Addiction and the Demand for Alcoholic Beverages:

Evidence from Italian Data

Pierpaolo Pieraniand Silvia Tiezzi
Department of Economics - University of Siena
October 14, 2004

Abstract
In this paper we estimate the demand for alcoholic beverages in Italy following the
rational addiction demand framework by Becker and Murphy (1988) and using the GMM
estimator. We use time series of national accounts data on an aggregate of alcoholic
beverages expenditures from 1960 to 2002 supplied by ISTAT. We take into account Auld
and Grootendorst (2004) recommendations when estimating the rational addiction model
using aggregate time series. The data seem to fit the model quite well implying that
alcohol consumers are actually forward looking. We also get a reasonable estimate of
the intertemporal rate of time preference at the sample mean. Short and long run price
elasticities of demand as well as the income elasticity of demand are also calculated. The
long run elasticity is higher than the short run one, but the demand is rather unelastic.
Surprisingly, the income elasticity of demand for alcohol, as derived from the rational
addiction model, is higher than one both in the short and in the long run so that alcoholic
beverages turn out to be luxury goods.
Keywords: rational addiction; alcohol consumption;
JEL Classification D12, C23

Acknowledgments:

We would like to thank, without implicating, Pier Luigi Rizzi for his insightful comments and critiques on earlier versions of the paper.
Corresponding author: Pierpaolo Pierani, Dipartimento di Economia Politica, Piazza San Francesco, 7,
Siena, Italy. E-mail: pierani@unisi.it.

Introduction

The consumption of alcoholic beverages has features and consequences that make this class
of commodities dierent from traditional consumption goods. As Cook and Moore (2000,
p. 1635) have emphasised, alcohol is an intoxicant: consumed in sucient quantity in a
single session it impairs mental and physical functioning and it is potentially toxic. Secondly,
alcohol consumption has direct intertemporal consequences: past consumption generates habit
formation and addiction. Thirdly, chronic alcohol use aects physical and mental health over
the course of years or decades. While the second and third consequences are also typical
of other addictive goods, such as cigarettes for instance, the first one is peculiar of alcohol
consumption and generates important social costs.
In this paper we estimate the demand for alcoholic beverages in Italy testing the rational
addiction model developed by Becker and Murphy in 1988. Its estimable version1 has produced a large bulk of empirical literature, because of its relevant policy implications. More
specifically, the model implies that the demand for addictive goods is quite sensitive to price
changes in the long run. Therefore, contrary to common sense beliefs, price changes could be
eective measures of curbing alcohol consumption, for instance.
Our study is primarily concerned with contributing towards this strand of literature as well
as investigating the demand for alcoholic beverages in Italy. To our knowledge, only very few
examples exist that confront the rational addiction hypothesis with Italian data and this paper
is the first attempt at providing such an evidence on alcohol consumption. The investigation
period covers the years 1960 - 2002. We use time series on an aggregate of alcoholic beverages
expenditures and estimate, with the GMM, a structural rational addiction demand equation
as well as a myopic demand function.
The paper is structured as follows. An overview of alcohol consumption and alcohol policies,
both in Italy and in Europe, is given in section 2. In section 3 we briefly review the analytical
framework and the existing empirical evidence on rational addiction to alcohol. Section 4
explains the empirical methodology and shows the results. Section 5 concludes.
1 The estimable version of the Rational Addiction Model has recently been criticised by Wangen (2004) who
has stressed how the estimable structural rational addiction equation is valid only provided that some rather
strong assumptions hold.

Alcohol Consumption and Policies

2.1

Alcohol Policies in the European Region

Since 1992 the World Health Organization (WHO) has advocated a combined approach to
reduce harm resulting from the use of alcohol, tobacco and drugs. To this aim, the second
European Alcohol Action Plan (EAAP) for the years 2000 - 2005 has been launched by the
WHO Regional Committee for Europe at its forty-ninth session in 1999. This is a first attempt
to establish a regional policy framework, a set of guidelines to which member states have
committed themselves. In February 2001 a joint conference WHO - European Union (EU)
has put forth a set of strategies aimed at reducing alcohol related harm among the young.
The objectives set forth in the Declaration on Young People and Alcohol are to be reached
by 2006. The debate that has followed in the EU has produced a Recommendation of the
Council of Europe (Drinking of alcohol by Young People, in particular children and adolescents,
2001/458/EC) and the following Conclusion (Council Conclusion on a Community strategy
to reduce alcohol-related harm 2001/C 175/01). A further contribution of the EU is given by
the new Public Health Project Community Action in the Field of Public Health 2003 - 2008
recently adopted by the European Parlamient (Decision of the European Parlamient and of
the Council n. 1786/2002/EC) that identifies, among others, the main initiatives to be taken
within the EU to modify individual behaviours related to alcohol, drugs, food and cigarettes
consumption.
The European Region of the WHO includes 51 countries2 and about 870 millions of people.
Within this region, alcohol products are estimated to be responsible for about 9% of the
total disease burden (Rehn, Room and Edwards, 2001, p. 21). They are linked to accidents
and violence and are responsible for a large proportion of the reduced life expectancy in the
countries of the former Soviet Union. Estimating the cost of alcohol consumption is very
dicult because of the problem in calculating the costs of unhappiness, pain, general ill health
that may be somehow linked to alcohol consumption. For this reason alcohol costs are likely
to be under estimated. Most often the social costs included in the calculations are: workplace
production losses, the costs of accidents, health care expenditures, the costs of treatment,
social welfare payments and social community costs (crime, enforcement and penal costs). It
is estimated that the total net (balanced against income from alcohol) social costs of alcohol
in the European region amount to between 1 and 3% of gross domestic product (Klingemann,
2001, p. 9). A first dierence between the social costs of smoking and the social costs of
2 The Russian Federation, most of the eastern european countries including the countries of the former
Soviet Union, Israel and the European Union.

drinking can be traced by noticing that smoking does not give rise to costs of accidents
and the social community costs mentioned above. These are more properly linked to the
consequences of binge drinking. Based only on the recorded level of alcohol consumption for
1998, the WHO Regional Oce for Europe has divided the Region into countries with a high
level of consumption (more than 10 litres of pure alcohol per person per year), middle level
of consumption (more than five litres per person per year) and low level of consumption (less
than five litres per person per year) with Italy belonging to the second group3 .
Alcohol policy in the enlarged European Region is implemented through dierent strategies: information and education; restrictions of alcohol consumption in dierent settings; drink
driving campaigns; restrictions on the availability of alcohol products; promotion and advertising; treatment of alcohol related health problems; taxation specifically aimed at reducing
alcohol consumption. The EU has adopted directives that relate to three aspects of alcohol
control policy: taxation, advertising and transportation between member states of alcohol
for personale use. The EU has also attempted to harmonize taxation on alcohol among its
member countries, but to date it has succeeded only in setting minimum excise rates4 . The
EU has also placed restrictions on the advertising of alcohol on television5 .

2.2
2.2.1

Alcohol consumption and policy in Italy

Alcohol Consumption

In the year 2000 the average per capita consumption of pure alcohol in Italy has been about
7.5 litres (Ministero della Salute, 2003, p. 12), but according to the WHO for the European
Region, the target of 6 litres per capita per year should be reached by the year 2015.
Alcohol consumption has followed a decreasing trend since the early eighties. If we look
at aggregate consumption of alcoholic beverages over time in Italy, the per capita level has
decreased, between 1970 and 2001, by 51.25%, as can be seen in figure 1.
3 Between 1988 and 1998 Bulgaria, Estonia, Italy, Switzerland and Ukraine have achieved a rather large
decrease in the recorded level of consumption, although Estonia and Ukraine have a high level of unrecorded
consumption (Rehn, Room and Edwards, 2001, p. 9). During the same decade, however, a number of countries
has shown a considerable increase in the recorded level of alcohol consumption: Turkey, the Russian Federation,
Belarus, Ireland, Latvia and the Czech Republic. Countries with a relatively stable level of alcohol consumption
include Denmark, Finland, Germany, Iceland, the Netherlands, Norway, Slovenia and the United Kingdom.
4 According to the Council Directive 92/83/EEC of 12 October 1992, the minimum rates, from January
1993, are as follows: for wine (still and sparkling): ECU 0; for beer: ECU 0,748 per hectolitre per degree Plato
or ECU 1,87 per hectolitre per degree of alcohol of the finished products; for intermediate products (beverages
with an alcohol content under 22% and not belonging to the group of wines and beers): ECU 45 per hectolitre
of product; for spirits: ECU 550 per hectolitre of pure alcohol (Rehn, Room and Edwards, 2001, p. 68).
5 The Council Directive 89/552/EC of 3 October 1989 restricts the content of alcohol advertisements on TV.

16

15

14

13

12

11

10
9

8
1970

1975

1980

1985

1990

1995

2000

Figure 1: Pure Alcohol Consumption: Litres per-capita

Source: Health for All Database, 2003.

These kind of data, however, present a number of drawbacks. On one hand, aggregate data
may be dominated by the behaviour of light and moderate drinkers and a decrease in aggregate
consumption could hide a rather dierent picture. For this reason, the use of micro-data is
recommended for this kind of studies. On the other hand, addictive behaviour, for instance,
could be more easily captured by data on spirits consumption only, because the distribution of
spirits is tipically more bimodal (as predicted by the Becker and Murphys model of addiction)
than that of wine and beer. Aggregate data separated for wine, beer and spirits are available,
for Italy, from 1981 to 2000. Table 2, taken from the Italian Institute of Health (Scafato and
Russo, 2004, p. 4), shows that the total per capita decrease in alcohol consumption from
1981 to 2000 results from a 40.8% decrease in wine consumption; a 65.7 % decrease in spirits
consumption and a 57% increase in beer consumption.
A rather dierent pattern of consumption emerges from the study of Bentzen, Eriksson
and Smith (1999) on alcohol consumption in four northern European countries: Denmark,
Sweden, Finland and Norway. Here the consumption of spirits has almost halved from 1960 to
1992 in the four countries. Beer and wine consumption have increased in Finland, Norway and
Sweden, whereas beer consumption has decreased in Denmark, where beer consumption has
always been very high. Overall, the consumption of pure alcohol over time in the four nordic
5

countries shows an increasing trend, because the decrease in spirits consumption is more than
compensated by the increase in wine and beer consumption. Italy seems to show an opposite
trend: here the increase in beer consumption does not compensate for the decrease in wine and
spirits consumption also because beer has the lowest alcohol content among the three types of
alcoholic beverages.
Despite the obvious drawbacks in aggregate data, we could check whether this evolution
in alcohol consumption in Italy bears some relationship with the evolution of the real price,
as would be expected when studying the consumption trend of a traditional commodity. In
figure 2 we have plotted an index (equal to 1 in 1995) of the real per capita (14 or older)
expenditure on all alcoholic beverages and the real alcohol price over the years 1960 - 2002.
A quick glance at the picture reveals that the consumption of alcoholic beverages does not
seem alike that of traditional commodities. It is only from around 1977 onwards that the
relationship between the real price and the real expenditure seems to be as expected: the real
price decreases sharply to reach a minimum in 1985, whereas real expenditure remains almost
stable at 1977 levels. From 1985 onwards the price starts increasing again and, as expected, real
consumption decreases continuously to reach a minimum in 2002. The correlation coecient,
calculated over the whole period, between these two variables is indeed rather low and equal
to -0.29. The graph seems then to suggest that alcohol consumption before 1977 was driven
by variables other than the price, whereas later on the role of the real price has become more
relevant in explaining trends of real consumption. The correlation coecient, calculated over
this subperiod, is indeed equal to -0.54.

ALC

1.7

1.6

1.5
1.4
1.3
1.2

1.1
1.0

1960

1965

1970

1975

1980

1985

1990

1995

2000

Figure 2: Index (1995=1) of per capita alcohol expenditure and real alcohol price.

Source: Italian National Accounts.

In 2001 75% of Italians older than 14 has consumed alcoholic beverages: 87% of men and
63% of females (table 1) (Scafato, Ghirini and Russo, 2004, p. 14).
Survey data on life styles and health conditions of individuals are collected every two years
by ISTAT, the Italian Statistical Oce, and may help understanding what are the variables,
other than the price, driving alcohol consumption. These survey data, however, are available
only for the years 1993 - 2001 (Indagine Multiscopo sulle famiglie - stili di vita e condizioni di
salute).
The observed sharp decrease in the per capita consumption has been accompanied, during
the nineties, by a remarkable change in the drinking habits. This change is characterised
by three main facts: a) an increase in the number of female consumers; b) an increase in
the number of young consumers (teen agers and people aged 18 to 24); c) an increase in the
number of people (and the increase is higher for females and the young) consuming alcohol
outside the main meals.
The increase in the number of alcohol consumers on one hand and the sharp decrease in
the per capita level of consumption, on the other, seem to testify a change in the consumption
habit. Italy is a wine producer country where, traditionally, wine has always and mainly been
consumed, on average, in moderate quantities and by all members of the household, to accom-

pany meals. The changing pattern seems to suggest a transition from a Mediterranean model
of alcohol consumption to a drinking pattern closer to the northern european one characterised
by binge drinking and by the use of alcohol as a bridge to ease personal relationships and wane
down social discomfort or as a means of female emancipation and cultural homologation. If
this is true, then the comforting picture of a continuously decreasing level of alcohol consumption could hide an increase in the number of people actually at risk, especially among the most
fragile groups of the population.
2.2.2

Alcohol Policy

Italys alcohol control policy seems to rely mainly on command and control policy options such as restrictions and prohibitions - and on voluntary agreements. Preventive and treatment
strategies are less developed whereas the use of economic policy instruments, such as taxes
aimed at reducing alcohol consumption, is not contemplated.
The majority of countries has a system of two taxes on alcoholic beverages: a value added
tax (VAT) and a specific alcohol tax. Italy has a VAT on alcohol (20%) higher than on other
products or services. The majority of the European countries have a VAT on alcohol close
to 20%. The specific alcohol tax is mostly based on the volume of alcohol: the stronger the
beverage, the higher the tax, but this is not the case in Italy. The ocial objective of alcohol
taxation in Italy is to raise revenues to finance public expenditure. Reducing total alcohol
consumption is not contemplated among the objectives of taxation. Moreover, Italy does not
tax wine. To ensure the collection of taxes and to counteract smuggling duty paid stamps for
spirits have been introduced.
Promotion and advertising of all types of alcohol is restricted on TV, radio, print media
and billboards with two exceptions: advertising of spirits on billboards and of beer on the
radio is only subject to a voluntary code. A fine is imposed as a punishment if advertising
regulations are breached.
The legal maximum blood alcohol concentration (BAC) in Italy is, since 2001, 50 mg%.
The increased risk for young and professional drivers is recognised and a lower BAC limit
for those categories is required. Random Breath Testing (RBT) has been introduced. The
police needs justified suspicion to breath test drivers and the penalties for drink driving are a
suspension of the driving licence from 15 days to 3 months, a fine and imprisonment of up to
one month.
As to strategies aimed at limiting the availability of alcohol products, Italy has a licensing
system for import, export, production, wholesale and retail sale of all types of alcoholics. There

is a formal procedure to obtain a licence for retail outlets, but all applicants get a licence, thus
the procedure is almost meaningless. Restrictions on sale of alcoholic beverages relates only to
sales at specific events: childrens festivals, sporting events and demonstrations. During these
events, the restriction is partial for beer, voluntary for wine and total for spirits. There is an
age limit of 16 for access to alcohol purchase.
Between 1994 and 1999 Italy, together with other European countries, has introduced
policy changes aimed at increasing alcohol consumption control. These changes have occurred
in two main areas: drink driving legislation and alcohol promotion. Random Breath Testing
(RBT) has been introduced to reinforce drink driving controls. However, the overall level of
enforcement of drink driving legislation is regarded as loose. To reduce alcohol promotion,
Italy has introduced more restrictions on sale and alcohol advertising: the age limit of 16 has
been introduced for alcohol purchase and the hours of sales of spirits have been restricted.
The most recent public measure concerning alcohol consumption is the law 125/2001 (Legge
Quadro in Materia di Alcohol e di Problemi Alcolcorrelati, 30.03.2001 n. 125). This law
introduces measures aimed at the prevention and treatment of alcohol related problems in line
with the recommendations of the European Alcohol Action Plan.
While these command and control strategies are important in controlling alcohol consumption and related problems, their monitoring and enforcement is the crucial key to determine
their eectiveness. Moreover, the implementation of the law provisions requires an extra expenditure of 2.130 Millions Euro per year.
The use of economic incentives, such as taxes, is not contemplated among the measures
that may help reaching the goal of reducing alcohol related social costs. In this respect, the
economic models of demand and, particularly, the addiction models, may help highlighting
the eectiveness of price mechanisms in reaching those targets.

Analytical Framework and Empirical Evidence

The usual way of modeling alcohol consumption is, nowadays, the rational addiction (RA)
framework developed in the seminal paper by Becker and Murphy (1988). Previous empirical
research developed the notion of dependency via models of habit formation, otherwise known
as myopic model of addiction, or else, it ignored the addictive nature of alcohol and modeled
alcohol demand as a traditional consumption good.
Following Becker, Grossman and Murphy (1994) (BGM henceforth), we assume that the
addicted consumers problem is to maximize, over the life cycle, the following utility function:

V =

t1 U (Yt , Ct , Ct1 , et )

(1)

t=1

where Yt is consumption of a non addictive good (used as numeraire) at time t, Ct is

consumption of the addictive good, alcohol in this case, at time t, Ct1 is consumption of
alcohol in the previous time period, et reflects unmeasured life cycle variables that aect
utility, whereas is the discount factor 1/(1 + r), with r being the intertemporal rate of time
preference. An increase in past consumption raises the marginal utility of current consumption
of the addictive good and leads to an increase in current consumption. When the instantaneous
utility function is quadratic and the intertemporal rate of time preference is equal to the market
interest rate equation (1) generates a structural demand equation of alcohol consumption at
time t of the following form:

Ct = Ct1 + Ct+1 + 1 Pt + 2 et + 3 et+1

(2)

where Pt is the price of the addictive good at time t. Thus, current consumption is
positively related to past and future consumption and negatively related to the current price.
In particular measures the eect of past consumption and, by symmetry, it also measures the
eect of an increase in future consumption on current consumption of the addictive good. The
larger the value of the greater is the degree of reinforcement or addiction. BGM recognize
that the et may be serially correlated. Even when they are not, they may aect utility in
each period and aect consumption at all dates through the optimizing behaviour implied by
equation (1). Therefore BGM treat Ct1 and Ct+1 as endogenous and use lagged and forward
prices as instruments. The statistical significance of the coecient of future consumption,
together with a reasonable estimate of the intertemporal rate of time preference, provides a
direct test of the rational addiction theory. Rational addicts will increase their consumption
when future prices are expected to fall.
Equation (2) has some important implications. First, complementarity between adjacent
periods of consumption or negative cross-price elasticities between alcohol consumption at
dierent points in time, holding prices in all other periods constant.
Secondly, there are important dierences between long and short run responses to permanent price changes (a price change in the current period which is not followed by an equal
change of the opposite sign in the following periods). The short run eect describes the impact
on consumption of a price change at time t and in all future time periods that is not anticipated until time t. The long run price eect pertains to a price change in all time periods.

10

Since Ct1 remains the same if a price change is not anticipated until period t, the long run
price eect must exceed the short run price eect. The BGM empirical implementation of
the RA theory has been recently criticised by Wangen (2004) who has argued that the tests
of significance rely on the assumption that the coecients under test are constant under the
alternative hypothesis. By using this testing scheme we implicitly make two rather strong
assumptions: a) a priori rationality of the consumer, i.e. that a price change at a point in
time had been perfectly anticipated in a distant past; b) the market interest rate and the
intertemporal rate of time preference are equal, so that the consumer has no inclination to
shift her consumption stream either into the past or into the future with respect to her income
stream, and that both are constant over time.
Tests of the RA model have been applied mainly to cigarettes consumption. Applications to
alcohol consumption are not as numerous. One reason for this is that alcohol is not supposed
to be as addictive as Tobacco. Another reason, put forward by Bentzen et al. (1999) is
probably the lack of long and consistent time series or panel data at the individual level.
Empirical tests of RA to alcohol have been carried out by Grossman, Chaloupka and Sirtalan
(1998); Waters and Sloan (1995); Baltagi and Grin (2002); Bentzen, Eriksson and Smith
(1999). Grossman et al. (1998) use surveys of high school seniors as part of the monitoring
of the future results to support the RA theory and they find a long run price elasticity which
is approximately 60% larger than the short run elasticity. However, their estimates of the
discount factor are negative and implausibly high yielding negative interest rates. Waters and
Sloan (1995) use a cross section of individual data to test the RA model applied to alcohol
consumption. Their estimations provide relatively strong support for the RA model. Baltagi
and Grin (2002), using aggregate consumption data on distilled spirits for 42 states over the
period 1959-1994, obtain results that are in general supportive of the RA hypothesis. However
their results are sensitive to the assumption of homogeneity across states or over time (Baltagi
and Grin, 2002, p. 486). Finally Bentzen et al. (1999) use aggregate time series data over the
period 1960-1994 to test the RA model for alcohol consumption across four northern european
countries and find support for the RA hypothesis mainly for wine and spirits whereas evidence
for beer is more controversial.

11

Empirical Strategy

4.1

Data

We use aggregate time series of alcoholic beverages expenditure (in millions Euro) for Italy for
the period 1960 - 2002 taken from the ISTAT National Accounts dataset. Per capita level of
expenditure is obtained by dividing the aggregate level by the population older than 14 (calculated in the middle of each year). The real price of alcohol is obtained by dividing the implicit
deflator, calculated as the ratio between current expenditure on alcohol and expenditure on
alcohol at 1995 prices, by the consumer price index (1995=1).
Figure 2 shows the index (1995=1) of per capita expenditure on alcohol and the real
price normalised in 1995. We have also added, to the set of explanatory variables, the real
expenditure on non durables as a proxy of disposable income. Summary statistics and details
of the data are presented in Table 3.
Recently, Auld and Grootendorst (2004) have demonstrated that estimable RA models
tend to yield spurious evidence in favor of the RA hypothesis when aggregate time series are
used. They do so by showing that even non addictive commodities such as milk, eggs and
oranges turn out to be rationally addictive when using this kind of data. More specifically
spurious evidence for rational addiction is likely to be obtained when: 1) the consumption
series is highly autocorrelated; 2) even a small amount of the variation in prices is endogenous;
3) the value of the discount rate is exogenously imposed and, 4), overidentified instrumental
variable estimators are used. To overcome these problems Auld and Grootendorst (2004, p.
17) make some tentative recommendations for researchers that we have tried to take into
account.
We have performed a battery of diagnostic tests on time series to avoid biases in the
direction of finding rational addiction due to the nature of the data used. First of all we have
tested for price endogeneity. We perform a Hausman-Wu (HW) test6 . This is a Likelihood
Ratio (LR) test distributed as a 2 with 34 degrees of freedom. The value of the test statistics
is 0.793 with a p-value of 1. We thus accept the null hypothesis of price exogeneity. Secondly,
we have tested for stationarity of the time series using two dierent types of unit root tests:
the Dickes-Fuller and Weighted Symmetric. We find a unit root for all the variables used:
P , C and Y . Therefore all estimations have been made with the variables in first dierences.
6 The HW test compares the original demand equation (estimated with OLS) with the unrestricted model
that includes, among the explanatory variables, the residuals of an auxiliary regression. In the auxiliary
regression the real price is the dependent variable whereas the explanatory variables include a linear time
trend, a constant and the aggregate quantity of Alcoholic beverages sold. See Davidson and McKinnon (1993)
for a description of the Hausman - Wu exogeneity test.

12

Finally we have tested for autocorrelation of the dierenced variables using two dierent tests:
a Durbinh alternative and a Breusch/Godfrey LM test. For all autocorretation tests we reject
the null of no autocorrelation7 and accept the hypothesis of serial correlation of order 2 in the
error terms. Results of these diagnostic tests are shown in table 4.

4.2

Estimation

For our empirical implementation we write a variant of (2) with all the variables in first
dierences as follows:

Ct = 0 + 1 Ct1 + 2 Ct+1 + 3 Pt + 4 Yt

(3)

where Yt is a proxy of disposable income and 0 is an intercept term. If the model is

correctly specified one should have 0 = 0. However, we allow for a non-zero intercept in all
estimations in order to avoid misspecification bias in the other parameters. We treat Ct1
and Ct+1 as endogenous and use the Generalised Method of Moments (GMM) which produces
consistent and asymptotically ecient estimates of the parameters of interest when the errors
are serially independent. As Auld and Grootendorst (2004, p. 17) have noted, estimating the
model in dierences is likely to yield better small-sample properties than estimation in levels
for commodities exhibiting moderate to high serial correlation in consumption. The GMM
does estimation on a set of orthogonality conditions which are the products of equations and
instruments. These orthogonality conditions are minimized in the metric of an estimate of
their expected covariance. Our instruments set includes three leads and three lags of both
the alcohol price and disposable income. We therefore have 10 extra instrumental variables
and test for the validity of the exceeding overidentifying conditions using the Sargan or J
test. Standard errors are computed from an heteroscedastic-consistent covariance matrix of
the orthogonality conditions using the White procedure. In the dierenced model we also
control for serial correlation of the disturbances using a covariance matrix of the orthogonality
conditions that incorporates Moving Average (MA) disturbances of order 2.

4.3

Results

The estimation results are shown in table 5. Looking at the dierenced model results, all the
estimated parameters have the expected sign and are statistically significant at the 95% level
7 The Durbins h alternative foolows a normal distribution and it is a valid test for autocorrelation when
more than one lagged dependent variable is included in the regressors.
The Breusch-Godfrey LM test for autocorrelation of order x follows a 2 distribution with DF = x + k 1,
where x is the number of lags and k is the number of identified coecients in the model, including the intercept.

13

of significance. The lagged consumption coecient is, consistently with the theory, larger
than the lead consumption coecient thus giving rise to a positive and reasonable rate of
time preference. The test of overidentifying restrictions is distributed as a 2 with degrees of
freedom equal to the number of exceeding instruments. The critical value of the 2 at the 95%
level of significance with 10 degrees of freedom is 18.31, we cannot therefore reject the null
of no overidentification. Moreover, the sum of the lead and lagged consumption coecient is
well below one and amounts to about 0.71.
These results are consistent with most of the recommendations of Auld and Grootendorst
(2004) concerning the estimation of the RA model when using aggregate time series data. We
have estimated the model in dierences to avoid the small sample bias that may arise when
the studied commodity may exhibit serial correlation. Moreover the value of the intertemporal
rate of time preference is not imposed ex ante as a constraint on the model.
BGM have derived formulas to compute the short run (SRE) and the long run (LRE) price
elasticity of demand. Taking into account equation (2), the resulting elasticities at the sample
mean, are:

SRE =

dCt P
21
P
i
=h

1/2
2
dPt C
C
1 2 + 1 4

(4)

1
dC P
P
=
(5)
dP C
1 (1 + ) C
In the rational addiction model the short run price elasticity of demand gives the percentage
LRE =

variation in the consumption of an addictive good in the first year after a permanent change
in the current price and all future prices, with past consumption held constant. The long run
price elasticity gives, instead, the percentage change in consumption following a permanent
price change in all time periods (Becker, Grossman and Murphy, 1991, p. 240).
The implied SRE and LRE at the sample mean of the dierenced model, are -0.26 and -0.47,
respectively, and the implied intertemporal rate of time preference for the mean estimates is
about 22% (table 6). We have also calculated the implied short and long run income elasticities
of demand. The algebraic form of the elasticities is the same as that of the price elasticities.
The income elasticity of alcoholic beverages turns out to be 1.17 and 2.17. The demand
for alcoholic beverages is thus a luxury good: a 1% increase in income causes a more than
proportional increase in the demand for alcohol. This response almost doubles when the
permanent income change pertains to all time periods.
According to the RA model, the marginal utility of income is a multiplying factor in the
current price coecient (Escario and Molina, 2001, p. 213), so that an increase in the marginal
14

utility of income will produce a greater increase in the price coecient. This implies that rich
people, who possess a lower marginal utility of income, will be less sensitive to price changes
than people with lower income and a higher marginal utility of income, who, instead, turn out
to be more sensitive to price changes.
Table 8 shows the short and long run price elasticities of demand (obtained from the
parameters of the dierenced rational addiction model) estimated at the sample mean of each
decade. The demand for alcohol appears to be rather sensitive to price changes over the
sixties, when Italy experienced a lower mean level of income, than during the eighties for
instance. From 1973 to 1992 both elasticities are rather low consistently with a higher, on
average, level of income in those years. The LRE starts to increase again from 1992 onwards,
a period of austerity and restrictive economic policy in Italy. The evolution over time of the
price elasticities of demand seems thus to be consistent with the predictions of the rational
addiction model.

4.4

The Myopic Addiction Model

Myopic models of addiction fail to consider the impact of future consumption on current
consumption, i.e. they are entirely backward looking. A myopic demand function for an
addictive good results from the solution of a static one period utility maximization problem
where utility at time t depends on past consumption of the addictive good, as well as on
current consumption, income and current events. The source of dynamic behaviour is the
stock of habits: St = (1 )St1 + Ct1 where is the rate at which consumption habits
decay. Assuming a 100% rate of decay ( = 1) reduces the stock of habits to last period
consumption and the solution to the static maximization problem produces a linear equation
to be estimated of the form:

Ct = + 1 Ct1 + 2 Pt + 3 Yt + 4 T + ut

(6)

where the subscript t denotes the time and where we have added the disposable income
Y and a time trend T . The coecient on the lagged dependent variable, in this restricted
model, can also be interpreted as habit persistence when the good in question is an addictive
good. The pioneering empirical work on habit persistence demand models is by Houthakker
and Taylor (1970) who estimated single equation dynamic demand functions for many goods.
This model can be estimated by OLS, because it takes the lagged dependent variable as given,
that is, as being uncorrelated with the equations error terms. In this case OLS estimation is
consistent (Verbeek, 2000, p. 280).
15

Table 7 shows the estimated coecients of the myopic model of addiction, both in levels
and in dierences, estimated as an alternative way of modeling alcohol demand. Although the
lagged consumption coecient is statistically significant in both models, the price coecient
turns out to be positive, in the model in levels, and it is negative but statistically insignificant,
in the dierenced model. These data, therefore, do not seem to support the myopic model of
addiction.

Conclusion

This study is mainly concerned with confronting the rational addiction hypothesis with Italian data. To our knowledge this is the first attempt at providing such evidence on alcohol
consumption. In the year 2000 the average per capita consumption of pure alcohol in Italy
was about 7.5 litres, but according to the WHO for the European Region, the target of 6 litres
per capita should be reached by the years 2015. Since the rational addiction model implies
a long run price elasticity of demand higher that alternative models of addiction, providing
such evidence might be of help when choosing policy measures aimed at reducing alcohol
consumption.
The demand for alcoholic beverages in Italy is modeled according to the rational addiction
theory by Becker and Murphy and using a long time series of data on per capita expenditure of
an aggregate of alcoholic beverages. We estimate a dierenced model, after having checked for
price exogeneity and autocorrelation of the residuals as recommended by Auld and Grootendorst (2004). Moreover, we do not impose a specific value of the discount rate as a constraint
on the model. Although alcohol consumption is not supposed to be as addictive as Tobacco,
for instance, our data fit the model quite well. The future consumption coecient is positive,
statistically significant and lower than the past consumption coecient, as predicted by the
theory and we therefore get a reasonable value for the intertemporal rate of time preference.
The myopic model, on the other hand, does not produce results as satisfying.
We also calculate short and long run price elasticities of demand as well as the income
elasticity of demand. Although the long run price elasticity of demand, calculated at the
sample mean, is substantially higher than the short run one, as predicted by the RA model,
the demand for alcoholic beverages is rather unelastic. Thus, even though alcohol consumers
turn out to be actually forward looking, a price increase would not be particularly eective
in reducing alcohol demand even in the long run: a 1% permanent price change in all time
periods would produce, according to our results, a decrease in consumption of about 0.5%.
Surprisingly, the long run income elasticity of demand at the sample mean is about 2 implying
16

that a permanent income change of 1% in all time periods causes the demand for alcoholic
beverages to increase by more than 2%. Alcoholic beverages are therefore to be considered a
luxury good the demand for which increases more than proportionally following an increase in
income. These results, however, are to be taken with caution, because the data do not allow to
separate wine, beer and spirits. Moreover, the distribution of income is not taken into account
here. Thus, alcohol could be a luxury good at low levels of income, but a necessary or inferior
good at high levels of income.

References
[1] Auld, M. C. and P. Grootendorst (2004) An Empirical Analysis of Milk Addiction. Journal
of Health Economics (Forthcoming).
[2] Baltagi, B. and J. Grin (2002) Rational Addiction to Alcohol: Panel data analysis of
liquor consumption. Health Economics, 11, pp. 485-491.
[3] Becker, G. S. and K. Murphy (1988) A Theory of Rational Addiction. Journal of Political
Economy, 96, pp. 675-701.
[4] Becker, G. S., M. Grossman, K.M. Murphy (1991) Rational Addiction and the Eect of
Price on Consumption. The American Economic Review, vol. 81, n. 2, pp. 237-241.
[5] Becker, G., M. Grossman and K. Murphy (1994) An Empirical Analysis of Cigarette
Addiction. American Economic Review, 84, n. 3, pp. 396-418.
[6] Bentzen, J., T. Eriksson and V. Smith (1999) Rational Addiction and Alcohol Consumption: Evidence from the Nordic Countries. Journal of Consumer Policy, 22, pp. 257-279.
[7] Cook, P. J. and M.J. Moore (2000) Alcohol. In: Handbook of Health Economics, Vol.
1B, edited by A.J. Culyer and J.P. Newhouse. Elsevier Science, 2000.
[8] Davidson, McKinnon (1993). Estimation and Inference in Econometrics. Oxford University Press, Oxford.
[9] Escario, J. J., J. A. Molina (2001) Testing for the Rational Addiction Hypothesis in
Spanish Tobacco Consumption. Applied Economics Letters, 2001, n. 8, pp. 211-215.
[10] Grossman, M., F. Chaloupka and I. Sirtalan (1998) An empirical analysis of alcohol
addiction: results from Monitoring the Future Panels. Economic Inquiry, 36, pp. 39-48.

17

[11] Houthakker, H. S., L. D. Taylor (1970) Consumer Demand in the United States, 19291970: Analysis and Projections, 2nd Edn. (Harvard University Press, Cambridge, MA).
[12] Klingemann, H. (2000) Alcohol and its social consequences - the forgotten dimension.
Copenhagen, WHO Regional Oce for Europe, 2000 (keynote paper for the WHO European Ministerial Conference on Young People and Alcohol, Stockholm, 19-21 February
2001).
[13] Ministero della Salute (2003) Relazione del Ministro della Salute al Parlamento sugli
Interventi Realizzati ai Sensi della Legge 30.3.2001 n. 125 Legge Quadro in Materia di
Alcol e Problemi Alcolcorrelati.
[14] Rehn, N., R. Room and G. Edwards (2001) Alcohol in the European Region - consumption, harm and policies. WHO Regional Oce for Europe, 2001.
[15] Scafato, E., S. Ghirini and R. Russo (2004) I consumi alcolici in Italia. Report 2004 sui
consumi e le tendenze (1998-2001). MIMEO, Istituto Superiore di Sanit; Osservatorio
Fumo, Alcol e Droga.
[16] Scafato, E., R. Russo (2004) La donna e lalcol. Tendenze nei consumi e strategie di
intervento. Annali dellIstituto Superiore di Sanit, vol. 40, n. 1, 2004.
[17] Verbeek, M. (2000) A Guide to Modern Econometrics. John Wiley and Sons Ltd, West
Sussex, England.
[18] Wangen, K. R. (2004) Some Fundamental Problems in Becker, Grossman and Murphys
Implementation of Rational Addiction Theory. Discussion Papers n. 375, April 2004;
Statistics Norway, Research Department.
[19] Waters, M. and F. A. Sloan (1995) Why do people drink ? Applied Economics, 27, pp.
727-736.

18

Table 1: Share of Alcohol Consumers in Italy

1998

1999

2000

2001

% Var. 98-01

83.4
Male
Female 58.6

83.1
59.1

87.2
63.6

87.7
63.1

5.2
7.7

Source: Scafato, Ghirini and Russo, 2004, p. 2

Table 2: Average per capita Consumption of Alcoholic

Beverages and Pure Alcohol in Italy (Litres/Year)

Wine
Beer
Spirits
Pure Alcohol

1981

1991

2000

% Var. 81-00

86.2
17.9
3.5
11.7

62.1
24.9
2.5
9.1

51
28.1
1.2
7.5

-40.8
+57
-65.7
-35.9

Source: Scafato and Russo, 2004, p. 4

19

Table 3: Definitions, Means (M) and Standard Deviations (SD) of Variables

Variable

Definition, Mean (M), Standard Deviation (SD)

Ct = (ALQt /N 14)100

Per capita (of persons aged >14 ) alcohol expenditure

per year (1995=1); (M=1.351; SD=0.241)
Real price of alcoholic beverages (1995=1)
(M=1.139; SD=0.161)
Real per capita disposable income per year (1995=1)
(M=0.783; SD=0.234)
Aggregate expenditure in alcoholic beverages, at constant
1995 prices (millions Euro); (M=5417.678, SD=837.055)
Aggregate expenditure in alcoholic beverages
at current prices (millions Euro); (M=2577.546; SD=1882.417)
Consumer Price Index (CPI) (1995=1)
(M=0.481; SD=0.406)
Households final consumption expenditure per year at constant
1995 prices (millions Euro); (M=391066.923; SD=145025.636)
Population older than 14 (in millions units)
calculated in the middle of each year; (M=44.050; SD=3.751)

Pt = (ALVt /ALQt )/P Gt

Yt = Y 95t /N 14
ALQt
ALVt
P Gt
Y 95t
N 14

20

Table 4: Diagnostic Tests on Time Series (p-values in parentheses)

1. Price
exogeneity
1. LR(HW) test
2a Wtd. Sym.
2b Dickey-Fuller

2. Unit Root
C

3. Autocorrelation
(dierenced model)

0.793
(1.000)
-1.515
(0.887)
-1.233
(0.903)

-0.798
(0.985)
-2.251
(0.461)

3a Durbins h Alt.

-1.166
(0.957)
0.567
(0.997)
-2.327
(0.020)
5.767
(0.056)

3b Breusch-Godfrey

21

Table 5: GMM Estimates of the Rational Addiction Model

Dependent variable Ct (standard errors in parentheses)
Dif f erences
Ct1

0.392
(0.083)*
0.322
(0.113)*
-0.161
(0.089)*
1.073
(0.336)*

Ct+1
Pt
Yt

Constant

-0.025
(0.007)*
8.379
0.592
0.314
37

Test of overident. restrict.

p-value
R2 (adj.)
n

*= standard errors are both heteroscedastic consistent and are also robust to autocorrelation:
the disturbance terms are sp ecified as a second order moving average process.

Table 6: Short and Long Run Elasticities of demand

at the sample mean (standard errors in parentheses)
SRE

LRE

r(%)

Price elasticity

-0.256
(0.190)

-0.474
(0.385)

21.952

Income elasticity

1.173
(0.205)

2.173
(0.596)

Standard errors have been computed using the Delta method.

Elasticities have b een computed from the parameters of the RA dierenced model;

22

Table 7: OLS Estimates of the Myopic

Model (standard errors in parentheses)

Ct1
Pt
Yt
T
Constant
R2 (adj.)
n

Levels

Dif f erences

0.818
(0.047)
0.104
(0.063)
1.178
(0.334)
-0.025
(0.006)
-0.257
(0.144)
0.976
42

0.337
(0.012)
-0.032
(0.207)
1.649
(0.423)
-0.035
(0.012)
0.344
41

Standard Errors are heteroscedastic consistent;

Table 8: Short and Long Elasticities of demand

over sub-periods (standard errors in parentheses)*
SRE

LRE

1963 1972

-0.27
(0.207)

-0.52
(0.419)

1973 1982

-0.22
(0.165)

-0.42
(0.334)

1983 1992

-0.21
(0.154)

-0.38
(0.311)

1993 1999

-0.31
(0.232)

-0.58
(0.470)

* Standard errors for elasticities have b een computed using ANALYZ in TSP45.
Elasticities have b een computed from the parameters of the RA dierenced model

23