Increase in the price of tobacco products in France from 2018 to 2020 : a forecast of the economic impact and the effect on tobacco consumption

(1)

1 MSc Economics: Public Policy

2017-2018

Increase in the price of tobacco products in France

from 2018 to 2020: A forecast of the economic

impact and the effect on tobacco consumption

MASTER THESIS – MSc ECO

by Maxime ROCHE (11769254) – maxime.roche@edhec.com under the supervision of Prof. Hessel OOSTERBEEK

second reader Prof. Erik PLUG

(2)

2

STATEMENT OF ORIGINALITY

This document is written by student Maxime Roche who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

ACKNOWLEDGMENT

I would like to thank my thesis supervisor Prof. Hessel Oosterbeek for his availability and invaluable advices and comments throughout the realization of this thesis.

In addition, I also thank Prof. Ward Romp and Prof. Massimo Giuliodori for useful preliminary comments.

Finally, I would like to thank the director of our Research Seminar for Public Policy, Prof. Jeroen Hinloopen.

(4)

4

ABSTRACT

This thesis forecasts the effects on tobacco consumption, the prevalence of smoking, tax revenues and the purchasing power of the new tobacco price increase policy implemented by the French government in end of 2017. This policy consists of a gradual increase in the price of cigarettes by €1 per year from 2018 to 2020 through increases in excise taxes. It is meant to contribute to the reduction of tobacco consumption. However, rising cigarette prices tend to increase the consumption of cheaper substitute products such as roll-your-own tobacco or illicit cigarettes.

Country fixed effects and year effects panel data regressions using cross-country data from the Euromonitor and WHO databases, and autoregressive distributed lag regressions employing monthly French aggregate level time series data from the French Observatory of Drugs and Drug Addiction (OFDT), are used to estimate the price elasticities of the demand for licit cigarettes and the prevalence of smoking, and the cross price elasticities of the demand for illicit cigarettes and roll-your-own tobacco.

The estimated aggregate level impact of the new policy is a decrease of approximately 12.5% in the quantity of tobacco consumed and of 0 to 6 percentage points in the prevalence of smoking. The high estimated impact on the consumption of licit cigarettes is mitigated by a substantial expected increase in the consumption of roll-your-own tobacco and illicit cigarettes. On the other hand, tax revenues from tobacco products are expected to rise with an estimated revenue of €16.4bn in 2020.

Finally, this thesis analyzes the consumption of cigarettes by income level and its evolution over time between 1995 and 2010 using the “Budget des Familles” surveys from INSEE. The results show that this policy is expected to have a low impact on the purchasing power of the lowest income deciles. Nevertheless, this last result is underestimated as it includes only the impact of the policy on the consumption of licit cigarettes, omitting the expected increase in the consumption of substitute products due to data unavailability at the microeconomic level.

(5)

5

TABLE OF FIGURES

Table 1 – Summary statistics Part III 17

Table 2 – Summary statistics Part IV 26

Figure 1 – Evolution of monthly sales (in millions of sticks) and price of the most sold pack of 20 cigarettes (in euros) between 2000 and 2017 27

Figure 2 – Evolution of monthly sales of roll-your-own tobacco (in tons) and price of the most sold pack of 20 cigarettes (in euros) between 2000 and 2017 28

Table 3 – Quantity of cigarettes per adult (age 15+) per day calculated with controls from the regression (yes or no) and based on macroeconomic data (yes or no) 35

Table 4 – Ratios of cigarette expenses on income per decile calculated from the regression with controls and based on macroeconomic data 36

Table 5 – Forecast of the 2020 aggregate tobacco consumption 38

Table 6 – Forecast of the 2020 aggregate prevalence 39

Table 7 – Forecast of the 2020 tax revenue 39

Table 8 – Forecast of the impact on the purchasing power per decile 41

(7)

7

TABLE OF APPENDICES

Appendix 1 – France yearly licit cigarette sales and illicit sales (in millions of sticks) as a function of the average price per pack of 20 cigarettes of the most sold brand (€) 51 Appendix 2 – France yearly roll-your-own sales (in tons) as a function of the average

price per pack of 20 cigarettes of the most sold brand (€) 51

Appendix 3 – Price and tax correlation 52

Appendix 4 – Description of the dataset for the cross-country analysis 53

Appendix 5 – Data source and variables description 54

Appendix 6 – Correlation graphs, illicit, licit and prevalence for 2016 55

Appendix 7 – Cigarette prices and illicit trade volume, 2008-2016 56

Appendix 8 – Country-demeaned cigarette prices and illicit trade, 2008-2016 57

Appendix 9 – Licit cigarette consumption – Pooled panel data 57

Appendix 10 – Licit cigarette consumption – FE 58

Appendix 11 – Licit cigarette consumption – IV 58

Appendix 12 – Illicit cigarette consumption – Pooled panel data 59

Appendix 13 – Illicit cigarette consumption – FE 59

Appendix 14 – Illicit cigarette consumption – IV 60

Appendix 15 – Prevalence total – Pooled panel data and FE 60

Appendix 16 – Prevalence total – IV 61

Appendix 17 – Stationarity checks for monthly licit cigarette sales and roll-your-own at the Macroeconomic level for France 61

Appendix 18 – Licit cigarette consumption and roll-your-own PED estimations 62

Appendix 19 – Expenditures and daily quantities smoked per household per decile and comparison to aggregate data 63

Appendix 20 – Revenue per decile (in 2010 euros) 64

Appendix 21 – Annual expenditure on cigarettes per household per decile (in 2010 euros) calculated with controls (yes or no) 64

Appendix 22 – Cigarette consumption per day per adult (aged 15+) in time with controls and based on macroeconomic data 65

Appendix 23 – Ratios of cigarette expenses on income per decile calculated with controls (yes or no) and based on macroeconomic data (yes or no) 65

Appendix 24 – Ratios of cigarette expenses on income by decile for each survey calculated from regression with controls and based on macroeconomic data 66

Appendix 25 – Ratios of cigarette expenses on income per survey for each decile calculated from regression with controls and based on macroeconomic data 66

(8)

8

I. INTRODUCTION

A. The tobacco epidemic

The tobacco epidemic is one of the most important public health threats humanity has ever faced. In the world today more than 1.1 billion people are current smokers. Smoking tobacco is a source of heart disease, strokes, and several cancers. It is responsible for premature illness and death. Each year, more than six million people die worldwide due to direct consumption of tobacco, and approximately 890,000 from exposition to second-hand smoking. According to Goodchild et al. (2017), smoking accounted for about 6% of global health spending and 2% of the worldwide gross domestic product in 2012. In addition, tobacco cultivation harms the environment due to the intensive use of fertilizer and pesticides, and the countless non-biodegradable cigarette butts strewing cities’ sidewalks.

In the past decades, a global effort has been put into controlling tobacco. Demand-reduction measures such as tobacco taxation, smoke-free policies or marketing bans have been implemented around the world (The Tobacco Atlas (2018)). These measures are significantly associated with lower smoking prevalence. Several studies concluded that the implementation or the increase of tobacco taxes are the most effective policies to reduce tobacco consumption as it represents an incentive for smoking cessation and participate in preventing initiation (Warner (2006) and Centers for Disease Control and Prevention (2007)). It leads to significant and immediate health and revenue benefits. The mechanism is quite simple. When a sufficiently large increase in the excise tax on tobacco products is implemented, it raises the prices of tobacco products making them less affordable. Most countries focus on raising the price of cigarettes as it is the most consumed tobacco product. However, increases in cigarette prices tend to lead to increases in the consumption of smuggled tobacco products, roll-your-own tobacco or smokeless products and e-cigarettes. Indeed, the price of these products per gram of tobacco is in general lower than the one of cigarettes. Therefore, the efficiency of such policies on consumption is reduced when these variations in the consumption of substitute products are taken into account. In addition, higher tobacco prices are especially effective in reducing tobacco consumption for vulnerable populations, such as low-income households, as they are particularly sensitive to price increases. As a consequence, an increase in tobacco prices harms the purchasing power of people in different ways according to their position in the income distribution.

(9)

9 B. The situation in France and the new policy implemented

In France, in 2016, around 18.5% of male deaths were due to tobacco use and 29.8% of the population was daily smoking1_{. According to Goodchild et al. (2017), in 2012, the health and}

social costs of tobacco were estimated to amount to approximately €120 bn. In the past decades, successive policies were implemented by the French government to help reducing tobacco consumption, such as banning smoking in public places and children’s play areas, the neutral pack of cigarettes, or the tripling of the reimbursement of nicotine substitutes such as patches. In addition, the government increased taxation on tobacco products several times. As can be seen in Appendix 1 and Appendix 2, while the price of cigarettes increased, the consumption of licit cigarettes decreased. In the meantime, the consumption of illicit cigarettes and roll-your-own tobacco rose.

In France, tobacco taxation consists of the excise tax (“Droit de Consommation”) and the VAT. The total amount of the excise tax cannot be lower than a minimum threshold of perception, which plays the role of a minimum tax expressed in euros (see Appendix A for more details about the taxation of tobacco products in France).

Although absent from Emmanuel Macron’s presidential program during the 2017 election, a rise in tobacco taxation was incorporated into the annual social security financing bill for 2018. The government has decided to launch a gradual price increase policy through increases in the minimum threshold of perception applied to tobacco products. By raising this minimum, the government will increase the price of packs of 20 cigarettes by approximately €1 per year from 2018 to 2020. The price of a pack of the most sold brand in France (Marlboro Red) being on average €7 in 2017, this increase will lead to a price of €10 by the end of 2020. The Ministry of Health specified that the rise in prices will also affect cigars, cigarillos, and packs of roll-your-own tobacco. I refer readers to the Appendix B for a more detailed explanation of the policy.

C. Research question and motivation

There is little doubt on the fact that this new policy will lead to a decrease in licit cigarettes consumption, however, the size of this decrease, the impact on illicit cigarettes or roll-your-own tobacco consumption, smoking prevalence, tax revenues or the purchasing power of consumers are not known precisely.

1_{The Tobacco Atlas (2018). Sixth Edition}

2_{In French: “Observatoire Français des Drogues et des Toxicomanies” (OFDT)} 3_{In French: “Institut National de la Statistique et des Etudes Economiques” (INSEE)} 4_{In English: “Family Budgets” surveys}

(10)

10 Therefore, my research question is the following: What will be the economic impact and the effect on tobacco consumption of the planned increase in the price of cigarettes in France?

I focus my analysis on cigarette prices as it is the most consumed tobacco product and data on others are not easily accessible or simply unavailable. I analyze the impact of increases in the price of cigarettes on consumption and not directly of the impact of increases in excise tobacco taxes, as the new policy implemented by the French government will mainly raise the minimum threshold of perception and not the proportional tax rate on tobacco products. In addition, as shown in Appendix 3, 85.3% of the variations in prices are indeed explained by variations in taxation in the sample used in my cross-country analysis.

The objective of this thesis is to estimate the price elasticity of the prevalence of smoking, the price elasticity of the demand (PED) for licit cigarettes, and the cross PED for illicit cigarettes and roll-your-own tobacco in order to forecast the impact of the policy at the aggregate and per decile level. It represents an addition to the existing literature as it updates the knowledge on the regressivity of tobacco taxation in France and on estimates of the cross PED for roll-your-own tobacco and illicit cigarettes. In addition, it provides an updated and a rather exhaustive forecast of the policy taking into account the impact on the consumption of substitute products.

The following section presents two theoretical models on tobacco consumption and the current state in the literature over the PED for tobacco products. Section 3 estimates the cross-country PED for licit cigarettes, the price elasticity of smoking prevalence, and the cross PED for illicit cigarettes using a panel dataset built from the Euromonitor and World Health Organization (WHO) databases. Section 4 estimates the PED for licit cigarettes and the cross PED for roll-your-own tobacco for France using a monthly time series dataset from the French Observatory of Drugs and Drug Addiction (OFDT)2_{. Section 5 provides an}

analysis at the microeconomic level in France, working with data from the “Budget des Familles” surveys carried out by the French National Institute of Statistics and Economic Studies (INSEE)3_{, exploring the effect of past price increases on the consumption of licit}

cigarettes per decile in the revenue distribution. The previous estimates are then used in section 6 to forecast the policy in terms of consumption of tobacco products, prevalence rates, tax revenues and purchasing power. Furthermore, section 7 provides a conclusion and a policy discussion. Finally, further sections are dedicated to references and appendices.

2_{In French: “Observatoire Français des Drogues et des Toxicomanies” (OFDT)} 3_{In French: “Institut National de la Statistique et des Etudes Economiques” (INSEE)}

(11)

11

II. LITERATURE REVIEW

The French government regularly raises the price of cigarettes through tax increases. Following an economic logic, cigarette taxation is a disincentive in terms of consumption because of the rising purchase cost and future socio-economic perspectives that may influence the decision to smoke or the level of consumption, especially in times of economic crisis. Consumers are addicted to tobacco because of the pleasure of smoking or socio-cultural criteria of belonging. Therefore, tobacco belongs to the category of goods known as addictive goods. Some theoretical models integrate this phenomenon of dependency by taking into account past consumption and consumption habits in a logic of maximizing satisfaction and budget allocation according to preferences. For example, the model of decisions by Becker and Murphy (1988) is based on a utility function that depends on current and past consumption of a dependency good. Following this life cycle model, a consumer plans his consumption over his entire life in order to maximize his intertemporal utility. Any increase in the price is then bad for him (even if it reduces his consumption). Only a compensation for the damage it causes to others (e.g. in the form of passive smoking) can thus justify the existence of taxes. In another model, however, Gruber and Köszegi (2002) suggest that tobacco consumption does not follow a "rational" scheme but a pattern of temporal incoherence as defined by Laibson (1997). Under this model, taxes would, under certain conditions, allow consumers to get out of the mechanism of overvaluation of current consumption and would actually increase the intertemporal utility of potential consumers.

Smoking is a public health issue in which the political and economic dimension is omnipresent (Etilé (2006)), mainly because of the taxes collected by the government. For example, tobacco taxation amounted to €14bn in France in 2013 (source: OFDT). The danger of smoking justifies and legitimizes the intervention of the public authorities, especially when the “rational” consumer hypothesis is put into question with the principle of delaying the realization of risk in the future. In this case, the public authorities give more weight to the future through prevention policies or price increases (Grignon and Pierrard (2004)). Increasing prices implies a decrease in the consumption of this dependency good and of well-being. These increases represent a relatively important source of income for the government, which allows, in particular, to deal with the perverse effects of smoking such as health costs.

(12)

12 Many studies have found that imposing excise taxes on cigarettes is an effective way to reduce consumption. In this paper, I am particularly interested in estimating the price elasticities of tobacco consumption in order to estimate the expected impact of the new policy to control tobacco in France.

First, in terms of prevalence, Ahmad and Franz (2008) employ a dynamic computer simulation model after estimating the price elasticity of smoking prevalence using survey data across states and weighted ordinary least-squares. They conclude that if the United States were to impose an excise tax rate of 40% on cigarettes, it would decrease smoking prevalence by 15.2%. Furthermore, the tax would generate US$365bn in new revenue. Using a cross-country dataset and a panel threshold regression, Hsieh et al. (2014) present evidence that cigarette prices and prevalence are negatively associated and that this association is stronger in lower-income countries with a price elasticity comprised in between -0.72 and -0.22.

For the estimation of the PED for licit cigarettes, Kostova et al. (2014) find an estimate of -0.53 (a 10% increase in price leads to -5.3% in consumption) for low- and middle-income countries. This result is based on a one-year nationally representative survey over 15 countries from GATS (Global Adult Tobacco Survey), tracking key tobacco control indicators. Another study from Gallus et al. (2006), also using data for a single year, finds a PED estimate of -0.4 for European Union (EU) countries and -0.8 for non-EU. This paper is based on data from the WHO for 52 European countries in 2000. However, these two previous papers do not investigate changes in consumption over time as Watkins (2015) does with a dataset containing 195 countries from 2008 to 2012. Using Ordinary Least Squares (OLS) and quantile regressions, he investigates differences in the PED accounting for variations in the relationship between smoking prevalence, price, and income level. His results show an average PED of -0.15 with a higher PED for high-income than low-income countries. The main finding of this paper is that countries where smoking prevalence is very high or very low are less price-sensitive than moderate consumption countries.

As I want to forecast the effect of an increase in price in France, the use of time series data could lead to a more accurate estimate of the PED for licit cigarettes. Anguis and Dubeaux (1997) calculate a PED of licit cigarettes of -0.3 using a VECM model controlling for household income, consumption, and inflation. This model is estimated on quarterly data for the period 1976-1995. More recently, Fromentin (2015) uses monthly time series data from OFDT to estimate the PED for licit cigarettes in France. He employs unit root tests with endogenous break and a structural break test over two periods, controlling for sales of stop-smoking medicines and the unemployment rate, as treatment and economic conditions could

(13)

13 influence cigarette consumption. The results show that the government should implement a strong increase in prices to obtain concrete and fast effects, as the PED during the high price increase period from 2000 to 2003 is of -0.45 compared to -0.18 for the afterward moderate price increase period.

Less attention has been given to the cross PED for illicit cigarettes in the literature, mainly because of the questionable reliability of the rare estimates of the illicit market share. Nevertheless, Prieger and Kulick (2018) estimate that a rise of €1 in the price of licit cigarettes would induce an increase of 25% to 125% in the consumption of illicit cigarettes for EU countries. They use a panel dataset from the Euromonitor Passport database for the EU from 1999 to 2013 and employ fixed effects regressions controlling for the Gross Domestic Product (GDP) and a corruption index. In order to account for the possible endogeneity of the price of licit cigarettes, they instrument it with the labor tax and the VAT. Finally, tobacco taxes, like all taxes, participate in the redistribution system. This aspect of tobacco taxes seems to have been little studied for France. However, different studies, for example Townsend et al. (1994) for the UK, show that low-income deciles are the most affected by tobacco-related diseases in developed countries. The assumption that tobacco consumption decreases with income, if it were to be verified, would make taxes on these products regressive, that is, taxes whose average rate by income category is decreasing with income. From the "Budget des Familles"4 surveys carried out by INSEE, Godefroy (2003) analyzes the evolution of household cigarettes consumption according to income level from 1978 to 2000. The results show a very strong regressivity of taxes on tobacco. For each of the years studied, this strong regressivity is characterized by the fact that the level of consumption per adult is itself decreasing according to the level of income. Moreover, this regressivity has not diminished over time. In a note from the French Observatory of Economic Conditions (OFCE)5_{, Madec (2017) forecasts the effect of the new policy on the}

purchasing power of each decile. Assuming a price elasticity of -0.3 as estimated by Besson (2006) and using Dauvergne (2012)’s estimated average amount of tobacco taxes paid per decile of standard of living in 2010, he estimates that the 10% most modest households should see their standard of living being reduced by €195 per year, which represents a loss of 2.4%. On the other hand, households with a standard of living above the median should experience a cut of less than 1% of their standard of living.

4_{In English: “Family Budgets” surveys}

(14)

14

III. CROSS-COUNTRY ANALYSIS

In this section, I estimate – using a cross-country panel dataset – the PED for licit cigarettes, the price elasticity of the prevalence of smoking, and the cross PED for illicit cigarettes. These estimates are used in the forecasting of the effects of the policy on consumption and prevalence in Part VI.

A. Data

1. Source

Data for the cross-country analysis comes from a variety of sources and covers the years 2008, 2010, 2012, 2014 and 2016.

The main sources of information are derived from the “WHO report on the global tobacco epidemic” published by the World Health Organization (WHO) every two years6_{. These}

reports focus on monitoring tobacco use and prevention policies. They provide an extensive analysis of the situation in each WHO Member States with the Tobacco Control Country Profiles (TCCP). For each country who signed the WHO FCTC treaty7_{, TCCP provides}

information about tobacco prevalence, preventive measures, and tobacco economics.

From the TCCP, I collect data for 34 countries (see Appendix 4 for a list of countries and years) on the following items: the price of a 20 cigarettes pack of the most sold brand (in international dollars 2016) and taxes (specific excise, ad-valorem excise, and VAT). International dollars 2016 is a hypothetical unit of currency with the same purchasing power parity than the U.S. dollar had in the United States in 2016. It is used to allow comparison of prices and GDP between countries and years.

Moreover, in order to detect more accurately the unique influence of price on consumption among the multitude of other tobacco control policies, I use the MPOWER indices from the WHO8_{. These indices, scaled from 0 to 3, help to determine whether or not a given country}

at a given time complies with a benchmark of policies that the WHO recommends implementing on one of the “Monitor”, “Protect”, “Offer”, “Warn H”, “Enforce” or “Raise” policy areas described in the MPOWER report. I do not include “Monitor” policies, as they are

6_{WHO (2009, 2011, 2013, 2015, 2017). “WHO report on the global tobacco epidemic”}

7

The WHO FCTC is a voluntary treaty for political commitment to develop, implement, and enforce interventions to control tobacco. As of end 2017, 181 parties had signed the treaty.

(15)

15 related to tracking and surveying tobacco use, and therefore should not be causally associated with lower cigarette consumption. In addition, I do not incorporate “Raise” policies, as they are directly related to the independent variable (price) in my estimations. Although these indices are an imperfect measure, they serve as proxies for other tobacco control policies. I refer readers to Appendix C for more information on the indices and to the Methods section of the TCCP for details on sources of data9.

For consumption data, the main source used is the Euromonitor Passport database, a global market research database, which provides insight on industries, economies, and consumers worldwide. From this database, I collect smoking prevalence rates10_{for the total population,}

aggregate sales volume of cigarettes and illicit trade volume for my sample countries for the period studied (see Appendix 4 for a list of countries and years). Prices for illicit cigarettes are unavailable.

The illegal nature of illicit trade makes it difficult to measure accurately. Euromonitor does not reveal the exact methodology for its estimates of the illicit tobacco market but mentions the use of estimates based on seizures of illicit cigarettes, estimates made by governments or companies from the tobacco industry, and on-the-ground analysis.

As this data source is non-academic, some researchers have questioned Euromonitor’s data on illicit tobacco arguing that it is artificially high. After analyzing the illicit market for tobacco in South Africa and Mexico and comparing their findings with Euromonitor estimates, Blecher et al. (2013) state: “Euromonitor's reliance on tobacco industry intelligence and an opaque modelling process may lead to biased estimates, especially if information provided by industry sources is influenced by their common narrative that increases in excise taxes cause increases in illicit trade”. It may also lead to an overestimation of the correlation between illicit trade and prices of licit cigarettes. Therefore, Blecher et al. (2013) recommend that “investigators exercise increased caution in using Euromonitor data for studies on illicit trade and that new well-documented and verifiable methods are developed to monitor illicit trade that are not dependent on industry data”. Despite this, no other comparable global data exists on illicit cigarettes trade. In addition, as a report for the European Commission noted: "Due to the market’s contentious nature, various parties have vested interests in either deflating or inflating illicit trade figures, though Euromonitor strives to present the most accepted and realistic estimate of the market" (Pedersen et al. (2014)).

9_{WHO (2009, 2011, 2013, 2015, 2017). “WHO report on the global tobacco epidemic”}

10_{Smoking prevalence by Euromonitor includes cigarettes, cigars, pipes or any other smoked tobacco products} for both daily and occasional smokers. Population aged 15+.

(16)

16 Finally, I collect other variables to control for potentially confounding factors associated with licit and illicit trade, and prevalence: a corruption index and the GDP per capita of each country for the period. The GDP per capita (in international dollar 2016) was collected from the World Bank International Comparison Program database. For corruption, I use a measure of control of corruption from the World Bank Worldwide Governance Indicators (WGI). Higher values of this variable – scaled from -2.5 to 2.5 – are associated with less corruption in a country as the measure reflects people’s perception of corruption in their own country.

For more details, please refer to the descriptive statistics in Appendix 5.

2. Summary statistics

Summary statistics for the panel dataset used in this section are shown in Table 1.

The licit sales of cigarettes average 127 billion of sticks and range from 218 million (Iceland in 2014) to over 2,549 billion of sticks (China in 2014). The illicit trade of cigarettes averages approximately 13 billion of sticks and ranges over the years from 74 million (Denmark in 2008) to over 198 billion of sticks (China in 2008). Countries with more than 20% illicit market share for at least two years in the period studied are Brazil, Greece, and Ireland. The countries with less than 5% illicit market share in at least two of the years are Denmark, Italy, Russia, Spain, Switzerland and the USA. In terms of prevalence, the average in the sample is 24% with a minimum of 10.9% (Sweden in 2016) and a maximum of 47.9% (Greece in 2008).

Prices average 6.29 international dollar 2016 per pack and range from 1.66 (China in 2010) to 12.69 international dollar 2016 (Ireland in 2016). Excise taxes average about 55.3% of the retail price; the same countries with high prices have similarly high taxes. The Scandinavian countries, Denmark, Norway, Finland, and Sweden have the highest values of Control of corruption. They are also top countries in terms of GDP per capita along with Switzerland and the US, while Bulgaria, China, Romania, and Russia have the lowest.

(17)

17 Table 1 – Summary statistics Part III

Note: Dataset covers years 2008, 2010, 2012, 2014, and 2016; some years or countries are missing for some variables.

The countries used in my sample represent approximately 71% of worldwide cigarettes consumption.

Appendix 6 presents the illicit, licit and prevalence levels as a function of the price of a pack of licit cigarettes in 2016. A clear negative correlation is observable between the price of a pack and the number of licit cigarettes smoked per adult per day. In addition, a negative correlation is noticeable between the price of a pack of licit cigarettes and the prevalence of smoking. Finally, for the illicit cigarette consumption, the association is not clear.

The correlations found in the pooled data in Appendix 6 do not answer the key question for policymakers: What would be the impact of a rise in cigarette prices on the licit and illicit consumption, and on prevalence? There are many confounding factors across countries and years, and their differences must be held constant to isolate the impact of prices.

Variable Obs Mean Std. Dev. Min Max

Total licit cigarette cales 153 127149.9 421701 217.619 2549500 Total illicit cigarette sales 105 13037.72 34791.27 74 198241

Price per pack 170 6.008 2.158 1.510 12.690

Gdp per capita 170 33943.88 13065.9 7635.073 71472.3

Control of corruption 170 0.960 0.918 -1.115 2.393

Total excise tax 170 0.553 0.113 0.151 0.758

Vat Protect Offer Warnh Enforce Prevalence total 170 151 170 170 170 105 0.166 0.980 2.129 1.618 1.788 24.038 0.039 1.208 0.621 0.800 0.808 8.050 0.048 0 1 0 0 10.9 0.250 3 3 3 3 47.9

(18)

18 B. Methodology

In this section, I present the econometric methodology used to estimate the cross-country PED for licit cigarettes, the price elasticity of the prevalence of smoking, as well as the cross PED for illicit cigarettes.

For each of these estimations, the regression has the following form:

ln(Y!") = 𝛽!+ 𝛽!ln 𝑃!" + 𝜶𝐖𝒊𝒕+ 𝜀!" (1)

I first regress using the pooled panel data. Then, in order to correct for omitted variable bias, I introduce additional controls 𝐖𝒊𝒕 such as GDP per capita, tobacco control policy

compliance MPOWER indices, or corruption.

In a second time, I control for unobserved differences among countries that may affect sales or consumers’ behaviors by adding fixed effects. For example, in the estimation of the cross PED for illicit cigarettes, there are potential confounding factors, such as resources devoted to law enforcement, attitudes toward participation in illicit markets, or income which vary across countries. In order to illustrate the idea behind the country fixed effects econometric approach, consider the data in Appendix 7 without controlling for country fixed effects. In Appendix 8 the within-country variation in prices and sales of illicit cigarettes is untainted by any confounding factors that vary among countries but not over time. The positive correlation for the demeaned variables is obvious. While Appendix 7 shows that the levels of prices and sales of illicit cigarettes have a little obvious relationship, Appendix 8 clearly indicates that increases in licit cigarette prices are associated with rising sales of illicit cigarettes. Some additional regressors will be added to control for possible confounding factors that change over time as well as across countries, such as corruption, income, and factors making prices potentially endogenous.

Even if the possible confounding factors vary much more across countries than over time, there may be also trends across all countries that lead to a spurious correlation between prices and sales or prevalence even after controlling for unobserved country factors. For example, for the estimation of the cross PED for illicit cigarettes, the general trends for prices and sales of illicit cigarettes are upward during the period studied. If the latter is actually caused by factors other than increasing prices, then the association between prices and sales of illicit cigarettes may emerge from a spurious correlation of trending variables. Therefore, year dummies are also included in the regression to control for year effects.

(19)

19 Finally, I employ instrumental variables (IV) regressions as the price coefficient might still be potentially biased. Indeed, the price variable used in the different regressions is measured as an aggregate average for each country and therefore might be subject to measurement error, which could lead to attenuation bias in the price coefficient11_{. In addition, there is a}

possible reverse causality between sales of licit or illicit cigarettes and prices. Indeed, the price of a pack of licit cigarettes is influenced by the minimum threshold of perception decided by the government. This minimum could be increased in response to an increase in sales of licit cigarettes or decreased after lobbying from the tobacco industry arguing that increases in prices are responsible for the rising share of the illicit market12_{. This would}

directly affect the price of licit cigarettes. However, decreases in prices are not observed in the sample over the period studied as prices generally rose or stayed constant in each country.

Therefore, country fixed effects and year effects IV regressions are performed in order to control for the possible endogeneity. The price of licit cigarettes is instrumented using the total excise tax as a percentage of the retail price, as it is common to instrument cigarette prices with excise taxes (eg. Stehr (2005) or Nonnemaker et al. (2009)). The relevance of this instrument is obvious as can be seen in Appendix 3. I also add another instrument: the VAT tax, as a fraction of the retail price. This instrument is positively correlated with the retail prices of all goods and therefore of cigarettes. In addition, it is not affected by changes in cigarette prices and exogenous to licit or illicit demand shocks.

1. Licit sales of cigarettes

The model used to estimate the PED for licit cigarettes follows Watkins (2015). In equation (1), 𝑌_!" refers to sales of licit cigarettes per capita13_{per day in number of sticks for country i}

at year t. The independent variable 𝑃_!" is the average price of licit cigarettes per pack in 2016 international dollar. Finally, the control vector 𝐖𝒊𝒕 is made of the per capita GDP in 2016

international dollar, and the four indices from the WHO’s MPOWER indicators – “Protect”, “Offer”, “Warn H”, and “Enforce” – to account for other policies aiming to control tobacco consumption.

11_{The price variable is calculated as an average at the aggregate level and is not accurate for any given} transaction.

12_{Joossens et al. (2000) mention that Canada and Sweden reduced their taxes on tobacco products in the 1990s} because of concern about increased illicit market share. Canada is not in the dataset and the tax reduction in Sweden happened in 1998, before the period studied.

13_{Due to the incapacity to find data on the population aged 15+ per country for each year for most of the} countries in my sample, I divide the aggregate licit cigarette consumption by the size of the entire population for each country and each period. Therefore, the dependent variable is undervalued.

(20)

20 Results of the different regressions to estimate the PED for licit cigarettes can be found in Appendix 9 for the pooled panel data regressions, Appendix 10 for the fixed effects and year effects regressions, and Appendix 11 for the IV regressions.

2. Illicit sales of cigarettes

In order to estimate the cross PED for illicit cigarettes, I follow the model of Prieger and Kulick (2018). In equation (1), 𝑌_!" refers to sales of illicit cigarettes in millions of sticks for country i at year t. The independent variable 𝑃!" is the average price of licit cigarettes per

stick in 2016 international dollar. Finally, the control vector 𝐖𝒊𝒕 is made of the per capita

GDP in 2016 international dollar, the square of the latter, and the measure of corruption from the World Bank WGI. The price of illicit cigarettes is theoretically lower than the price of licit cigarettes. Therefore, the price differentials represent a source of profit from smuggling. However, the appeal of the illicit market depends on enforcement, in other words, the probability and consequences of being caught. Some analysts find that prices weigh heavily on the incidence of smuggling (Goel 2008), while others like Merriman et al. (2000) argue that the level of corruption explains more of the variance in the consumption of illicit cigarettes than licit cigarette prices. Therefore, the perceived level of corruption is incorporated as a control in the model.

Results of the different regressions to estimate the cross PED for illicit cigarettes can be found in Appendix 12 for the pooled panel data regressions, Appendix 13 for the fixed effects and year effects regressions, and Appendix 14 for the IV regressions.

3. Prevalence of smoking

The model used to estimate the price elasticity of the prevalence of smoking employs equation (1), in which 𝑌_!" refers to the prevalence of smoking for the total population aged more than 14 years old in country i at year t. The independent variable 𝑃_!" is the average price of licit cigarettes per pack in 2016 international dollar. Finally, 𝐖_𝒊𝒕 is the per capita GDP in 2016 international dollar, and the four indices from the WHO’s MPOWER indicators: “Protect”, “Offer”, “Warn H”, and “Enforce”. Indeed, as for the estimation of PED for licit cigarettes, I control for other policies aiming to control tobacco consumption to isolate the effect of an increase in price.

(21)

21 Results of the different pooled panel data, fixed effects, and year effects regressions to estimate the price elasticity of the prevalence of smoking can be found in Appendix 15. In addition, results of the IV regressions are displayed in Appendix 16.

C. Results

In this section, I present the results of the different regressions derived from equation (1). No comments are given on the results of the IV regressions for the estimations of the PED for licit and the cross PED for illicit cigarettes, as the coefficients of interest are mostly insignificant or too high to reflect the reality. Nevertheless, the results of these IV regressions are displayed respectively in Appendix 11 and Appendix 14.

My approach has several important limitations. I use an aggregate cross-section dataset, hence individual-level trends among smokers may be different from those I measure. In addition, the control variables for the presence of non-price tobacco policies are relatively crude.

1. Licit sales of cigarettes

The pooled panel data regressions, although not controlling for country fixed effects, allow me to check the sign of the price coefficient. This sign is negative as expected, as can be seen in Appendix 9.

After controlling for unobserved differences among countries, I estimate, in Appendix 10, the cross-country average PED for licit cigarettes to be -0.563 (regression 1.4). This means that a 10% increase in the price per pack is associated, on average, with a 5.6% reduction in cigarettes consumed per capita. However, this result does not take into consideration how countries with different levels of income might respond differently. Hence, I estimate equation (1) again but restricting the sample to high-income countries14. As France is classified as a high-income country, this new regression uses a more representative sample for the future forecast in Part VI. The results (regression 1.6) give a cross-country average PED for licit cigarettes of -0.673 in high-income countries. This shows that the population in those countries is more responsive to increases in prices of cigarettes than middle-income countries, at least in my sample, which is in line with the results of Watkins (2015).

(22)

22 In addition, the results show significant negative coefficients for the “Protect” and “Warn H” policy indices. This demonstrates that smoke-free policies and health warnings participate in reducing tobacco consumption.

Nevertheless, when including year effects, the results show significant negative coefficients on the year dummies, and lower and not significant PEDs for licit cigarettes (regressions 1.5 and 1.7). Hence, the PEDs for licit cigarettes obtained using only the country fixed effects specification are influenced by decreasing aggregate trends which are not associated with the causal relationship of price on consumption.

2. Illicit sales of cigarettes

The sign of the price coefficient is positive for the pooled panel data regressions, as can be seen in Appendix 12. Although not controlling for country fixed effects, this shows the positive correlation between the price of licit cigarettes and illicit cigarettes consumption. In addition, the sign and significance of the corruption variable coefficient highlight the expected negative correlation between corruption and illicit cigarettes trade.

After controlling for unobserved differences among countries, I estimate, in Appendix 13, the cross-country average cross PED for illicit cigarettes to be 0.499 (regression 2.5). This means that a 10% increase in the price per pack of licit cigarettes is associated, on average, with a 5.0% rise in illicit cigarettes consumed at the aggregate level. However, this result does not take into consideration how countries with different border controls and laws regarding customs might respond differently. Hence, I estimate again equation (1) but restricting the sample to members of the European Economic Area (EEA). This area is characterized by freedom of movement of persons and goods, therefore making smuggling easier. As France is a member of the EEA, this new regression uses a more representative sample for the future forecast of the policy. The results (regression 2.7) give a cross-country average cross PED for illicit cigarettes in EEA countries of 0.602. This shows that the population in EEA countries is more responsive to increases in prices of cigarettes than non-EEA countries, at least in our sample, which demonstrates the importance of cross-border shopping and the transport of contraband or counterfeit cigarettes inside the area15_.

15_{Although, Transcrime (2015) finds that the highest volume of illicit cigarette flows emanates mainly from} outside the EEA, often entering through Russia, Turkey, and the northeast criminal hubs of Lithuania, Estonia, and Latvia.

(23)

23 When including year effects, the results show non-significant coefficients on the year dummies. The size of the price coefficients changes only slightly, although their significance drops (regression 2.6 and 2.8).

Finally, since the corruption variable varies more between countries than across time, the results of the country fixed effects and year effects regressions show no significant impact of corruption on illicit trade. However, the sign of the coefficient is still negative as expected. In line with Prieger and Kulick (2018), my results show that the corruption level of a country matters less than income per capita as determinants of the illicit cigarettes consumption.

There are some limitations to my analysis. I estimate the cross PED for illicit cigarettes without controlling for or analyzing the effects of other policy measures that may accompany price increase policies. It is also important to notice that the estimated impacts of increasing prices, forecasted in Part VI, are for a country unilaterally raising its prices. However, the price differentials among countries affect the volume of illicit cigarettes traded. Therefore, a harmonized raise or decrease of prices across the EU would not necessarily lead to similar changes in smuggling.

3. Prevalence of smoking

Once again, the pooled panel data regressions confirm the expected sign of the price coefficient. Although not controlling for country fixed effects, this shows the negative correlation between the price of licit cigarettes and the prevalence of smoking.

After controlling for unobserved differences among countries in Appendix 15, I estimate the cross-country average price elasticity of the prevalence of smoking to be -0.26 (regression 3.3). This means that a 10% increase in the price per pack of licit cigarettes is associated, on average, with a 2.6% decrease in smoking prevalence for the entire population of a country. However, these results do not take into consideration how countries with different levels of income might respond differently. Hence, I estimate again equation (1) but restricting the sample to high-income countries16. The results (regressions 3.5) give a cross-country average price elasticity of prevalence of smoking in high-income countries of -0.534. This shows that the population in high-income countries is more responsive to increases in prices of cigarettes than middle-income countries, at least in our sample.

(24)

24 In addition, the results show a negative significant coefficient for “Warn H”. This demonstrates that health warnings participate in reducing smoking prevalence. The coefficients on the other policy indices are not significant.

The general trend for the prevalence of smoking is downward during the entire period. The association between prices and the prevalence of smoking may stem from a spurious correlation of trending variables. When including year effects, the results show significant negative coefficients on the year dummies and lower and not significant PEDs for smoking prevalence (regressions 3.4 and 3.6). Hence, the PEDs obtained using only the country fixed effect specification are influenced by aggregate trends which are not related to the causal relationship of smoking prevalence.

In order to correct for the remaining possible endogeneity explained in the previous Methodology section, I present the results from different IV regressions in Appendix 16. The two first regressions (3.7 and 3.8) using only the total excise tax as instrument have a significant first stage F-statistic, however, the estimates are not significant. Therefore, I do not use the results from these regressions in the forecast in Part VI. The country fixed effects IV regression, adding the VAT tax as additional instrument (3.9), has a significant first stage F-statistic. The significant estimate of the price elasticity of smoking prevalence obtained through this specification is higher than the one without instruments. It is used in Part VI for the forecast of the policy.

(25)

25

IV. MACROECONOMIC ANALYSIS – FRANCE

This section is dedicated to the analysis of the French aggregate trend in consumption of cigarettes and roll-your-own tobacco for the past two decades. Using monthly time series data, I estimate the PED for licit cigarettes and the cross PED for roll-your-own tobacco in France. These estimates are used in the forecasting of the effects of the policy on consumption in Part VI.

A. Data

1. Source

In order to study the impact of the price of cigarettes on the sales of licit cigarettes and roll-your-own tobacco in France, I use monthly data compiled by the OFDT between January 2000 and December 2017: cigarette sales in millions of units (source: Directorate General of Customs and Excise (DGDDI)17_{) and the selling price of a pack (in euros) of the currently}

most sold brand: Marlboro red 20 cigarettes pack for cigarettes and Fleur du Pays Blond 40g for roll-your-own tobacco (source: Journal Officiel). The database integrates other indicators which can also influence the evolution of cigarette sales: sales of stop-smoking medicines quantified as the equivalent number of smokers treated (source: Group for the development and production of statistics (GERS)18, sales of roll-your-own tobacco (source: DGDDI) and the average unemployment rate (source: Federal Reserve Bank of St. Louis (FRED)).

For more details, please refer to the descriptive statistics in Appendix 5.

2. Summary statistics

Summary statistics for the time series used in this section are shown in Table 2.

Unemployment ranges from 7.2% to 10.5% with a decreasing trend from 2000 to 2008. After the economic crisis, it increases strongly and remains high for the rest of the period. The number of smokers treated has been continuously increasing throughout the period studied. It is interesting to notice that this variable has a seasonal trend with a high consumption of

17_{In French: “Direction Générale des Douanes et Droits Indirects” (DGDDI)}

(26)

26 stop-smoking medicines in January, which may be linked with some smokers taking the good resolution to stop smoking for the New Year.

Table 2 – Summary statistics Part IV

Note: Monthly data from January 2000 to December 2017; data on roll-your-own sales only available starting January 2004

Figure 1 shows the evolution of cigarette sales (dotted line) and cigarette prices (solid line) between January 2000 and December 2017. It highlights recurring increases, especially the successive increases in the price of tobacco in October 2003 (from €3.9 to €4.6) and in January 2004 (from €4.6 to €5). Between July 2003 and February 2004, cigarette sales dropped by 43%, probably due to the mentioned price increases. The month before a price increase, consumers seem to be buying more packs of cigarettes to limit the impact of rising future prices. This double increase in prices seems to have stabilized cigarette sales from March 2004 onwards. Consumers appear to be affected by a rise in the price in the short-term before gradually returning to a logic of consumption of "dependency" goods.

Over the 2004-2017 period, tobacco sales remain relatively stable in trend (around 4,250 million units per month) and even though some more moderate increases in the price of cigarettes occurred, they had a smaller impact on sales.

Variable Obs Mean Std. Dev. Min Max

Total cig sales FR 216 4781.072 1114.485 3155.589 7933.285

Price cig per pack FR 216 5.43 1.25 3.20 7.30

Total roll sales FR 168 657.352 76.597 480.221 856.155

Price roll per pack FR 216 6.67 2.50 2.97 11.70

Unemployment 216 9.18 0.81 7.20 10.50

(27)

27 Figure 1 – Evolution of monthly sales (in millions of sticks) and price of the most sold pack of 20 cigarettes (in euros) between 2000 and 2017

The evolution of sales is difficult to interpret due to the development of contraband and cross-border purchases, which are inherently difficult to estimate. Customs seizures of tobacco have increased steadily since 1999 (194 tons in 1999, 462 tons in 2011 and 630 tons in 2015, source DGDDI). However, cross-border shopping is the main component of the parallel market for tobacco products. These cross-border purchases (mainly cigarettes) represent between 10,000 and 11,000 tons for the year 2007, one in five cigarettes consumed by French smokers and two billion euros annual tax evasion according to Lakhdar et al. (2011). This cross-border trade has been intensified following the increases in price, which significantly broadened the price differential with neighboring countries. As can be seen in Appendix 1, the consumption of illicit cigarettes increased strongly from 2002 to 2007 before becoming more stable.

The sales of roll-your-own tobacco are increasing slightly over the sample period as shown in Figure 2. This could be explained by the transfer of some cigarette smokers' consumption towards this type of products, which remains cheaper when comparing the price per gram of tobacco. Nevertheless, this transfer of consumption must be put into perspective, as cigarette sales remain largely predominant in France, accounting for 81% of the volume of tobacco sales in 2017 (source: DGDDI).

3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 2000 3000 4000 5000 6000 7000 8000 9000 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 year

(28)

28 Overall, even though the number of cigarettes smoked fell significantly over the period, smoking among the French population increased between 2004 and 2011 when adding licit cigarettes, illicit cigarettes and roll-your-own tobacco in volume. It decreased afterward.

Figure 2 – Evolution of monthly sales of roll-your-own tobacco (in tons) and price of the most sold pack of 20 cigarettes (in euros) between 2000 and 2017

Note: Data on roll-your-own sales only available starting January 2004.

It should be noted that the sales time series has a seasonal component with an upward trend between May and September, due to the summer period, marked notably by the increased purchase of tobacco by French consumers and foreign tourists.

3 4 5 6 7 400 500 600 700 800 900 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 year

(29)

29 B. Methodology

Since the dataset used is a time series, I proceed to Dickey-Fuller tests for unit root for the series of sales of cigarettes and roll-your-own tobacco. Results of the tests are shown in Appendix 17. For both series, the null hypothesis of unit root is rejected. Therefore, the series of sales of cigarettes and roll-your-own tobacco are stationary. This is important because it helps to identify the driving factors and enable to infer a correlation between the two time series that would otherwise be misleading.

1. Licit cigarette consumption

In order to estimate the PED for licit cigarettes in France, I use the same equation as Fromentin (2015) and regress monthly licit cigarettes consumption (Y!) on the previous

month consumption (𝑌_!!!), the price of a pack of 20 cigarettes of the most sold brand (𝑃_!), and a vector of control variables (𝐖𝒕): the unemployment rate and sales of stop-smoking

medicines (regression 4.1 in Appendix 18).

ln(Y!) = 𝛽!+ 𝛽!ln (𝑃!) + 𝛽!ln 𝑌!!! + 𝜶 ln 𝐖𝒕 + 𝜀! (2)

The current consumption is a function of the price and past consumption. Past sales are incorporated into the model to account for the inertia of cigarette sales. An omission of past sales in the model would overestimate the response of consumers to price changes. The integration of the unemployment rate series aims to take into account the socio-economic situation that could influence consumption. Sales of stop-smoking medicines are also included in the model to estimate the influence of treatment on tobacco use.

2. Roll-your-own tobacco consumption

To estimate the cross PED for roll-your-own tobacco in France, I first use equation (2) replacing the dependent variable by the monthly sales of roll-your-own tobacco (regression 4.2 in Appendix 18). For the same reason than for licit cigarettes, past sales are incorporated into the model. In addition, in another regression, the price of roll-your-own tobacco is also added to the regression as a control (regression 4.3 in Appendix 18).

(30)

30 The coefficient of interest is 𝛽_! as it is the short-term price elasticity estimate. The long-term price-elasticity estimate can be obtained by !!

!!!! .

C. Results

The results are presented in Appendix 18. The short-term PED for licit cigarettes is -0.574, inducing that a 10% increase in the price of cigarettes results in an average sales decrease of approximately 5.7%. By calculating the long-term PED from the price coefficient and the coefficient of past sales, I obtain that a 10% increase in the price of cigarettes results in an average sales reduction of 8.2%, all other things being equal.

The short-term cross PED for roll-your-own tobacco obtained in regression 4.2 (without controlling for the price of roll-your-own tobacco) is 0.339, meaning that a 10% increase in the price of cigarettes results in an average sales increase of 3.4% of roll-your-own tobacco. Using the coefficient on past sales of roll-your-own tobacco and the short-term cross price elasticity estimate, I find a long-term cross price elasticity of 0.50.

It is interesting to notice that when the price of roll-your-own tobacco is incorporated into the regression (4.3), its coefficient is not significant. This leads to the conclusion that sales of roll-your-own tobacco are mainly driven by the price of cigarettes in France, as roll-your-own tobacco represents a clear substitute for cigarettes when the price of the latter increases. As in Part III, the price of illicit cigarettes could not be obtained and therefore not controlled in the estimation of the cross PED for illicit cigarettes. Therefore, only the estimate of the cross PED for roll-your-own tobacco from regression 4.2 is used for forecasting the policy.

For all the regressions, the long-term price effect is greater than the short-term price effect. Becker et al. (1991) argue that this is natural as a price increase has a negative effect on current consumption and future dependence.

Previous works also show that price seems to be the most important predictor of cigarette sales variations in France, with a price elasticity close to -0.40 (Godefroy (2003)) and -0.46 (Fromentin (2015)). Therefore, my estimate of -0.574 is slightly above these findings. This can be explained by the fact that the period used in my analysis incorporates data for the most recent years, for which sales have constantly been decreasing.It should also be noted

(31)

31 that in a meta-analysis presenting the conclusions of 86 studies on the economic aspects of tobacco, Gallet and List (2003) find an average price elasticity of -0.48.

Finally, the estimate found in my result is close to the one obtained in Part III using a cross-country analysis (-0.563) and a bit lower than the one obtained when restricting the sample to only high-income countries (-0.673). Watkins (2015) found that for high-income countries, the highest quantiles of consumption have stronger price elasticities. The sample restricted to high-income countries used in Part III has a higher average consumption than France throughout the period, therefore, it justifies the higher cross-country estimate of price elasticity.

The variable "unemployment", which represents the socio-economic situation, seems to have an influence on tobacco sales with a significant coefficient in regression 4.1. It was conceivable that the deterioration of the French economic situation could influence consumption downward or upward. Indeed, a gloomy economic situation could encourage smokers to reduce their tobacco consumption to save money. Conversely, the loss of a job could result in increased "idle" behavior, which could result in a rise in tobacco use.

On the contrary, changes in sales of stop-smoking medications do not seem to have an influence on tobacco sales. A significant downward influence on cigarette sales could have been expected and would have shown the efficiency of this type of treatment.

Finally, the selling price of cigarettes is an efficient regulation tool on aggregated quantities sold, and concomitantly on the consumption of individuals. However, the magnitude of price increases determines the impact on cigarette sales. Indeed, Fromentin (2015) estimates the same equation over two different periods, one from January 2000 to October 2003 and one from November 2003 to December 2012. He notices that the former period was marked by relatively large recurring price increases. The first period has a rather strong price elasticity in the short-term (-0.45) and in the long-term (-0.47). On the other hand, during the period from November 2003 to December 2012, the weaker and more spaced price increases have a limited influence on tobacco sales, with a short-run price elasticity of -0.177 and a long-term of - 0.2.

(32)

32

V.

MICROECONOMIC ANALYSIS – FRANCE

In this section, I study data on cigarette consumption at the household level in France. Although I am not able to estimate the PED for licit cigarettes for each decile, I analyze the past trend in tobacco consumption per decile and the share of purchasing power it represents.

A. Data

The dataset used to study consumption at the microeconomic level comes from the "Budget des Familles" surveys. Five of these surveys have been carried out since 1995 by INSEE, for the following years: 1995, 2000, 2005, 2010 and 2015. I was unable to be granted access to the last one. Approximately 10,000 to 15,000 households are interviewed for each survey. The questions asked relate to the incomes and expenses of households. In addition, the surveys contain a set of socio-demographic variables (see Appendix D for more information on the surveys).

Expenditure is broken down into several consumption items, including one item for cigarettes, cigars, and cigarillos. Each resource or consumption item is associated with an amount corresponding to the annual sum obtained or spent on it. Expenditures on cigarettes per household in 2010 euros are reported in Appendix 19.

To have a comparable measure of income per decile for the different surveys, I follow the definition of INSEE of the total revenue (variable revtot). I refer readers to Appendix D for more information on the income variable. Revenues per decile in 2010 euros are shown in Appendix 20. Over the period studied, household incomes have remained relatively stable except in 2010 with a significant drop for low-income deciles. This drop – particularly for the first decile – must be due to a change in the sample of households surveyed with on average poorer households in the 2010 survey. The economic crisis that hit France in 2008 cannot be the only explanation for such a decrease.

By analyzing consumption data from Appendix 21, I find that the consumption of cigarettes by household does not decrease according to the deciles, but on the contrary seems to increase with the decile - this increase being, however, less important than the rise in income itself from one decile to the next. This phenomenon can be interpreted in two ways.

Increase in the price of tobacco products in France from 2018 to 2020 : a forecast of the economic impact and the effect on tobacco consumption