer (broken line). Index 1984:1=1.
2.2
2.0
1.8
1.6
1.4
1.2
1.0
.8 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1
1984 -85 -86 -87 -88 -89 -90 -91 -92 -93 -94 -95 -96 -97 -98 -99 2000 -01 -02 -03 -04
Figure 2. Sales of wine (solid line) and real price of wine (broken line). Index 1984:1=1.
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T
91

---

The price elasticity for alcohol in Sweden 1984–2003
1.4
1.2
1.0
.8
.6
.4
.2
Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1 Q1
1984 -85 -86 -87 -88 -89 -90 -91 -92 -93 -94 -95 -96 -97 -98 -99 2000 -01 -02 -03 -04
Figure 3. Sales of spirits (solid line) and real price of spirits (broken line). Index 1984:1=1.
400 000
300 000
200 000
erenced)
100 000
0
, litres (seasonally diff
-100 000
-200 000
Beer sales
-300 000
-.3
-.2
-.1
-.0
.1
.2
Real price, beer (seasonally differenced)
Figure 4. Relationship between real price and sales of beer (litres 100%). Seasonally
differenced quarterly data 1984:1–2004:1.
92
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2. 2 0 0 5 . E N G L I S H
S U P P L E M E N T

---

The price elasticity for alcohol in Sweden 1984–2003
600 000
400 000
erenced)
200 000
0
, litres (seasonally diff
-200 000
Wine sales
-400 000
-.08 -.06 -.04 -.02 0.00
.02 .04 .06 .08
Real price, wine (seasonally differenced)
Figure 5. Relationship between real price and sales of wine (litres 100%). Seasonally
differenced quarterly data 1984:1–2004:1.
400 000
200 000
erenced)
0
, litres (seasonally diff -200 000
its sales -400 000
Spir
-600 000-.08 -.06 -.04 -.02 0.00 .02 .04 .06 .08 .10
Real price, spirits (seasonally differenced)
Figure 6. Relationship between real price and sales of spirits (litres 100%). Seasonally
differenced quarterly data 1984:1–2004:1.
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T
93

---

The price elasticity for alcohol in Sweden 1984–2003
Table 1. Correlation between sales and real
where Assarsson’s estimate was somewhat
price of beer, wine and spirits. Based on
stronger. It should be mentioned that As-
quarterly data for the period 1984:1–2004:1
sarsson used another techique for time
Raw data
Seasonally
differenced
series analysis, and that he also included
Beer
-0.62
-0.18
additional explanatory variables.
Wine
-0.12
-0.33
Models including cross-elasticitities
Sprits
-0.51
-0.38
were estimated as well (not shown); e.g.,
the model for beer included the prices of
wine and spirits, in addition to the beer
▀ Estimation of the price elasticitities
price. In neither of the models for beer,
Table 2 summarizes the results for the wine and spirits were the cross-elasticiti-
price elasticitities (complete model es-
ties statistically signifi cant.
timates are found in the Appendix). All
It is of interest to note that the elastici-
estimates but one are negative and statisti-
tities estimated on monthly data hardly
cally signifi cant. It can be noted that the differ from those based on quarterly data;
elasticity for beer becomes markedly lower
the differences are unsystematic and not
after 1994; during the fi rst period it is -1.4,
statistically signifi cant. This would imply
compared to -0.6 during the period after that people respond quickly to a change in
1994. The elasticity for wine is at approxi-
the alcohol price.
mately the same level during both of the
The decreasing price elasticitity for beer
periods. The elasticity for spirits is close after 1994 is intriguing. It can hardly be due
to -1 for the early period; the fact that it is
to limited variation in price, as in the case
insignifi cant after 1994 is probably due to
of spirits. On the contrary, the variation in
the small variation in the price of spirits beer price was largely due to the price cut
during that period.
of 15% in January 1997. This price cut has
In comparison it can be mentioned that
rather the form of a natural experiment,
Assarsson (1991) for the period 1970–1988
and inspires a separate analysis. Thus the
estimated the elasticitities for beer, wine elasticitity for beer was estimated for a pe-
and spirits at -1.3, -0.9, and -0.9, respec-
riod that was dominated by this price cut,
tively. The agreement between Assarsson’s
that is, 1995:1–1998:4. Figure 7 indicates
estimates and those presented above for a clear and negative relationship between
the period prior to 1995 is thus fairly good;
prices and sales (seasonally differenced);
the largest discrepancy is noted for wine,
this is also verifi ed by the model estima-
Table 2. Estimated price elasticities for various time periods
Quarterly data
Quarterly data
Quarterly data
Monthly data
1984:1–1994:4
1995:1–2004:1
1984:1–2004:1
1984:1–2004:3
Beer
-1.36***
-0.55*
-0.79***
-0.90***
Wine
-0.62**
-0.81(*)
-0.57**
-0.63**
Sprits
-1.16***
0.34
-0.96***
-0.81***
***
**
*
(*)
p<0.001; p<0.01; p<0.05; p<0.10
94
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2. 2 0 0 5 . E N G L I S H
S U P P L E M E N T

---

The price elasticity for alcohol in Sweden 1984–2003
300 000
erenced) 200 000
100 000
, litres (seasonally diff
0
Beer sales -100 000-.3
-.2
-.1
0.0
.1
Real price, beer (seasonally differenced)
Figure 7. Relationship between real price and sales of beer (litres 100%). Seasonally
differenced quarterly data 1995:1–1998:4.
tion (table 3). The estimated elasticitity decreases, and 0 if the price remains the
(-0.60) is on a par with the one found for
same. The models were estimated for spir-
the entire period after 1994.
its only, since this is the beverage where
the presence of this sort of asymmetry
▀ Is the price effect asymmetric?
seems most probable. Table 4 shows that
Two models were estimated to elucidate the dummy variable did not have any sig-
the issue of whether a price increase and
nifi cant effect, nor did it affect the esti-
a price decrease are equivalent in terms of
mate of the price elasticity (the difference
absolute effects. The fi rst model (Model 1)
between the estimated price elasticity in
included the price series only as explana-
Model 1 and Model 2 is not statistically
tory variable. The second model (Model signifi cant).
2) included in addition to the price series
a dummy variable (denoted Change), that
▀ Geographical differences in price
(after ordinary differencing) took the val-
elasticity
ue 1 if the price increases, -1 if the price
To examine the issue of geographical dif-
ferences in price elasticity, county specifi c
Table 3. Price elasticity for beer estimated
on seasonally differenced data for the period
analyses were performed for beer sales
1995:1–1998:4
during two periods of time: 1984–1994
Coeff
SE
and 1995–2004. Beer is the beverage for
which the availability of alternatives to
Price
-0.60***
0.12
Systembolaget ought to be largest, espe-
Q(4)+
2.18; p> 0.70
cially during the latter period with its in-
*** p<0.001
creased travellers’ allowances. According
+Box-Ljung test for autocorrelated residuals (lag 4)
to Norström’s study (2000) it is particu-
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2 . 2 0 0 5 . E N G L I S H
S U P P L E M E N T
95

---

The price elasticity for alcohol in Sweden 1984–2003
Table 4. Estimated price elasticity for spirits. Change is a dummy variable (see text). Based on
regularly differenced quarterly data 1984:1 – 1994:4
Model 1
Model 2
Coeff
SE
Coeff
SE
Price
-0.79**
0.26
-0.90*
0.35
Change
0.004
0.01
AR1
-0.64
0.11
-0.64
0.11
SAR1
0.48
0.15
0.48
0.15
SAR2
0.45
0.16
0.45
0.16
Q(4)+
3.35; p> 0.50
3.29; p> 0.51
** p<0.01; * p<0.05
+Box-Ljung test for autocorrelated residuals (lag 4)
larly in southern Sweden that one can ob-
a hint of a relationship in the expected di-
serve an increased private import of beer
rection, but the spread around the regres-
from Denmark after the increased quotas sion line is considerable.
in January 1995. Thus, it is in this area we
should expect an excess sensitivity to the
▀ Projections of trends in consumption
beer price, particularly after 1994. Accord-
How would the consumption of beer, wine
ing to the results, this is not the case; the
and spirits have evolved during the last
average price elasticity for these counties
ten years if it had been determined solely
is about the same as in the remainder of
the country during both of the time peri-
Price elasticity
ods (table 5). Further analyses of the coun-
-1.3
ty specifi c elasticity estimates for southern
Sweden show no relationship between the
-1.4
price elasticity and the distance to Hels-
ingborg during the early period (fi gure 8).
-1.5
During the latter period (fi gure 9) there is
Table 5. Price elasticity for beer in southern
-1.6
Sweden and in the rest of Sweden estimated
on data for two different time periods.
Average (standard deviation in parentheses)
-1.7
of county specifi c estimates.
1984:1–
1995:1–
-1.8
1994:4
2004:1
0
100
200
300
Southern Sweden*
-1.48 (0.13)
-0.59 (0.14)
Kilometres
Rest of Sweden
-1.38 (0.15)
-0.56 (0.17)
Figure 8. Relationship between county
* Southern Sweden includes the municipality of Helsingborg,
specifi c price elasticity for beer (estimated
the municipality of Malmö, and the following counties: Malmö-
for the period 1984:1–1994:4) and distance
hus, Halland, Kristianstad, Kronoberg, Göteborg and Bohus,
Blekinge, and Jönköping.
to Helsingborg in kilometres.
96
N O R D I S K A L K O H O L - & N A R K O T I K A T I D S K R I F T V O L . 2 2. 2 0 0 5 . E N G L I S H
S U P P L E M E N T

---