how well targeted are soda taxes - university of manchester · soda taxes i increased interest in...

63
How well targeted are soda taxes Pierre Dubois, Rachel Griffith and Martin O’Connell Institute for Fiscal Studies QMUL, May 2017 Introduction Demand Estimates Analysis Summary 1 / 37

Upload: others

Post on 25-Jul-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

How well targeted are soda taxes

Pierre Dubois, Rachel Griffith and Martin O’Connell

Institute for Fiscal Studies

QMUL, May 2017

Introduction Demand Estimates Analysis Summary 1 / 37

Page 2: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Corrective Taxes

I Corrective taxes have long been seen as a tool to improve socialwelfare when consumption imposes costs on others

I externalities (Pigou, 1920)

I taxes on fuel, alcohol, tobacco, gambling are examples

I More recently have been advocated to reduce consumption thatimposes costs on your future self (sin taxes)

I “internalities”

I Gruber and Koszegi, 2004; ODonoghue and Rabin, 2006; Haavio andKotakorpi, 2011; Allcott, Mullainathan and Taubinsky, 2014, ...

I Soda taxes are a leading example

Introduction Demand Estimates Analysis Summary 2 / 37

Page 3: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda taxesI Increased interest in soda taxes to reduce consumption of sugar

I introduced in France in 2012, Mexico in 2013, UK in 2016

I four cities in California (Albany, Berkeley, Oakland, and San Francisco)

I Boulder, Philadelphia, Cook County in Illinois and the Navajo Nation

I World Health Organisation (WHO) in 2016 urged countries to taxsugary drinks to reduce sugar consumption, especially in children

I considerable evidence that sugar is over consumed, and that this

I leads to obesity, type 2 diabetes, heart disease, cancers, etc...

I is associated with poor mental health and poor school performance,particularly in children

I poor childhood nutrition thought to be an important determinant oflater life health, social and economic outcomes and of presistentinequality

Introduction Demand Estimates Analysis Summary 3 / 37

Page 4: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Corrective Taxes

I There is a sharp divide between

I libertarian view

I harm accrues to the individuals themselves, so there is no rationale forgovernment intervention

I sin taxes are ineffective, regressive and largely aimed at raising revenue

I paternalistic view

I people make errors (due to cognitive constraints, inattention, ...), withinformation on when and how we can design policies to help them makebetter choices (e.g. O’Donoghue and Rabin, 2003, 2006)

I not to be confused with the public health literature where health risksshould be reduced, even where individuals are making optimal choices

I We want to empirically evaluating the efficiency and equityimplications of recently introduced soda taxes

Introduction Demand Estimates Analysis Summary 4 / 37

Page 5: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Overconsumption of sugar, mean consumption

Recommended maximum is 37.5 grams for men and 25 grams for women

Source: Euromonitor

Introduction Demand Estimates Analysis Summary 5 / 37

Page 6: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Overconsumption of sugar in the USCumulative density, share of calories from added sugar

Introduction Demand Estimates Analysis Summary 6 / 37

US government recommended maximum

←− 65% are aboverecommended maxas share of calories

Source: NHANES

Page 7: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Overconsumption of sugar in the UKCumulative density, share of calories from added sugar

Introduction Demand Estimates Analysis Summary 7 / 37

US government recommended maximumUK government recommended maximum

←− 70% are above USrecommended maximum

←− 90% are above UKrecommended maximum

Source: LCFS

Page 8: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Overconsumption of sugar in the UKMean by age group, sugar consumption particularly high amongst children

Introduction Demand Estimates Analysis Summary 8 / 37

Source: NDNS

years old: 1.5-3 4-10 11-18 19-64 65+

Page 9: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda and overconsumption of sugar in the USSugar from soda represents a large share of calories from added sugar

Introduction Demand Estimates Analysis Summary 9 / 37

40%

0%

Share ofcalories fromadded sugarthat comefrom soda

Share of calories from added sugarSource: NHANES

Page 10: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda and overconsumption of sugar in the UKSugar from soda represents a large share of calories from added sugar

Introduction Demand Estimates Analysis Summary 10 / 37

50%

0%

Share ofcalories fromadded sugarthat comefrom soda

Share of calories from added sugarSource: NDNS

Page 11: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda taxes

I Soda taxes look to be (potentially) well motivatedI (very) high consumption of sugar

I soda represents a substantial share of sugar consumption

I soda consumption is higher for those consuming a lot of sugar

I soda has no redeeming nutritional characteristics

I There is potential for equity concernsI consumption of sugar and soda are higher amongst lower income

households

I We want to evaluate efficiency and equity implications of a soda tax

Introduction Demand Estimates Analysis Summary 11 / 37

Page 12: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Our contribution

I We study the on-the go market

I less well studied than the at home market

I interesting because may be stronger reasons for intervention

I and avoid stockpiling issues

I We provide non-parametric estimates of key preference parametersthat determine demand for soda and responsiveness to taxes

I typically we include random coefficients and make distributionalassumptions

I We study how well soda taxes targets those consumers whoover-consume by the most

Introduction Demand Estimates Analysis Summary 12 / 37

Page 13: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Outline

1. Introduction

I motivation and contribution

I optimal corrective tax

2. Demand model and specification

3. Estimates of preference parameters

4. Analysis

5. Summary and conclusions

Introduction Demand Estimates Analysis Summary 13 / 37

Page 14: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I i ∈ 1, ...,N consumers with income yi

I j = 1, ..., j ′, j ′ + 1, ..., J ∈ Ω food and drink products

I j ∈ 1, ..., j ′ = Ωw sodas

I j ∈ j ′ + 1, ..., J = Ωnw other products that contain sugar

I p = (p1, . . . , pJ)′: post tax prices

I zj sugar in each product

I τ tax rate on the sugar in soda

I vi (p, yi ): indirect decision utility

I demands are: qij(p, yi ) = −∂vi/∂pj∂vi/∂yi

Introduction Demand Estimates Analysis Summary 14 / 37

Page 15: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I Si (p, yi ) =∑

j∈Ω zjqij(p, yi ): total sugar in a consumer’s diet

I includes sugar from soda and non-sodaSi (p, yi ) = Swi (p, yi ) + Snwi (p, yi )

I vi (p, yi ) governs choice, but not necessarily long term welfare

I sugar consumption may give rise to future costs, both to the consumer(internalities) and to others (externalities, that consumers do not takeaccount of at the point of consumption

I φi (Si (p, yi )): the externality + internality from sugar consumption, weassume it is a positive, convex function

I consumers ignore this when making choices

Introduction Demand Estimates Analysis Summary 15 / 37

Page 16: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I tax policy has the potential to improve welfare by inducing consumersto internalise φi (Si (p, yi ))

I how effective it is depends on

I how successfully it averts internalities by lowering sugar consumption ofthose prone to suffer them

I how much it distorts the behaviour of those that do not suffer frominternalities

I extent it has undesirable distributional consequences

Introduction Demand Estimates Analysis Summary 16 / 37

Page 17: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I Policy maker chooses τ to maximise a utilitarian social welfarefunction,

maxτ

W =

∑i

[vi (p, yi + ri )− φi (Si (p, yi + ri ))]

I where vi (p, yi + ri ) is indirect decision utility, ri is any rebate toconsumer i

I pj denotes the pre-tax price

I for soda products (j ∈ Ωw ), pj = pj + τzj

I for non-soda products (j ∈ Ωnw ) pj = pj

Introduction Demand Estimates Analysis Summary 17 / 37

Page 18: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I A marginal change in the soda tax effects welfare in the following way:

dW

dτ=∑i

(φ′i − τ λ)|S ′wi | −∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I φ′i ≡ φ′i (Si (p, y + ri )) is the marginal internality of consumer i

I S ′i effect of marginal change in tax rate on consumer’s sugar demand

I λi is the marginal (decision) utility of income of consumer i

I λ =∑

i βiλi captures tax revenue that is redistributed back toconsumers

Introduction Demand Estimates Analysis Summary 18 / 37

Page 19: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

dW

dτ=∑i

(φ′i − τ λ)|S ′wi | −∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I The direct efficiency effect of the tax

I For consumers with a marginal internality that exceeds the tax rate thisterm is positive

I the reduction in sugar from soda from tax leads to a welfare gain

I the size of the gain is proportional to how response the consumer’sdemand for sugar in soda is to the tax instrument

I For consumers with marginal internalities below the tax rate this term isnegative

I The higher the correlation between φ′i and |S ′wi | the more effective the

tax is

Introduction Demand Estimates Analysis Summary 19 / 37

Page 20: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

dW

dτ=∑i

(φ′i − τ λ)|S ′wi | −∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I The indirect efficiency effect associated with how taxing the sugar insoda affects demand for sugar in non soda products

I If S′nwi > 0, so taxing the sugar in soda increases demand for untaxed

sugar, this will be negative

I If consumers with large marginal internalities strongly switch to otherforms of sugar this inefficiency cost from only taxing a subset of sugarwill be large

I The lower the correlation between φ′i and |S ′nwi | the more effective the

tax is

Introduction Demand Estimates Analysis Summary 20 / 37

Page 21: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

dW

dτ=∑i

(φ′i − τ λ)|S ′wi | −∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I Redistributive effect

I If consumers with high marginal utility of income have high demands forsugary soda then tax will be regressive and this term will be negative

Introduction Demand Estimates Analysis Summary 21 / 37

Page 22: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in sodaI Internality taxes are rationalised as a way to help people who will later

regret their consumption choices

I if effective the tax will lead to fewer regrets about unhealthy choices,but they will also have less income

I (we think) high internality individuals tend to be lower income

I poverty, lack of self-control and low cognition are correlated (andpossibly causally related)

I tax might serve a self-control function that benefits lower incomegroups more

I but only when they are more price sensitive so respond to the tax

I if lower income consumers have inelastic demand (as could be withpreference heterogeneity, if inattentive, with habits, ...) they may paythe tax while also bearing most of the internal costs

I Taxes will also fall on consumers with low internalities

Introduction Demand Estimates Analysis Summary 22 / 37

Page 23: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

A tax on sugar in soda

I To assess the effect of taxes on soda empirically we need

I estimates of substitution away from the sugar in soda (S′wi )

I how strongly switch to alternatives (S′nwi )

I measure of marginal internalities (φ′i )

I measure of the marginal utility of income (λi )

I the correlations between all of these

Introduction Demand Estimates Analysis Summary 23 / 37

Page 24: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Outline

1. Introduction

I motivation and contribution

I optimal corrective tax

2. Demand model and specification

3. Estimates of preference parameters

4. Analysis

5. Summary and conclusions

Introduction Demand Estimates Analysis Summary 24 / 37

Page 25: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Demand model overviewI Model demand for drinks on-the-go

I Model choice of drink conditional on buying a drink

I Discrete choice model, individual chooses one of several options

I one of several soda products

I one of two outside options

I non-soda drinks with sugar

I non-soda drinks without sugar (mainly water)

I Characteristics of products determine choice

I preferences for price, sugar and soda will be individual specific,exploiting long panel

I other preferences (pack size, brand-time effects) will vary bydemographic group (gender-age)

Introduction Demand Estimates Analysis Summary 25 / 37

Page 26: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Data

I Use data on purchases made by a panel of consumers of food anddrink bought “on-the-go”

I collected by market-research firm Kantar

I participants use mobile phones to record all purchases

I Random sample of 5373 individuals taken from households thatparticipate in Worldpanel survey

I 1478 never purchase drinks while on-the-go (though do purchasenon-drinks)

I 1332 only purchase non-soda drinks (fruit juice, flavoured milks, water)

I 2563 are soda purchasers

I Longitudinal: we observe each consumer making purchases on at least25 separate days (81 on average)

Introduction Demand Estimates Analysis Summary 26 / 37

Page 27: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda products

Product

Brand Variety Size Market Price g sugarshare (£) per 100ml

Soda optionsCoca Cola 38.1%

Regular 330ml can 6.2% 0.62 10.6Regular 500ml bottle 11.2% 1.08 10.6Diet 330ml can 7.1% 0.63 0.0Diet 500ml bottle 13.6% 1.09 0.0

Cherry Coke 4.3%Regular 330ml can 0.8% 0.63 11.2Regular 500ml bottle 2.4% 1.08 11.2Diet 500ml bottle 1.1% 1.08 0.0

Pepsi 15.1%Regular 330ml can 1.6% 0.61 11.0Regular 500ml bottle 3.5% 0.96 11.0Diet 330ml can 1.9% 0.62 0.0Diet 500ml bottle 8.2% 0.95 0.0

Fanta 5.9%Regular 330ml can 0.9% 0.60 6.9Regular 500ml bottle 4.5% 1.08 6.9Diet 500ml bottle 0.5% 1.07 0.6

other soda products: Lucozade, Oasis, Ribena, ...

Non-soda options Fruit juice 330ml 4.0% 1.39 10.6Flavoured milk 500ml 2.2% 0.96 10.6

Outside option 12.3%

Introduction Demand Estimates Analysis Summary 27 / 37

Page 28: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Empirical specificationI i : consumer; t: purchase occasion

I Inside option decision utility (j ∈ 1, ..., J):

Uijt = αi + βipjrt + γi sj + gi (xjt) + εijt

I pjrt price of product j at time t in store r

I sj indicator of sugary vs. diet

I xjt additional product attributes (pack size; brand-time effects)

I εijt type I extreme value deviate

I Outside option decision utility (j = 0):

Ui0t = ζdrt + εi0t

ζdrt : gender-age-store-time effect

Introduction Demand Estimates Analysis Summary 28 / 37

Page 29: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Preference heterogeneity

I Soda (αi ), price (βi ) and sugar (γi ) preferences are consumer specific

I We treat α = (α1, ...αN)′, β = (β1, ...βN)′ and γ = (γ1, ...γN)′ asparameters

I using large T dimension of data to recover estimates of (α,β,γ)

I and large N dimension to construct nonparametric estimate off (αi , βi , γi )

I We also allow for the possibility of infinite regions of the parameterspace

I for instance, consumers that never purchase sugary (non-diet) productshave γi = −∞

Introduction Demand Estimates Analysis Summary 29 / 37

Page 30: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Our approach vs. random coefficient logit

I It’s well understood that incorporating preference heterogeneity isimportant for capturing realistic substitution patterns

I Standard approach uses random coefficients, requires distributionassumptions, allows us to recover distribution of preferences byestimating a small number of parameters

I Strength of our alternative approach is

I don’t need to specify the functional form of preference distribution

I we recover consumer specific parameters and therefore can relate themto other information about consumers

I this is key if we want to assess how well targeted

Introduction Demand Estimates Analysis Summary 30 / 37

Page 31: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issuesI Identifying the effect of price

I brand-level time effects control for time varying shocks to demand

I we exploit

I cross-retailer variation in the relative prices of different drinks and

I time series variation in products price within brands that is driven byfactors other than shocks to consumers’ soda demands

Details

I Incidental parameters problem

I in non-linear panel data model with individual parameters, even if bothT and N go to infinity, the maximum likelihood estimates may sufferfrom an incidental parameters problem

I we show that our results are not overly affected by this problem

Details

Introduction Demand Estimates Analysis Summary 31 / 37

Page 32: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Outline

1. Introduction

I motivation and contribution

I optimal corrective tax

2. Demand model and specification

3. Estimates of preference parameters

4. Analysis

5. Summary and conclusions

Introduction Demand Estimates Analysis Summary 32 / 37

Page 33: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

EstimatesMoments of distribution of consumer specific preferences

StandardVariable Estimate error

Price Mean -1.7856 0.0727Standard deviation 4.3500 0.0898Skewness 0.6999 0.1892Kurtosis 6.6126 0.9868

Soda Mean -0.6297 0.0938Standard deviation 2.8590 0.0622Skewness 0.1663 0.1994Kurtosis 5.5891 1.0256

Sugar Mean -0.0027 0.0218Standard deviation 2.0513 0.0836Skewness -0.8439 0.5754Kurtosis 9.3688 4.3680

Price-Soda Covariance -5.6252 0.3427Price-Sugar Covariance -1.0102 0.2236Soda-Sugar Covariance 0.4631 0.2928

Introduction Demand Estimates Analysis Summary 33 / 37

Page 34: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

EstimatesConsumer group specific preferences

Estimate Standard Estimate StandardVariable error error

Female - <40 Female - 40+288ml carton 1.2407 0.0469 0.6903 0.0698380ml bottle 2.3507 0.0532 2.2498 0.0568500ml bottle 2.2819 0.0577 2.2436 0.0669Fanta -1.7720 0.1469 -1.7825 0.1479Cherry Coke -1.6180 0.1408 -2.5570 0.1922Ribena -1.1724 0.1177 -1.3078 0.1191Pepsi -0.9854 0.0957 -0.9844 0.1003Lucozade -2.0206 0.1710 -1.5247 0.1418Oasis -2.1829 0.1527 -1.9083 0.1391Fruit juice -0.1399 0.2867 1.4397 0.3273Flavoured milk -3.5404 0.2616 -2.9586 0.3426

Male - <40 Male - 40+288ml carton -0.1527 0.0634 -0.1002 0.0658380ml bottle 2.1269 0.0452 2.3867 0.0502500ml bottle 2.5041 0.0544 2.2437 0.0587Fanta -1.8289 0.1238 -1.3700 0.1059Cherry Coke -2.1613 0.1489 -1.9045 0.1428Ribena -2.3484 0.1506 -1.1716 0.1091Pepsi -1.5897 0.1046 -0.9880 0.0873Lucozade -1.6734 0.1323 -1.8384 0.1177Oasis -2.3463 0.1582 -2.8384 0.1747Fruit juice 1.4702 0.3233 -0.6937 0.3473Flavoured milk -2.3191 0.2501 -3.7719 0.3130

Time-demographic-brand effects YesRetailer-demographic-brand effects Yes

Introduction Demand Estimates Analysis Summary 34 / 37

Page 35: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Marginal preference distributions(a) Price (b) Soda

(c) Sugar (d)

Introduction Demand Estimates Analysis Summary 35 / 37

Page 36: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Preference parameters and % calories from added sugar(a) Price (b) Soda

(c) Sugar (d) Predicted sugar from soda

Introduction Demand Estimates Analysis Summary 36 / 37

Page 37: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Preference parameters and “income”(a) Price (b) Soda

(c) Sugar (d) Predicted sugar from soda

Introduction Demand Estimates Analysis Summary 37 / 37

Page 38: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Price effects

Effect of 1% price increase on:own cross demand for: total

demand sugary products diet products demand

Coca Cola 330 -2.182 0.141 0.057 0.013Coca Cola 500 -1.077 0.219 0.089 -0.064Coca Cola Diet 330 -2.033 0.047 0.153 0.015Coca Cola Diet 500 -0.916 0.073 0.269 -0.043Pepsi 330 -2.485 0.056 0.021 0.005Pepsi 500 -1.670 0.133 0.058 -0.035Pepsi Diet 330 -2.369 0.017 0.071 0.006Pepsi Diet 500 -1.412 0.047 0.161 -0.027Cherry Coke 330 -2.538 0.021 0.007 0.002Cherry Coke 500 -1.125 0.022 0.012 -0.008Cherry Coke Diet 500 -1.029 0.010 0.028 -0.005Oasis 500 -1.074 0.050 0.019 -0.018Oasis Diet 500 -0.975 0.015 0.044 -0.009

Introduction Demand Estimates Analysis Summary 38 / 37

Page 39: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Relationship between own price elasticities and added sugar

(a) Soda (b) Sugary soda

Introduction Demand Estimates Analysis Summary 39 / 37

Page 40: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

The impact of a tax on soda

dW

dτ=∑i

(φ′i − τ λ)|S ′wi | −∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I The higher the correlation between φ′i and |S ′wi | the more effective thetax is

I These figures suggest the correlation is zero (soda tax) or negative(sugary soda tax)

Introduction Demand Estimates Analysis Summary 40 / 37

Page 41: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Outline

1. Introduction

I motivation and contribution

I optimal corrective tax

2. Demand model and specification

3. Estimates of preference parameters

4. Analysis

5. Summary and conclusions

Introduction Demand Estimates Analysis Summary 41 / 37

Page 42: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Effects of a soda tax

I We simulate two taxes

I soda tax: 25p tax per litre on all soda (Philadelphia style)

I sugary soda tax: 48p tax per litre on only sugary soda (UK style)

I Rates chosen to be revenue equivalent

I We assume tax is fully passed-through to consumers

Introduction Demand Estimates Analysis Summary 42 / 37

Page 43: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Average effect of tax: % change in demand

Soda tax Sugary soda tax

Sugary soda -3.42 -5.78[-3.62, -3.24] [-6.16, -5.32]

Diet soda -2.60 3.53[-2.77, -2.40] [3.14, 3.79]

Sugary non-soda 5.72 4.95[4.36, 6.13] [4.02, 5.35]

Outside option 15.77 9.37[14.94, 16.55] [8.97, 9.82]

Introduction Demand Estimates Analysis Summary 43 / 37

Page 44: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Average effect of tax: grams of sugar

Total Change following:Sugar (kg) from pre tax soda tax sugary soda tax

soda 1.46 -49.06 -83.88[1.42, 1.45] [-51.95, -45.73] [-88.40, -76.83]

(%) -3.36 -5.75[-3.57, -3.18] [-6.15, -5.33]

all drinks 1.62 -39.83 -75.89[1.59, 1.62] [-42.03, -38.01] [-79.61, -69.56]

(%) -2.46 -4.68[-2.60, -2.37] [-4.96, -4.34]

Introduction Demand Estimates Analysis Summary 44 / 37

Page 45: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Average effect of tax: grams of sugar

Total Change following:Sugar (kg) from pre tax soda tax sugary soda tax

soda 1.46 -49.06 -83.88[1.42, 1.45] [-51.95, -45.73] [-88.40, -76.83]

(%) -3.36 -5.75[-3.57, -3.18] [-6.15, -5.33]

all drinks 1.62 -39.83 -75.89[1.59, 1.62] [-42.03, -38.01] [-79.61, -69.56]

(%) -2.46 -4.68[-2.60, -2.37] [-4.96, -4.34]

Introduction Demand Estimates Analysis Summary 44 / 37

Page 46: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Average effect of tax: grams of sugar

Total Change following:Sugar (kg) from pre tax soda tax sugary soda tax

soda 1.46 -49.06 -83.88[1.42, 1.45] [-51.95, -45.73] [-88.40, -76.83]

(%) -3.36 -5.75[-3.57, -3.18] [-6.15, -5.33]

all drinks 1.62 -39.83 -75.89[1.59, 1.62] [-42.03, -38.01] [-79.61, -69.56]

(%) -2.46 -4.68[-2.60, -2.37] [-4.96, -4.34]

Introduction Demand Estimates Analysis Summary 44 / 37

Page 47: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Average effect of tax: welfare

Soda tax Sugary soda tax

Comp var 6.00 3.42(GBP/consumer) [5.96, 6.01] [3.38, 3.42]

Tax revenue 5.91 3.32(GBP/consumer) [5.86, 5.91] [3.28, 3.32]

Consumer cost 15.07 4.51(GBP/100g sugar reduction) [14.24, 15.74] [4.25, 4.89]

Net consumer cost 0.24 0.13(GBP/100g sugar reduction) [0.23, 0.24] [0.13, 0.14]

Introduction Demand Estimates Analysis Summary 45 / 37

Page 48: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Sugar taxes: how well targeted(a) Soda tax - change in g (b) Sugary soda tax - change in g

(c) Soda tax - % change (d) Sugary soda tax - % change

Introduction Demand Estimates Analysis Summary 46 / 37

Page 49: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Sugar taxes: distributional effects(a) Soda tax - change in g (b) Sugar-soda tax - change in g

(c) Soda tax - % change (d) Sugar-soda tax - % change

Introduction Demand Estimates Analysis Summary 47 / 37

Page 50: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Sugar taxes: distributional effectsCompensating variation assuming lump sum redistribution of revenue

Introduction Demand Estimates Analysis Summary 48 / 37

Page 51: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Summary

I A tax on sugary soda achieves larger mean reductions in sugar

I it encourages switching from sugary to diet sodas

I but diet soda is more attractive for consumers who have weaker sugarpreferences

I so it exacerbates the weaker switching away from sugar of those withhigh added sugar in their diets relative to those with lower added sugar

I Neither tax is well targeted

I consumers with a high sugar consumption both have systematicallystronger preferences for sugar in drinks and are also less sensitive toprice changes compared to those with less added sugar in their diets

I this leads them to be less willing to switch away from sugary soda whenfacing higher prices due to tax

Introduction Demand Estimates Analysis Summary 49 / 37

Page 52: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Summary

I Model demand in the on-the-go soda market, estimating consumerspecific preference parameters for soda, price and sugar

I Work in progress

I upper level substitution to sugar in non-drinks products

dW

dτ=∑i

(φ′i − τ λ)|S′wi | −

∑i

φ′iS′nwi −

∑i

(λi − λ)Swi

I better specification of the internality-externality function

Introduction Demand Estimates Analysis Summary 50 / 37

Page 53: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Additional slides

Introduction Demand Estimates Analysis Summary 51 / 37

Page 54: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Participation in market

Female, <40 Female, 40+ Male, <40 Male, 40+ TotalNever purchase drink 166 340 204 393 1103

14.0 23.3 17.6 28.1 21.2Only purchase non-soda drinks 397 550 331 455 1733

33.6 37.6 28.6 32.5 33.3Purchase soda 619 571 623 550 2363

52.4 39.1 53.8 39.3 45.5Total 1182 1461 1158 1398 5199

100.0 100.0 100.0 100.0 100.0

Introduction Demand Estimates Analysis Summary 38 / 37

Page 55: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Soda consumers

Purchase:Non soda and soda Only soda Total

Purchase: (αi ∈ −∞,∞) (αi =∞)Only diet (γi = −∞) 170 24 194

7.6 21.4 8.2Both diet and sugary (γi ∈ −∞,∞) 1563 49 1612

69.4 43.8 68.2Buy only sugary (γi =∞) 518 39 557

23.0 34.8 23.6Total 2251 112 2363

100.0 100.0 100.0

Introduction Demand Estimates Analysis Summary 38 / 37

Page 56: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Sugar taxes: how well targetedChange in sugar from drinks (in grams)

(a) Soda tax (b) Sugary soda tax

Introduction Demand Estimates Analysis Summary 38 / 37

Page 57: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Sugar taxes: distributional effectsChange in sugar from drinks (in grams)

(a) Soda tax (b) Sugar-soda tax

Introduction Demand Estimates Analysis Summary 38 / 37

Page 58: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: Identification of price effects

I prices, pjrt measured at the retail outlet level

I the price of Coca Cola varies across the 330ml can and 500ml bottleversion of the product and across retail outlets

I we include brand-level time effects to control for time varying shocksto demand

I the price variation we exploit is

I cross-retailer variation in the relative prices of different drinks and

I time series variation in products price within brands that is driven byfactors other than shocks to consumers’ soda demands

Back

Introduction Demand Estimates Analysis Summary 38 / 37

Page 59: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: IdentificationPrice variation for Coca Cola

(a) 330ml can (b) 500ml bottle(c) Within brand price vari-ation

Back

Introduction Demand Estimates Analysis Summary 38 / 37

Page 60: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: Incidental parameters problem

I Identification of individual preference parameters relies on the large Tdimension of the data

I We may suffer from an incidental parameter problem that is commonin non linear panel data

I even if both T and N go to infinity, the maximum likelihood estimatesmay suffer from an incidental parameters problem (Arellano and Hahn(2007)).

I There are several methods for bias correction; we apply a paneljackknife method (Hahn and Newey (2004), Dhaene and Jochmans(2015)) and show that our results are not overly affected by theincidental parameters problems

Back

Introduction Demand Estimates Analysis Summary 38 / 37

Page 61: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: Incidental parameters problemBias for sugar preference parameters relationship with T

Back

Introduction Demand Estimates Analysis Summary 38 / 37

Page 62: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: Incidental parameters problemBias for sugar preference parameters relationship with sugar

Back

Introduction Demand Estimates Analysis Summary 38 / 37

Page 63: How well targeted are soda taxes - University of Manchester · Soda taxes I Increased interest in soda taxes to reduce consumption of sugar I introduced in France in 2012, Mexico

Econometric issues: Incidental parameters problemDistribution of sugar preference parameters

Back

Introduction Demand Estimates Analysis Summary 38 / 37