a sales tax is better at promoting healthy diets than the ...a sales tax is better at promoting...
TRANSCRIPT
A Sales Tax Is Better at Promoting Healthy Diets
than the Fat Tax and the Thin Subsidy
October 9, 2019
Appendix
A Proof of Proposition 1
To derive the effects of the policy parameters on consumption, we totally differentiate
the system of first-order conditions (11) and (12) with respect to H∗, F ∗, τF , σ and τC .
The total differential results in the following system of equations1:
(a11 a12a21 a22
)︸ ︷︷ ︸
=J
(dH∗
dF ∗
)=
(0 −VZpGα VZpGα
VZpF 0 VZpF
) dτFdσ
dτC
, (A.1)
where
a11 ≡∂2NetU
∂H2= UCCC
2H − 2UC`βCH + UCCHH + U``β
2 + VZZ p̃Gα2pG < 0, (A.2a)
a12 ≡∂2NetU
∂H∂F= (UCCCH − UC`β)CF + UCCHF + VZZ p̃GαpF , (A.2b)
a21 ≡∂2NetU
∂F∂H= (UCCCH − UC`β)CF + UCCHF + VZZ p̃FαpG, (A.2c)
a22 ≡∂2NetU
∂F 2= UCCC
2F + UCCFF + VZZ p̃FpF < 0. (A.2d)
In order to simplify the remaining analysis, we use the functional forms of U(C, `, Z)
and C(H,F ) to derive the following expressions:
CHF = (1− ρ)CHCFC
, (A.3a)
1Before deriving the total differential, we use the government’s budget constraint and (9) to expressZ as Z = I − pFF − pGαH.
1
CHH = −CHFF
H, CFF = −CHF
H
F, (A.3b)
UCC = −ηUCC, UC` = η
UC`, (A.3c)
U` =η
1− ηUCC
`, U`` = −ηUCC
`2. (A.3d)
Using Equations (A.2a)-(A.2d) and the expressions (A.3a)-(A.3d), we can calculate the
determinant of the matrix J , which is given by
|J | = a11a22 − a12a21
= −VZZUCC
[η(βCF )2ππτC +
(1− ρ)CHCF (p̃FF + p̃GαH) (pFF + pGαH)
HF
+ηβC
`
(βCF (pFπ + p̃FπτC ) + pF p̃F
βC
`
)]+U2CCHCFη(1− ρ)
HF
(1 +
βH
`
)2
> 0.
(A.4)
In simplifying Equation (A.4), we used the condition C = CHH + CFF which follows
from the constant returns property of the consumption basket C. Moreover, the term π
in Equation (A.4) denotes the opportunity cost of leisure and is positive by assumption,
i.e., π = (p̃FCH/CF − p̃Gα)/β > 0, while πτC = π(τF = σ = 0)/(1 + τC) and equals
the lowest value that π can take (since π is increasing in τF and σ according to (13a)
and (13b)) and, thus, πτC > 0. Note that the matrix J is the Hessian matrix and the
second-order conditions require a11 < 0, a22 < 0, and |J | > 0, which is unambiguously
satisfied according to Equations (A.2a), (A.2d) and (A.4). In order to derive the effects
of tax changes on optimal consumption, we apply Cramer’s rule. Consider first a change
in the fat tax dτF 6= 0. It has the following impact on consumption:
dF ∗
dτF=
1
|J |
∣∣∣∣∣ a11 0
a12 VZpF
∣∣∣∣∣ =VZpFa11|J |
< 0, (A.5)
dH∗
dτF=
1
|J |
∣∣∣∣∣ 0 a12VZpF a22
∣∣∣∣∣ = −VZpFa12|J |
(A.6)
= −VZpF
[UCCFC
(CH(1− ρ)− ηCH − ηβC/`) + VZZ p̃GαpF]
|J |> 0, if ρ > ρ.
According to Equation (A.5), the fat tax unambiguously lowers unhealthy consumption
F ∗. On the other hand, Equation (A.6) contains only positive terms with the exception
of the term containing (1−ρ). If ρ = 1, i.e. if the elasticity of substitution εC = 1/(1−ρ)
is infinite, then this term is zero and dH∗ > 0. On the other hand, if H and F are nearly
perfect complements, i.e., ρ → −∞, then the negative term dominates and dH∗ < 0.
2
Define the value of ρ that satisfies dH∗ = 0 as ρ and the corresponding elasticity of
substitution as εC = 1/(1− ρ). Then for εC > εC , healthy consumption is increasing in
the fat tax and for εC < εC , it is decreasing.
Next, we derive the effects of a change in the thin subsidy. Applying Cramer’s rule
to the system of equations (A.1), we get
dH∗
dσ=
1
|J |
∣∣∣∣∣ −VZpGα a120 a22
∣∣∣∣∣ = −VZpGαa22|J |
> 0, (A.7)
dF ∗
dσ=
1
|J |
∣∣∣∣∣ a11 −VZpGαa21 0
∣∣∣∣∣ =VZpGαa12|J |
(A.8)
=VZpGα
[UCCFC
(CH(1− ρ)− ηCH − ηβC/`) + VZZpGαp̃F]
|J |< 0, if ρ > ρ.
According to Equation (A.7), the thin subsidy unambiguously raises healthy consump-
tion. Equation (A.8) is proportional to the negative of Equation (A.6) if τF = σ = 0.
For positive τF and σ, (A.8) again behaves qualitatively opposite to (A.6). Thus, there
exists a value ρ̂, such that (A.8) is positive, zero or negative if ρ is smaller than, equal
to or larger than ρ̂, where ρ̂ = ρ for τF = σ = 0. Thus, we have proven the second part
of Proposition 1.
Lastly, we consider a change in the sales tax. It results in the following consumption
changes:
dH∗
dτC=
1
|J |
∣∣∣∣∣ VZpGα a12VZpF a22
∣∣∣∣∣ =VZ [pGαa22 − pFa12]
|J |(A.9)
=VZ|J |
[UCCFC
(ηCFβπτC +
ηpFCβ
`− CH(1− ρ)
(αpG
H
F+ pF
))+VZZp
2FαpG(τF + σ)
]> 0, if ρ > ρ̃, τF = σ = 0,
dF ∗
dτC=
1
|J |
∣∣∣∣∣ a11 VZpGα
a21 VZpF
∣∣∣∣∣ =VZ [pFa11 − pGαa12]
|J |(A.10)
= −VZ|J |
[UCC
(ηCHCFβπτC +
ηCβCF`
(pFCHCF
+ βπτC
)+ ηpF
(Cβ
`
)2
+CHCF (1− ρ)
(αpG + pF
F
H
))+ VZZp
2Gα
2pF (τF + σ)
]< 0, if τF = σ = 0.
Equation (A.9) is positive if the term in brackets is greater than zero. The term in
the last row of (A.9) is zero if there is no fat tax or thin subsidy. Moreover, πτC > 0
from Equation (13c). The only negative term is the one containing (1− ρ). Thus, if H
3
and F are perfect substitutes (ρ = 1), then dH∗ > 0. On the other hand, if the two
types of meals are nearly perfect complements (ρ → −∞), then dH∗ < 0. Define the
degree of substitutability that satisfies dH∗ = 0 as ρ̃ and the corresponding elasticity of
substitution as ε̃C = 1/(1− ρ̃). Then, a sales tax raises healthy consumption if εC > ε̃C ,
leaves it unchanged if εC = ε̃C and lowers H∗ if εC < ε̃C . Its effect on F ∗ is negative if
τF = σ = 0, as in this case all terms in Equation (A.10) are negative. �
B Proof of Proposition 2
The effect of the policy instrument x = τF , σ, τC on the weight gain is given by
dS
dx= ξ
(δdF ∗
dx+ ε
dH∗
dx
), x = τF , σ, τC . (B.1)
Since each policy instrument may simultaneously increase H∗ and lower F ∗, the sign
of Equation (B.1) is ambiguous. We calculate the effect of τF on obesity by inserting
Equations (A.5) and (A.6) in (B.1). We derive after some manipulation
dS
dτF= ξ
VZpF|J |
{−UCC
[η
(CH +
βC
`
)(δ
(CH +
βC
`
)− εCF
)+ CHCF (1− ρ)
(δF
H+ ε
)](B.2)
+VZZ p̃Gα(δpGα− εpF )
}< 0, if δ
(CH +
βC
`
)> εCF ∧ δpGα > εpF .
The right-hand side of Equation (B.2) consists of two terms in the first row and one
term in the second row. Only the second term in the first row containing (1 − ρ) is
unambiguously negative. A sufficient condition for dS/dτF < 0 is that the other two
terms are also negative. This proves the first part of Proposition 2.
Next, we derive the impact of a change in the thin subsidy. We insert Equations
(A.7) and (A.8) in (B.1) and derive after some manipulation
dS
dσ= ξ
VZpGα
|J |
{UCC
[ηCF
(εCF − δ
(CH +
βC
`
))+ CHCF (1− ρ)
(δ + ε
H
F
)](B.3)
+VZZ p̃F (δpGα− εpF )
}< 0, if δ
(CH +
βC
`
)> εCF ∧ δpGα > εpF ∧ ρ = 1.
Equation (B.3) consists of one nonnegative term in the first row (containing (1 − ρ))
and two other ambiguous terms. The ambiguous terms are negative under the same
conditions as in Equation (B.2). A further sufficient condition for dS/dσ < 0 is that
the nonnegative term is zero, i.e. ρ = 1 and εC →∞. Thus, we have proven the second
4
part of Proposition 2.
In order to prove the last part, we insert Equations (A.9) and (A.10) in (B.1). The
resulting expression can be simplified to
dS
dτC= ξ
VZ|J |
{UCC
[(εCF − δ
(CH +
βC
`
))(ηCβpF
`+ ηCFβπτC
)− CHCF (1− ρ)·
·(δ + ε
H
F
)(pGα + pF
F
H
)]− VZZpFpGα(τF + σ)(δpGα− εpF )
}< 0,
if δ
(CH +
βC
`
)> εCF ∧ τF = σ = 0. (B.4)
According to Equation (B.4), if there is no fat tax and no thin subsidy, then only one of
the conditions sufficient for (B.2) and (B.3) to be negative, is sufficient for dS/dτC < 0.
This proves the last part of Proposition 2. �
C General production function
This section extends the model by assuming a more general production function. We
replace the Leontief production function for H (defined in Equation (2)) by the CES
function
H = (ψkkν + ψGG
ν)1ν , (C.1)
where ψk, ψG > 0 and ν ≤ 1. The elasticity of substitution between time and ingredients
is 1/(1 − ν). The goods are perfect complements if ν → −∞ and perfect substitutes
if ν → 1. The main model considers only the case of perfect complements. Prior to
solving the model, we redefine the opportunity cost of time. Using Equation (3) and
the expression F = (C − CHH)/CF , we can rewrite (3) as
p̃FCF
C +H
k
(p̃GG
H− CHCF
p̃F
)k + Z = I + TR. (C.2)
Using furthermore the time constraint k = T − `, we can express the price of leisure as
π ≡ H
k
(CHCF
p̃F − p̃GG
H
). (C.3)
Note that in the special case of the Leontief production function (2), H/k = 1/β and
G/H = α, such that π from (C.3) coincides with its value in the main text.
5
The consumer solves the following maximization problem:
max`,G,Z
NetU = U(C, `, Z)− bS(F,H) s. t. (C.4)
k = T − `, (C.5)
F =I + TR− p̃GG− Z
p̃F. (C.6)
The three first-order conditions are given by
∂NetU
∂`= U` − UCCHHk + bξεHk = 0, (C.7)
∂NetU
∂G= UCCHHG − UCCF
p̃Gp̃F− bξ
(HG − δ
p̃Gp̃F
)= 0, (C.8)
∂NetU
∂Z= VZ − UCCF
1
p̃F+ bξ
1
p̃F= 0. (C.9)
To prove Proposition 1, we follow the same steps as in Appendix A. First, we totally
differentiate the first-order conditions (C.7)-(C.9) with respect to `,G, Z and the tax
instruments (as in Appendix A, we insert the government’s budget constraint in the
consumer’s budget constraint prior to differentiation): a11 a12 a13a21 a22 a23a31 a32 a33
︸ ︷︷ ︸
=J
d`∗
dG∗
dZ∗
=
0 0 0
b21 b22 b23b31 0 b33
dτF
dσ
dτC
, (C.10)
where
a11 ≡∂2NetU
∂`2= UCC(CHHk)
2 − 2UC`CHHk + UCCHHH2k + U`` + (UCCH − bξε)Hkk
= UCC(CHHk)2 − 2UC`CHHk + UCCHHH
2k + U`` +
U`Hk
Hkk < 0, (C.11a)
a12 ≡∂2NetU
∂`∂G= (U`C − UCCCHHk)
(CHHG − CF
pGpF
)− UCHk
(CHHHG − CHF
pGpF
)− (UCCH − bξε)HkG
= (U`C − UCCCHHk)
(CHHG − CF
pGpF
)− UCHk
(CHHHG − CHF
pGpF
)− U`Hk
HkG,
(C.11b)
a13 ≡∂2NetU
∂`∂Z= − 1
pF[(U`C − UCCCHHk)CF − UCCHFHk] , (C.11c)
a21 ≡∂2NetU
∂G∂`= (U`C − UCCCHHk)
(CHHG − CF
p̃Gp̃F
)− UCHk
(CHHHG − CHF
p̃Gp̃F
)
6
− (UCCH − bξε)HkG
= (U`C − UCCCHHk)
(CHHG − CF
p̃Gp̃F
)− UCHk
(CHHHG − CHF
p̃Gp̃F
)− U`Hk
HkG
(C.11d)
a22 ≡∂2NetU
∂G2= UCC
(CHHG − CF
p̃Gp̃F
)(CHHG − CF
pGpF
)+ UC
(CHHHG − CHFHG
(p̃Gp̃F
+pGpF
)+ CFF
p̃Gp̃F
pGpF
)+ (UCCH − bξε)HGG
= UCC
(CHHG − CF
p̃Gp̃F
)(CHHG − CF
pGpF
)+ UC
(CHHHG − CHFHG
(p̃Gp̃F
+pGpF
)+ CFF
p̃Gp̃F
pGpF
)+U`Hk
HGG < 0, (C.11e)
a23 ≡∂2NetU
∂G∂Z=
1
pF
[UCCCF
(CHHG − CF
p̃Gp̃F
)+ UC
(CHFHG − CFF
p̃Gp̃F
)],
(C.11f)
a31 ≡∂2NetU
∂Z∂`= − 1
p̃F[(U`C − UCCCHHk)CF − UCCHFHk] , (C.11g)
a32 ≡∂2NetU
∂Z∂G=
1
p̃F
[UCCCF
(CHHG − CF
pGpF
)+ UC
(CHFHG − CFF
pGpF
)],
(C.11h)
a33 ≡∂2NetU
∂Z2= VZZ +
UCCC2F + UCCFFp̃FpF
< 0. (C.11i)
The terms on the right-hand side of (C.10) are given by
b21 ≡ −∂2NetU
∂G∂τF= [UCCF − bξδ]
[− p̃GpF
p̃2F
]= −VZ p̃GpF
p̃F< 0, (C.12a)
b22 ≡ −∂2NetU
∂G∂σ= [UCCF − bξδ]
[−pGp̃F
]= −VZpG < 0, (C.12b)
b23 ≡ −∂2NetU
∂G∂τC= [UCCF − bξδ]
p̃FpG − p̃GpFp̃2F
=VZpGpF (τF + σ)
p̃F≥ 0, (C.12c)
b31 ≡ −∂2NetU
∂Z∂τF= [UCCF − bξδ]
[−pFp̃2F
]= −VZpF
p̃F< 0, (C.12d)
b33 ≡ −∂2NetU
∂Z∂τC= [UCCF − bξδ]
[−pFp̃2F
]= −VZpF
p̃F< 0. (C.12e)
We continue by simplifying Equations (C.11a)-(C.11i), where we use the expressions
(A.3a)-(A.3d) together with the following expressions regarding the production function
(C.1):
HkG = (1− ν)HkHG
H, Hkk = −HkG
G
k, HGG = −HkG
k
G. (C.13)
7
The simplification of the aij terms gives the following results:
a11 = −UCC
[η
(C
`+ CHHk
)2
+ (1− ρ)CHCFH2k
F
H+ (1− ν)
HGGηC2
Hk(1− η)`
], (C.14a)
a12 =UCC
[η
(C
`+ CHHk
)(CHHG − CF
pGpF
)+ (1− ρ)CHCFHk
(F
HHG +
pGpF
)− (1− ν)
HGηC2
H(1− η)`
], (C.14b)
a13 =UCCpF
[−ηCF
(C
`+ CHHk
)+ (1− ρ)CHCFHk
], (C.14c)
a21 =UCC
[η
(C
`+ CHHk
)(CHHG − CF
p̃Gp̃F
)+ (1− ρ)CHCFHk
(F
HHG +
p̃Gp̃F
)− (1− ν)
HGηC2
H(1− η)`
], (C.14d)
a22 = −UCC
[η
(CHHG − CF
p̃Gp̃F
)(CHHG − CF
p̃Gp̃F
)+ (1− ρ)CHCF
F
H
(HG +
H
F
pGpF
)(HG +
H
F
p̃Gp̃F
)+ (1− ν)
HGkηC2
HG(1− η)`
],
(C.14e)
a23 =UCCFCpF
[η
(CHHG − CF
p̃Gp̃F
)− (1− ρ)CH
(HG +
H
F
p̃Gp̃F
)], (C.14f)
a31 =UCCp̃F
[−ηCF
(C
`+ CHHk
)+ (1− ρ)CHCFHk
], (C.14g)
a32 =UCCFCp̃F
[η
(CHHG − CF
pGpF
)− (1− ρ)CH
(HG +
H
F
pGpF
)], (C.14h)
a33 = VZZ −UC
Cp̃FpF
[ηC2
F + (1− ρ)CHCFH
F
]. (C.14i)
First, we show that the determinant |J | is negative, as required by the second-order
conditions. After some tedious calculations, we derive
|J | = −U3C(1− ν)HGη
2(1− ρ)CHC2FT
p̃2FpF (1− η)`3GF
[π`+
p̃FCF
C +`
k(p̃FF + p̃GG)
]+ VZZU
2C ·
·{
(1− ρ)CHCFηF
H
[1
`
(HG +
HpGFpF
)+HkpGFpF
] [1
`
(HG +
Hp̃GF p̃F
)+Hkp̃GF p̃F
]+
(1− ν)HGkη
H(1− η)`G
[η
(C
`+CFp̃F
πτC
)(C
`+CFp̃F
π
)+(1− ρ)CHCF
F
H
(H
k+GHpGkFpF
)(H
k+GHp̃GkF p̃F
)]}< 0. (C.15)
8
We can now calculate the effect of the fat tax on at-home meals H∗. Applying the
Cramer rule, we get
dH∗
dτF= −Hk
d`∗
dτF+HG
dG∗
dτF(C.16)
=1
|J |
−Hk
∣∣∣∣∣∣∣0 a12 a13
−VZpGpFp̃F
a22 a23
−VZpFp̃F
a32 a33
∣∣∣∣∣∣∣+HG
∣∣∣∣∣∣∣a11 0 a13a21 −VZpGpF
p̃Fa23
a31 −VZpFp̃F
a33
∣∣∣∣∣∣∣
=VZU
2C
|J |p̃F
{(1− ν)HGη
Gk(1− η)`
[(1− ρ)CHCFH − ηCF
(CHH + k
C
`
)]+
(1− ρ)CHCFηH2G
`2
}− VZ p̃GpFVZZUC
|J |p̃FC
[(1− ρ)CHCFH
2k
pGpF
−η(C
`+ CHHk
)(HkCF
pGpF
+HGC
`
)− (1− ν)HGηC
2
(1− η)`k
].
All the negative terms in (C.16) are proportional to (1−ρ). They vanish for ρ→ 1 and
tend to infinity for ρ→ −∞. Hence, there exists a value ρ such that dH∗/dτF > (< 0)
for ρ > (<)ρ.
To derive the change in F ∗, we use the budget constraint and apply Cramer’s rule:
dF ∗
dτF=
1
pF
[−pG
dG∗
dτF− dZ∗
dτF
](C.17)
=1
|J |pF
−pG∣∣∣∣∣∣∣a11 0 a13a21 −VZpGpF
p̃Fa23
a31 −VZpFp̃F
a33
∣∣∣∣∣∣∣−∣∣∣∣∣∣∣a11 a12 0
a21 a22 −VZpGpFp̃F
a31 a32 −VZpFp̃F
∣∣∣∣∣∣∣
=VZU
2C
|J |p̃F
{(1− ν)HGη
H(1− η)`
[(1− ρ)CHCF
HF
kG+ηk
G
(CH
H
k+C
`
)2]
+(1− ρ)CHCFηH
2GF
H`2
}+VZpGp̃GVZZa11|J |p̃F
< 0.
According to (C.17), the effect of the fat tax on the demand for away-from-home food
is unabiguously negative. This concludes the proof of Proposition 1 (i).
To prove Proposition 1 (ii), we follow the same steps and derive the following results:
dH∗
dσ= −VZpGU
2C
|J |p̃FpF
{(1− ν)HGη
k(1− η)`
[(1− ρ)CHCF
H
F+ ηC2
F
]+
(1− ρ)CHCFηHGH
F`
(1
`+Hk
H
)}− VZpGVZZUC
|J |C
[(1− ρ)CHCFH
2k
pGpF
−η(C
`+ CHHk
)(HkCF
pGpF
+HGC
`
)− (1− ν)HGηC
2
(1− η)`k
], (C.18)
9
dF ∗
dσ=
VZU2CpG
|J |p̃FpF
{(1− ν)HGη
H(1− η)`
[ηCF
(CH
H
k+C
`
)− (1− ρ)CHCF
H
k
]−(1− ρ)CHCFηHG
`
(1
`+Hk
H
)}+VZp
2GVZZa11|J |pF
. (C.19)
There is one negative term in (C.18) which is proportional to 1− ρ. If ρ is sufficiently
large, the term becomes small and dH∗/dσ > 0. Otherwise, dH∗/dσ is undetermined.
All the positive terms in (C.19) are proportional to (1− ρ). Hence, dF ∗/dσ is positive
for ρ→ −∞ and negative for ρ→ 1. Hence, a critical value ρ′ exists, such that dF ∗/dσ
is negative (positive) if ρ > (<)ρ′.
Analogously, we derive the effect of τC on the demand for H and F :
dH∗
dτC=
VZU2C
|J |p̃F
{(1− ν)HGη
(1− η)`
[(1− ρ)CHCF
(H
Gk+Hp̃GF p̃F
)− ηCF
G
(CFHπ
kp̃F+C
`
)]+
(1− ρ)CHCFηHG
`
[1
`
(HG +
Hp̃GF p̃F
)+Hkp̃GF p̃F
]}+(τF + σ)
VZpGpFVZZUC|J |p̃FC
[(1− ρ)CHCFH
2k
pGpF
−η(C
`+ CHHk
)(HkCF
pGpF
+HGC
`
)− (1− ν)HGηC
2
(1− η)`k
]+(τF + σ)
VZpGpFU2C
|J |p̃F 2pF
[(1− ρ)CHCFηHGH
F`
(1
`+Hk
H
)+
(1− ν)HGη
(1− η)k`
(ηC2
F + (1− ρ)CHCFH
F
)], (C.20)
dF ∗
dτC=
VZU2C
|J |p̃F
{(1− ν)HGη
H(1− η)`
[ηk
G
(CH
H
k+C
`
)(CFHπ
kp̃F+C
`
)+(1− ρ)CHCF
F
G
(H
k+HGp̃GFkp̃F
)]+
(1− ρ)CHCFηHGF
H`
[1
`
(HG +
Hp̃GF p̃F
)+Hkp̃GF p̃F
]}−(τF + σ)
VZpG|J |p̃F
{VZZa11pG −
U2C
p̃F
[(1− ρ)CHCFηHG
`
(1
`+Hk
H
)+
(1− ν)HGη
H(1− η)`
(ηCF
(CHH
k+C
`
)+ (1− ρ)CHCF
H
k
)]}. (C.21)
When τF = σ = 0, the terms in the third to sixth rows in (C.20) vanish. The only
negative terms in the first and second rows of (C.20) are proportional to (1 − ρ).
Following the same steps as in part (i) of Proposition 1, it follows that for τF = σ = 0,
there exists ρ̃, such that the demand for at-home food increases (declines) following an
increase in τC if ρ > (<)ρ̃. It is futhermore easy to see that in the case τF = σ = 0, all
10
terms in (C.21) are negative and dF ∗/dτC < 0. This concludes the proof of Proposition
1.
To prove Proposition 2, we follow the same steps as in Appendix B. Using Equation
(B.1), we derive the following effect of τF on the weight gain S:
dS
dτF= ξ
VZU2C
|J |p̃F
{(1− ρ)CHCFηH
2G
H`2(δF + εH) +
(1− ν)HGη
H(1− η)`· (C.22)
·[ηk
G
(CH
H
k+C
`
)(δ
(CH
H
k+C
`
)− εCF
H
k
)+
(1− ρ)CHCFH
kG(δF + εH)
]}−ξVZZ p̃GUC|J |p̃FC
{(1− ν)HGηC
2
(1− η)`k
(δpG
G
H− εpF
)+ (1− ρ)CHCFH
2kpG
(F
Hδ + ε
)+η
(CHHk +
C
`
)[HkpGk
H
(δ
(CH
H
k+C
`
)− εCF
H
k
)+CHG
`
(δpG
G
H− εpF
)]}< 0, if δ
(CH
H
k+C
`
)> εCF
H
k∧ δpG
G
H> εpF .
The conditions for a negative dS/dτF in (C.22) are identical to the conditions in (B.2)
when one substitutes β for H/k and α for G/H in the case of a Leontief production
function.
The effect of the thin subsidy on the gain weight is derived analogously and equals
dS
dσ= ξ
VZU2CpG
|J |p̃FpF
{−(1− ρ)CHCFηHG
F`(δF + εH)
(1
`+Hk
H
)+
(1− ν)HGη
H(1− η)`· (C.23)
·[ηCF
(δ
(CH
H
k+C
`
)− εCF
H
k
)− (1− ρ)CHCFH
Fk(δF + εH)
]}−ξVZpGVZZUC
|J |pFC
{(1− ν)HGηC
2
(1− η)`k
(δpG
G
H− εpF
)+ (1− ρ)CHCFH
2kpG
(F
Hδ + ε
)+η
(CHHk +
C
`
)[HkpGk
H
(δ
(CH
H
k+C
`
)− εCF
H
k
)+CHG
`
(δpG
G
H− εpF
)]}< 0, if δ
(CH
H
k+C
`
)> εCF
H
k∧ δpG
G
H> εpF ∧ ρ→ 1.
Similarly to the fat tax case, the conditions for a negative dS/dσ in (C.23) are analogous
to the conditions in (B.3).
Lastly, the impact on S of a change in the sales tax is the following:
dS
dτC= ξ
VZU2C
|J |p̃F
{(1− ρ)CHCFηHG
H`
[1
`
(HG +
Hp̃GF p̃F
)+Hkp̃GF p̃F
](δF + εH) +
(1− ν)HGη
H(1− η)`·
·[ηk
G
(CFHπ
kp̃F+C
`
)(δ
(CH
H
k+C
`
)− εCF
H
k
)
11
+(1− ρ)CHCFH
kG(δF + εH)
F p̃F +Gp̃GF p̃F
]}+(τF + σ)ξ
VZpG|J |p̃F
εpF
{VZZUC|J |C
[(1− ρ)CHCFH
2k
pGpF
−η(C
`+ CHHk
)(HkCF
pGpF
+HGC
`
)− (1− ν)HGηC
2
(1− η)`k
]+
U2C
p̃FpF
[(1− ρ)CHCFηHGH
F`
(1
`+Hk
H
)+
(1− ν)HGη
(1− η)k`
(ηC2
F + (1− ρ)CHCFH
F
)]}−(τF + σ)ξ
VZpG|J |p̃F
δ
{VZZa11pG −
U2C
p̃F
[(1− ρ)CHCFηHG
`
(1
`+Hk
H
)+
(1− ν)HGη
H(1− η)`
(ηCF
(CHH
k+C
`
)+ (1− ρ)CHCF
H
k
)]}< 0, if δ
(CH
H
k+C
`
)> εCF
H
k∧ τF = σ = 0. (C.24)
Equation (C.24) is negative under analogous conditions to Equation (B.4) in Appendix
B. This concludes the proof of Proposition 2 in the case of a general production function.
D Derivation of price elasticities
In this Section, we derive the price elasticities reported in Table 4. We use the estimates
of Okrent and Alston (2012), which are shown in Table D.1.
The category “FAFH and Alcohol” includes food-away-from-home and alcoholic
drinks. However, Okrent and Alston (2012) report that alcoholic drinks represent only
15.5% of the expenditures in that category. For that reason, we assume that this
category corresponds to food of type F in our model. Thus, the own-price elasticity of
the category “FAFH and Alcohol” gives εFF . There are six food-at-home categories,
which together constitute our food of type H. To aggregate their own- and cross-price
elasticities, we use their expenditure shares as weights. These shares are reported by
Okrent and Alston (2012), and we show them in Table D.2. In the first row of Table
D.2, we report the average share of food category i in expenditures in the period 1998-
2010, as reported by Okrent and Alston (2012). In the second row, we normalize the
sum of the expenditure shares to equal unity, such that we can calculate the share
of each category within the food group H. Thus, the value 0.152 for the normalized
expenditure share of cereals and bakery says that 15.2% of the overall expenditures for
H are spent on cereals and bakery. Use index i, j = 1, . . . , 6 to denote the six food-at-
home categories and denote as sj the normalized expenditure share of category j. We
12
Table D.1: Own-price and cross-price elasticities
With respect to price of
Elasticity ofdemand fora
Cerealsandbakery
Meat andeggs Dairy
Fruitsand veg-etables
NonalcoholicBever-ages
OtherFAHb
FAFHandalcoholc
Cereals andbakery
−0.58(0.25) 0.05(0.12) 0.36(0.08) −0.31(0.13) −0.09(0.15) 0.25(0.29) 0.16(0.34)
Meat andeggs
0.03(0.07) −0.31(0.17) 0.02(0.05) 0.11(0.09) 0.08(0.04) 0.26(0.11) 0.17(0.32)
Dairy 0.49(0.10) 0.06(0.12) −0.05(0.09) −0.03(0.11) −0.16(0.07) −0.44(0.16) 0.23(0.30)
Fruits andvegetables
−0.30(0.13) 0.18(0.15) −0.02(0.08) −0.79(0.19) 0.02(0.09) 0.58(0.20) 0.24(0.39)
Nonalcoholicbeverages
−0.21(0.32) 0.31(0.16) −0.25(0.12) 0.05(0.20) −0.65(0.39) 0.65(0.46) −0.05(0.51)
Other FAH 0.15(0.18) 0.27(0.11) −0.19(0.07) 0.36(0.12) 0.18(0.12) −0.98(0.30) 0.45(0.34)
FAFH andalcohol
0.03(0.07) 0.05(0.11) 0.03(0.04) 0.05(0.08) −0.01(0.05) 0.14(0.11) −0.71(0.38)
Nonfood −0.02(0.01) −0.05(0.01) −0.02(0.00) −0.02(0.01) −0.01(0.00) −0.04(0.01) −0.06(0.04)
a Source: Okrent and Alston (2012). Standard errors in parentheses.b FAH = Food-at-home.c FAFH = Food-away-from-home.
calculate the remaining food products elasticities according to the following formulae:
εHH =6∑i=1
(6∑j=1
sjεji
), (D.1)
εHF =6∑j=1
sjεjF , (D.2)
εFH =6∑j=1
εFj, (D.3)
where εij denotes the elasticity of food category i with respect to the price of category
j.
The elasticity of non-food products with respect to the price of F εZF is taken
directly from Table C1. The second non-food price elasticity is derived according to
εZH =6∑j=1
εZj, where j = 1, . . . , 6 denotes the at-home food categories. The resulting
expressions are reported in Table 4.
13
Table D.2: Own-price and cross-price elasticities
Cerealsandbakery
Meat andeggs Dairy
Fruitsand veg-etables
NonalcoholicBever-ages
OtherFAH Total
Absoluteexpenditure sharea
1.66 1.21 2.88 1.69 0.75 2.76 10.95
Normalizedexpenditure share
0.152 0.263 0.11 0.154 0.068 0.252 1
a Source: Okrent and Alston (2012).
E Calibration: tables and figures
Figure E.1: Average kcal per meal at- and away-from-home for male and female individ-
uals with 95% confidence intervals. Source: Author’s calculations based on NHANES
2009− 2010 (CDC, 2010).
14
Figure E.2: Average kcal per meal at- and away-from-home for different BMI groups of
male and female individuals with 95% confidence intervals. Under- and normalweight if
BMI less than 25, overweight if BMI ∈ [25, 30) and obese if BMI ≥ 30. The under- and
normalweight individuals are represented in one category due to the small number of
underweight individuals in NHANES. Source: Author’s calculations based on NHANES
2009− 2010 (CDC, 2010).
15
Figure E.3: Average kcal per meal at- and away-from-home for different age groups of
male and female individuals with 95% confidence intervals. Source: Author’s calcula-
tions based on NHANES 2009− 2010 (CDC, 2010).
16
Figure E.4: Average kcal per meal at- and away-from-home for different income groups
of male and female individuals with 95% confidence intervals. Income is defined as “ratio
of family income to poverty threshold” (variable INDFMPIR in NHANES), where a
value 1 indicates income equal to the poverty threshold. Source: Author’s calculations
based on NHANES 2009− 2010 (CDC, 2010).
Table E.1: Revenue Neutral Introduction of a Fat Tax and a Thin Subsidy
τF = 10%, σ = 6.7%
Male Female
Consumption, weight and
utility effectsa:
%∆F : −12.1 −6.44
%∆H: 3.31 1.28
%∆W : −3.24 −1.39
CV ($/day): 0.0096 0.034
EB/kcal (cent/kcal): 0.048 0.057a Source: Author’s calculations.
17
Table E.2: Sensitivity Analysis: εC = 2 and εC = 5
τF = 10% σ = 10% τC = 10%
Male Female Male Female Male Female
A. εC = 2
Consumption, weight andutility effectsa:%∆F : −4.21 −2.19 −0.74 −0.23 −3.49 −1.96%∆H: 0.54 0.17 1.38 0.56 −0.8 −0.38%∆W : −1.67 −0.68 0.85 0.36 −2.52 −1.04
CV ($/day): 0.4 0.35 −0.55 −0.46 0.95 0.81EB/TR: 0.022 0.011 0.007 0.0027 0.0099 0.0054EB/kcal (cent/kcal): 0.026 0.034 − − 0.019 0.025
B. εC = 5
Consumption, weight andutility effectsa:%∆F : −13.85 −7.85 −3.89 −1.5 −10.34 −6.44%∆H: 2.76 1.13 2.34 0.82 0.51 0.32%∆W : −4.64 −2.04 0.11 0.095 −4.86 −2.16
CV ($/day): 0.38 0.34 −0.56 −0.46 0.94 0.81EB/TR: 0.078 0.042 0.011 0.004 0.021 0.013EB/kcal (cent/kcal): 0.03 0.041 − − 0.02 0.03a Source: Author’s calculations.
Fat Tax
Sales Tax
2 3 4 5ϵC
0.018
0.020
0.022
0.024
0.026
0.028
EB/kcal
(a)
Fat Tax
Sales Tax
2 3 4 5ϵC
0.020
0.025
0.030
0.035
0.040
EB/kcal
(b)
Figure E.5: Excess burden per kcal reduction in consumption of men (Panel (a))
and women (Panel (b)) as functions of the elasticity of substitution εC , measured in
cent/kcal.
18
Fat Tax
Sales Tax
2 3 4 5ϵC
0.02
0.04
0.06
0.08
EB/TR
(a)
Fat Tax
Sales Tax
2 3 4 5ϵC
0.01
0.02
0.03
0.04
EB/TR
(b)
Figure E.6: Excess burden per one dollar of tax revenues for men (Panel (a)) and
women (Panel (b)) as functions of the elasticity of substitution εC .
(a) (b)
Figure E.7: Excess burden per kcal reduction in consumption of men (Panel (a)) and
women (Panel (b)) as functions of the kcal per meal δi and εi, where the domains of δi
and εi are defined by their 95% confidence intervals. The upper (red) plane represents
the fat tax and the lower (blue) plane the sales tax.
19
(a) (b)
Figure E.8: Excess burden per one dollar of tax revenues for men (Panel (a)) and
women (Panel (b)) as functions of the kcal per meal δi and εi, where the domains of δi
and εi are defined by their 95% confidence intervals. The upper (red) plane represents
the fat tax and the lower (blue) plane the sales tax.
20