Master Advanced Course: Public Economics:

University of Bonn, Spring term 2024

Pavel Brendler

Lecture 2:
Distortionary Taxation of Commodities

Contents
1 Introduction 2

2 Environment 3
2.1 Households . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Government . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Uncompensated and compensated demand functions 5

3.1 Uncompensated demand function . . . . . . . . . . . . . . . . 5
3.2 Compensated demand function . . . . . . . . . . . . . . . . . 6
3.3 Income and substitution effects . . . . . . . . . . . . . . . . . 6
3.4 Graphical derivation of the Hicksian and Marshallian demand
curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Deadweight losses 8
4.1 Measuring deadweight losses using indifference curves . . . . . 14
4.2 Measuring deadweight loss using compensated demand curves 14

5 Ramsey taxation 17
5.1 Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Derivation of the Ramsey tax formula . . . . . . . . . . . . . 20

1
 Readings (not obligatory):
• Stiglitz, Joseph and Jay Rosengard (2015): Economics of the public
sector , W.W. Norton & Company. Ch. 7 Efficiency and Equity, Ch.
19 Taxation and Economic Efficiency, Ch. 20 Optimal taxation.

• Jehle, Geoffrey A. and Philip J. Reny (2010): Advanced Microeconomic

Theory, Pearson. Ch. 1 Consumer Theory.

1 Introduction
In the previous lecture Baseline General Equilibrium Model, we learned that
the government intervention is justified in two cases:
1. When the First Welfare Theorem fails, i.e., when the competitive equi-
librium is not Pareto efficient, then the government intervention could
potentially lead to a Pareto improvement.

2. Even if the First Welfare Theorem applies, the government may want
to intervene to implement its preferences for inequality. In this case,
the Second Welfare Theorem tells us that the government can achieve
any allocation on the Pareto frontier by setting appropriate lump-sum
transfers and lump-sum taxes.
However, implementing a lump-sum tax in the real world is challenging be-
cause governments do not observe the household’s type (ability, health, etc.).1
For example, if the government promised a lump-sum transfer to all dumb
households and a lump-sum tax to all smart households, everyone would have
incentives to claim they are dumb. As a result, nobody would be working,
and there would be no means to finance any transfers. This situation is re-
ferred to as the informational friction or the informational constraint of the
government.
As you can see, the Second Welfare Theorem shows how an optimal allo-
cation can be decentralized using lump-sum taxes and lump-sum transfers,
but these instruments are missing in the real world. Therefore, the key as-
sumption we maintain throughout this lecture is that lump-sum taxes are
1
Recall that a lump-sum tax is a fixed amount of resources taken away from a consumer
that doesn’t depend on the consumer’s actions but only on those characteristics that are
outside of the household’s control.

2
not available to the government. Instead, the government must resort to
alternative policy instruments, such as proportional or progressive income
taxes, consumption taxes, etc. Such tax schemes directly link the tax bur-
den to labor income or consumption. However, ALL these instruments will
be distortionary and located underneath the Pareto frontier that we derived
for the social planner’s problem in the previous lecture.
Recall the “Four Central Questions in Public Economics” from the intro-
ductory lecture. The goal of this lecture is to answer the WHAT question,
i.e., What is the distortionary effect of interventions on economic outcomes?
There are two types of taxable transactions: consumption of goods and
factor incomes. In the literature, taxation of consumption goods is referred to
as indirect taxation; taxation of factor prices is referred to as direct taxation.
In this lecture, we will focus on indirect taxation. More specifically, we will
derive the optimal Ramsey tax on commodities that minimizes the total
deadweight loss in the economy. In a later Lecture: Optimal Linear Labor
Income Taxation, we will analyze the optimal taxation of labor income. You
will notice that intuition is very similar in both cases.
Below we proceed as follows. In sections 2-4, we will understand how
consumers respond to taxes and why distortions (equivalently, deadweight
losses or excess burden) arise. The analysis of how taxation changes the
consumer’s behavior is referred to as Positive analysis in public finance. We
do not judge what the tax should be; we solely document the effects of taxes
on the agents’ behavior. Once we learn this behavior, we will turn in section
5 to the Normative analysis and discuss the optimal design of a tax system.

2 Environment
Let’s start off with the baseline general equilibrium model from Lecture 1:
Baseline General Equilibrium Model (see Section Market solution in those
lecture notes). We will consider a simplified version of that set-up. In par-
ticular, there is only one household type but two sectors of production (and
therefore two types of goods). We choose two sectors of production because
our goal is to analyze the effect of a change in relative prices due to a tax. At
this stage, our goal is solely to understand the distortionary nature of taxes
and to do this, heterogeneity among households is not necessary. Later in
this course, when we analyze how taxes and transfers help reduce inequality,
we will introduce two different types of households.

3
2.1 Households
Suppose that there is a continuum of identical households. Therefore, we
drop subscript j. However, there are two types of goods indexed by i. As
opposed to the notation in the lecture, we index goods by a subscript (not
a superscript) i. Household takes the prices of the output goods p1 , p2 as
given. Let’s normalize household’s productivity to e = 1. Suppose that the
household always works one hour, i.e. the labor supply is exogenous with
l = 1, so that household’s labor income is simply given by w.
There is a government who imposes a linear excise tax t1 on good 1.
Excise taxes is a per unit tax (a fixed tax per package of cigarettes, for
example). Excise taxes are collected from producers or wholesalers, and are
embedded in the price paid by the consumer. Since the consumers bear the
full amount of the excise tax, the price of the taxed good, p1 , rises by a full
amount of the tax. Denote by qi the producer price of good i. Then we must
have p1 = q1 + t1 . For the untaxed good, we must have p2 = q2 . Household’s
budget constraint becomes:

p1 x1 + p2 x2 = y, (1)

where y = w + T denotes the total resources available to the household –

the sum of the wage income and the lump-sum transfer T .2 Note that even
though we have lump-sum transfers, the taxes are not lump-sum! Therefore,
the second welfare theorem doesn’t apply.
Household’s utility over the two goods is given by the utility function
u(x1 , x2 ) with the usual assumption that ∂u(x1 , x2 )/∂xi > 0 and ∂ 2 u(x1 , x2 )/∂ (xi )2 <
0 for i = 1, 2. Note that since labor is fixed at 1, we exclude it from the utility
function.

2.2 Firms
There are infinitely many firms operating in the two sectors. Each firm in
sector i has access to the technology:

fi (Li ) = Ai Li .
2
In the previous lecture, I used y to denote a sector’s output. Do not confuse it with
the household’s total resources, also denoted by y, in the current lecture. The two have
nothing in common.

4
 Firms maximize their profits. Recall from problem set 1 that with a linear
technology above an equilibrium exists only if
w
= Ai , (2)
qi
where qi is the producer price of good i. Combining together the optimality
condition for each sector, we obtain:
q1 A2
= (3)
q2 A1
Recall that the labor supply by households is inelastic. At the same time,
any demand for labor satisfies the optimization problem of a firm in sector i
as long as w/qi = Ai . Therefore, the exogenous labor supply by households
pins down the equilibrium quantity of labor.

2.3 Government
The government collects taxes and rebates them back to the household as a
lump-sum transfer. The government’s budget constraint is:

t1 x1 = T (4)

3 Uncompensated and compensated demand

functions
In order to understand the distortionary nature of taxes that we will address
in the subsequent section, we first need to introduce two concepts.

3.1 Uncompensated demand function

The uncompensated demand function for goods can be obtained by solving
a utility maximization problem.
Taking the prices (p1 , p2 ) as well as the government transfer T as given,
households solve:
max u(x1 , x2 )
x1 ,x2

5
subject to the budget constraint (1). The first-order conditions lead to:

u1 (x1 , x2 ) p1 q1 + t1
M RS1,2 = = = (5)
u2 (x1 , x2 ) p2 q2

This condition together with agent’s budget constraint delivers the Marshal-
lian (uncompensated ) demand functions for two goods denoted by xi (p1 , p2 , y).

3.2 Compensated demand function

Alternatively, we can set up household’s expenditure minimization prob-
lem. The compensated demand curve gives the agent’s demand for a good
assuming that as the price of the good is increased due to the tax, income is
being given to the agent in such a way as to leave her on the same indifference
curve. The household solves:

min {p1 x1 + p2 x2 }
x1 ,x2

subject to:
u(x1 , x2 ) = ū
where ū is some desired level of utility. The Lagrangian to this problem is:

L = p1 x1 + p2 x2 + λ [ū − u(x1 , x2 )]

where λ is a Lagrangian multiplier on household’s utility constraint. The

first-order condition remains the same as in the utility maximization problem
(see eq. 5). This condition together with the agent’s budget constraint pin
down the Hicksian (compensated ) demand function for two goods, which we
denote by xci (p1 , p2 , ū). The minimum expenditure required to attain utility
level ū is e(p1 , p2 , ū).3

3.3 Income and substitution effects

Any change in the price for a good may change the demanded quantity
for the good. The change in the demanded quantity can be decomposed
into two effects: income and substitution effect. To illustrate both effects,
3
Do not confuse the expenditure function e(p1 , p2 , ū) with household’s ability endow-
ment e which we assume to be 1 throughout this lecture.

6
look at figure 1.4 The figure plots the household’s budget constraint in a
(x1 , x2 ) diagram. Let’s start with the case without taxes (p1 = q1 ). The
preferred consumption bundle is (xold old
1 , x2 ) which is located at the tangent
point between the original budget line and the household’s indifference curve.
At this point, the optimality condition
u1 (x1 , x2 ) p1 q1
M RS1,2 = = =
u2 (x1 , x2 ) p2 q2
holds. This condition characterizes the household’s ideal consumption be-
havior.
Suppose that the government introduces an excise tax on good 1, t1 . The
new preferred consumption bundle is now (xnew new
1 , x2 ). It is located at the
tangent point between the new budget line and the household’s indifference
curve. At this point, the optimality condition changes from the one we have
just derived to:
u1 (x1 , x2 ) p1 q1 + t1 q1
M RS1,2 = = = 6=
u2 (x1 , x2 ) p2 q2 q2
holds. It shows that the ratio between the consumed quantities of both goods
deviates from the ideal ratio!
The total change in consumption of each of the two goods due to the tax
on good 1 can be decomposed into two distinct effects: the substitution effect
and the income effect. Below we describe each effect.
1. Substitution effect (SE) is the amount by which individual’s con-
sumption is reduced due to the relative price change of the goods.
To compute the substitution effect, we give the consumer exact same
amount of money that she needs to reach the old indifference curve
but we keep the relative prices at the new level, i.e. after the tax
is introduced. The direction of substitution effect is always unique.
The magnitude of the substitution effect depends on the shape of in-
difference curves which is captured by the elasticity of substitution
between the two goods. If it is easy to substitute for the other good
(i.e. the cross-price elasticity of compensated demand, cxi ,pj with i 6= j
in eq. (16), is high in absolute terms), then the substitution effect is
large. The substitution effect is zero if preferences are of Leontief type:
u(x1 , x2 ) = min {ax1 , bx2 }, where a, b > 0 are some constants.
4
In the Appendix to these notes, we will derive both effects analytically.

7
 2. Income effect (IE). Any tax increase makes a consumer worse off
because it reduces the amount of goods that can be purchased. The
income effect is the amount by which individual’s consumption is re-
duced due to a change in a consumer’s purchasing power. The income
effect is the residual effect after computing the substitution effect. The
direction of income effect may be different depending on whether we
deal with inferior goods or normal goods. In the first case, the income
effect is negative as the tax increases. In the second case, the income
effect is positive as the tax increases.

3.4 Graphical derivation of the Hicksian and Marshal-

lian demand curves
Figure 3 illustrates how the Hicksian and Marshallian demand curves are
derived. Take a moment to understand how this is done.5
You can see from figure that the Hicksian and the Marshallian demand
curves look very similar, with the Marshallian demand curve being slightly
flatter. As the price of the good increases from q1 to q1 + t, the reduction
in the demanded quantity of the good is higher under Marshallian demand
function (the absolute difference between xold new
1 and x1 ) than under the Hick-
sian demand function (the absolute difference between xold c
1 and x1 ). This is
due to the additional income effect that reduces the amount of good 1 under
Marshallian demand function. This argument holds as long as the good is a
normal good; the opposite would be true (i.e. the Hicksian demand curve is
flatter than the Marshallian demand curve) if the good is an inferior good.

4 Deadweight losses
To understand the distortionary effects of the excise tax, it is useful to con-
trast the results with the case of a lump-sum tax. Suppose that the govern-
ment imposes a lump-sum tax Z on the consumer. Taking the prices (p1 , p2 )
5
As you can see from Figure 3, we assume that both demand functions are linear.
In reality, the demand curves might be highly non-linear. However, if the tax change is
sufficiently small, we can approximate the demand function at a given point using the
first-order Taylor approximation, which results in a linear function.

8
Figure 1: Income and substitution effects with an excise tax on good 1 (part
1)

9
Figure 2: Income and substitution effects with an excise tax on good 1 (part
2)

10
Figure 3: Graphical derivation of the Hicksian and the Marshallian demand
curves
11
as well as the government transfer Z as given, households solve:
max u(x1 , x2 )
x1 ,x2

subject to the budget constraint

p1 x1 + p2 x2 = y,
where y = w + Z denotes the total resources available to the household. The
first-order condition is, of course:
u1 (x1 , x2 ) p1 q1
M RS1,2 = = = . (6)
u2 (x1 , x2 ) p2 q2
Hence, the lump-sum tax doesn’t change the ideal ratio of quantities of both
goods consumed! This optimality condition is identical to the one we derived
in Lecture 1 meaning that the resulting allocation would be Pareto efficient.
Figure 4 illustrates the effect of a lump-sum tax. The lump-sum tax
induces a parallel downward shift of the budget line. The optimal bundle of
goods shifts from A to B. However, at each of these points the optimality
condition (6) continues to hold. As you can see, there is only the income effect
at work, with the substitution effect being equal to zero! If the government
rebates the tax revenues back to the consumer, it will restore the original
allocation.
This scenario is in stark contrast to the effect of an excise tax shown in
eq. (5). This optimality condition deviates from the one we derived in Lec-
ture 1 Baseline General Equilibrium Model (section Market solution). This
deviation is sometimes referred to as a wedge. Intuitively, the substitution ef-
fect persuades the consumer to move consumption away from the taxed good
toward the untaxed good. Even if the government rebates the tax revenues
to the consumer, she will not choose the original, ideal bundle. With an ex-
cise tax (or any other tax, except for a lump-sum tax), there are deadweight
losses induced by the substitution effect. In the next section, we will learn
to measure these losses.
Without excise tax, the first welfare theorem holds, and the competitive
equilibrium is Pareto efficient. With excise taxes, however, the allocation
will not be Pareto efficient, since the optimality condition (5) is not the same
as in (6). Note that the second welfare theorem doesn’t hold because the
government uses non lump-sum taxes to redistribute incomes!
Next we learn two approaches to measure the deadweight losses.

12
Figure 4: Effects of a lump-sum tax

13
4.1 Measuring deadweight losses using indifference curves
This method goes as follows:

1. Suppose that the government imposes a distortionary tax (in our case,
the excise tax) to collect a given amount of tax revenues denoted by R.

2. Compute the household’s utility given the distortionary tax, ū.

3. Suppose that the government is able to impose a lump sum tax instead
of the excise tax. Find the amount of the lump-sum tax, T , such that
the consumer achieves the same level of utility as under the excise tax
(ū).

4. The deadweight loss (or equivalently, the excess burden of taxa-

tion) is the difference in tax revenues between the case of lump-sum
taxes and excise taxes:
T − R.
This difference T − R always non-negative. In other words, the gov-
ernment is able to extract more resources from a household using a
lump-sum tax than an excise tax (assuming that in both cases the
consumer maintains the same utility level). Intuitively, the amount
of taxes collected using the excise tax on a good is lower because the
substitution effect incentivizes the agent to consume relatively more of
the other (untaxed) good. By contrast, the relative quantities of the
goods consumed remain unchanged under lump-sum taxes.

You will practice this method in Problem set 2.

4.2 Measuring deadweight loss using compensated de-

mand curves
Another way to measure distortions is to use directly the compensated (Hick-
sian) demand function.
The compensated demand function plays a very important role in eco-
nomics. It reflects changes in consumer demand for a good as a result of a
price change only due to substitution effect. In figure 3, the area enclosed
among the initial price (horizontal line at q1 ), the price after tax (horizontal
line at q1 + t1 ) and the Hicksian demand function measures the amount an

14
individual is willing to pay to have the price of good 1 reduced from q1 + t1
to q1 .
Consider figure 5. Focus on the Hicksian demand function only and ignore
the Marshallian demand function. Assume that the supply of good 1 is
infinitely elastic at producer price q1 (meaning that the supply curve is a
horizontal line at price q1 ). Recall from the basic microeconomics course
that the area enclosed between the demand curve and the supply curve is
the consumer surplus. Under the excise tax, the supply curve shifts upwards
by t1 as we can see in the figure. The area A + D + C denotes the total loss
in consumer surplus due to taxation. However, the deadweight loss does not
equal the total loss in consumer surplus. In fact, area A + D denotes the
tax revenue collected by the government, t1 × xc1 . Hence, this area reduces
the welfare losses because the government can rebate these resources back to
the consumer. The deadweight loss is, therefore, only area C. This area is
sometimes referred to as a Harberger triangle.
Let’s derive analytically the expression for the excess burden (area C
in figure 5). When it comes to the policy analysis, it is useful to express
the deadweight loss as a function of some simple statistic that we could also
measure in the data. Below, we will use the elasticity of the Hicksian demand
for good 1 in response to a change in price p1 as such statistic. Of course,
the Hicksian demand is not observable in the real world. In the Appendix
to these notes, you will see that we can alternatively apply the Marshallian
elasticity, measurable in the data, to obtain an upper bound on the size of
the deadweight loss. Please study the Appendix on your own.
We define the Hicksian price elasticity as follows:
4xc1 4p1
cx1 ,p1 = old / ,
x1 q1

where 4x1 = xc1 − xold 1 and 4p1 = pnew

1 − pold
1 = t1 . Note that sometimes
the elasticity is defined as an absolute value of the term above. However, we
do not follow this alternative definition, so bear in mind that cx1 ,p1 will be a
negative number.

15
Figure 5: Deadweight losses under the Hicksian demand function (ignore the
Marshallian demand function)

16
 The excess burden the reads:
1
EB c = × t1 × (xold c
1 − x1 )
2
1
= − × 4p1 × 4xc1
2
1 4p1 q1 q1 xold1
= − × 4p1 × 4xc1 × × × × old
2 4p q 1 q 1 x1
| 1 {z }
=1
1 4xc1 q1 4p1 1 q1 xold
= − × 4p1 × × × × × × 1
2 xold
1 4p1 1 q1 q1 1
| {z }
cx1 ,p1
 2
4p1
1
= − × cx1 ,p1 × q1 xold
1
2q1
 2
1 t1
= − × cx1 ,p1 × q1 xold
1 (7)
2 q1

Several observations emerge from eq. (7):

• Since cx1 ,p1 < 0, we have that EB c > 0.

• The excess burden rises quadratically in the tax rate t1 . This is to say
that distortions become over-proportionally more severe as taxes rise.

• The deadweight loss increases in the demand elasticity. If the elasticity

is high (in absolute terms), then a tax increase will incentivize the
consumers to switch away from the taxed good.

5 Ramsey taxation
After discussing the effects of taxation on consumer’s behavior (positive anal-
ysis), we move to the question of optimal taxation (normative analysis).
Suppose that there is a given level of government revenue to be raised
which must be financed solely by taxes upon commodities. When decid-
ing upon the taxes, the government is minimizing the total deadweight loss
caused by taxation (this is equivalent to saying that the government is max-
imizing the total utility of households). For a given amount of revenue to

17
be extracted from a consumer, what excise taxes would the government set?
This is the Ramsey problem of efficient commodity taxation, first addressed
in the 1920s by economist Frank P. Ramsey (1903-1930).
We will assume that all the tax revenues collected are going to be wasted
(i.e. not rebated back to the household). Think of these expenditures as
government spending on activities that are outside of the model, such as
pension system, education, defense, etc. This means that at this stage we will
ignore any distributional (equivalently, equity or equality) effects of taxation.
This simplifying assumption helps us focus solely on the efficiency aspect of
taxation (i.e. minimization of distortions). Later in this course, in the lecture
Equity-efficiency trade-off, we will add to the problem of choosing the optimal
tax rate the equity (i.e. equality) considerations, i.e. a motive to redistribute
incomes in order to reduce inequality. This will bring us to the prominent
equity-efficiency trade-off. Since we ignore the distributional effects, we must
require that the government has to pay for these activities. Otherwise, of
course, the government would set all taxes equal to zero, given that it is
minimizing the total deadweight loss in the economy, and the entire problem
would be trivial and uninteresting.

5.1 Environment
Assume for simplicity that there is a single consumer and N different con-
sumption goods. For good i, the (inverse) Hicksian demand function is given
by:
pi (xci ) = ai − bi xci , (8)
where xci is the quantity of good i consumed under the compensated demand
function, pi is the price paid by the consumer and ai , bi are constants. As
before, we assume that the supply of good i is infinitely elastic at producer
price qi (meaning that the supply curves are horizontal lines at price qi ).
Figure 6 illustrates the demand and the supply curve for good i.
Observe that the demand function in eq. (8) implies that the Hicksian
cross-elasticities are zero. This assumption is very important. It simplifies
algebra, since we exclude the case when a tax on good i might affect the

18
Figure 6: Compensated demand curve for good i

19
demand for good j.6
The last step we need to make before deriving the Ramsey formula is to
obtain the price elasticity of demand evaluated at price qi (i.e. before tax
increase):
4xci pi 1 qi qi
cxi ,pi |pi =qi = × c = − × ai −qi = − ,
4pi xi bi bi
ai − q i

In deriving this expression, we used the (direct) Hicksian demand function,

xci (pi ) = (ai − pi )/bi .
Note that instead of postulating the demand function in (8), we could
have instead specified consumer’s utility function over N goods. Afterwards,
we would have derived the Hicksian demand functions as a solution to the
household’s expenditure minimization problem and then proceed as we will
do below. Instead, we take a short-cut and directly postulate the Hicksian
demand function.

5.2 Derivation of the Ramsey tax formula

We assume that the government has to collect R in tax revenues from the con-
sumer. In doing so, the government imposes a set of excise taxes (t1 , t2 , ..., tN )
on N goods. The objective of the government is to minimize the total dead-
weight loss that arises from taxation, so we need to derive it first before we
set up government’s problem.
The deadweight loss that arises from taxing good i is the shaded area in
figure 6. This area is formally given by:
1
EBic = × ti × (xold
i − xi
new
)
2  
1 ai − qi ai − qi − ti
= × ti × −
2 bi bi
1 ti
= × ti ×
2 bi
2
1 t
= × i
2 bi
6
Independent goods are goods that have a zero cross elasticity of demand. Recall
one example of preferences which implies independent goods – the Cobb-Douglas utility
1−α
function u(x1 , x2 ) = xα
1 x2 . In this case, the optimal demand for goods is given by:
x1 (p2 , y) = αy/p1 and x2 (p1 , y) = (1 − α)y/p2 , where y is the agent’s income.

20
where EBic denotes the excess burden under the Hicksian demand function
for good i. Note that in the previous section, when we introduced the excess
burden under the Hicksian demand function, we used xci instead of xnew i .
Both are the same.
The objective of the government is therefore:
N
X
min EBic , (9)
{t1 ,t2 ,...,tN }
i=1

subject to its budget constraint:

N
X
ti xnew
i ≥R (10)
i=1

where R is the total amount of revenues to be raised. We can re-write the

inequality constraint as an equality constraint and furthermore substitute for
xnew
i to arrive at:
N  
X ai − qi − ti
ti =R (11)
i=1
b i

Alternatively, we could have let the government maximize the utility of

the household over N goods. This maximization problem, however, is equiva-
lent to the deadweight loss minimization problem in our case. The advantage
of the deadweight loss minimization approach is that we can explicitly see
how the distortions from taxation arise and measure them.
The Lagrangian function for the government’s minimization problem reads:
N  
" N #
1 X t2i

X ai − q i − t i
L= +λ R− ti
2 i=1 bi i=1
bi

where λ > 0 is a Lagrangian multiplier on the government budget constraint

(11).
The first-order condition with respect to ti reads:
ti ai − qi − ti ti
= λ −λ (12)
bi bi bi
ti ai − q i ti
⇒ = λ − 2λ (13)
bi bi bi

21
 Eq. (13) equates the marginal deadweight loss (left-hand side) to the
marginal revenue of a change in good i0 s tax rate. Observe that the terms
depend only on the parameters of good i (such as price qi , tax ti , demand
parameters ai , bi ) and not those of other goods because we assumed that the
Hicksian cross price elasticities are zero. The first term on the right-hand side
of eq. (13) shows the increase in the tax revenue when the demanded quantity
of good i is fixed. This effect is sometimes referred to as a mechanical (or
direct) effect of taxation. The second term on the right-hand side of eq. (13)
reflects the tax revenue loss because a tax increase decreases the demand for
good i. This effect is sometimes referred to as a behavioral (or indirect) effect
of taxation.
Rearranging the terms in eq. (13), we can derive the optimal tax rate:

ti ti qi ai − q i
+ 2λ = −λ × × (− )
bi bi qi bi
ti qi 1
[1 + 2λ] = −λ × ×
bi bi xi ,pi
λ 1
ti = − × qi × (14)
1 + 2λ xi ,pi
Observe that since xi ,pi < 0 and λ > 0, we have that ti > 0.
Summarizing, an elastically demanded good has a high marginal dead-
weight loss plus it is a poor source of tax revenue. Therefore, the theory of
optimal taxation suggests that such a good should not be taxed too heavily.
This is the key idea behind the Ramsey optimal taxation formula in eq. (14).

Appendix
To compute the deadweight loss, we must use the Hicksian demand function.
However, the Hicksian demand is unobservable in the real world. By contrast,
the Marshallian demand function can be estimated in the data. The key
question is: By how much would we underestimate/overestimate the true
deadweight loss if we used the Marshallian demand function?
We can derive an analytical expression for excess burden (EB) under the
Marshallian demand function. Consider figure 7 (ignore the Hicksian demand
function). Here, the area A + B denotes the total loss in consumer surplus
due to taxation. Area A is the tax revenue collected by the government.

22
Area B is therefore the deadweight loss (excess burden) under Marshallian
demand.
To derive the excess burden analytically, first note that the elasticity of
the Marshallian demand for good 1 in response to a change in price p1 is
defined as:
4x1 4p1
x1 ,p1 = old / ,
x1 q1
where 4x1 = xnew
1 − xold
1 and 4p1 = p1
new
− pold
1 = q1 + t1 − q1 = t1 .
Then, the excess burden under the Marshallian demand function reads:
 2
M 1 t1
EB = − × x1 ,p1 × q1 xold
1 . (15)
2 q1
Finally, let’s compare the excess burden under the Marshallian demand
function to the true deadweight loss under the Hicksian demand function.
To do so, we need first to introduce a new concept – the Slutsky equation.

Slutsky equation
It can be shown that:
∂xi (p1 , p2 , y) ∂xci (p1 , p2 , ū) ∂xi (p1 , p2 , y)
= − × xj (p1 , p2 , y)
∂pj ∂pj ∂y
| {z } | {z }
substitution effect income effect
which is the familiar Slutsky equation. This equation decomposes analytically
the effect of a price increase for good j on the demand for good i into two
distinct effects.
We can re-write this equation in terms of elasticities:
p j xj
xi ,pj = cxi ,pj − ηxi · (16)
y
where
• xi ,pj – uncompensated price elasticity of good i with respect to price
pj ;
• cxi ,pj – compensated price elasticity of good i with respect to pj ;
• ηxi – income elasticity of good i with respect to y;
• pj xj /y – share of the total household’s budget spent on good j.

23
Figure 7: Deadweight losses under Marshallian demand function (ignore the
Hicksian demand function)

24
Comparing the deadweight losses
We can re-write the deadweight loss in (5) as follows:

 2
M 1 t1
EB = − × x1 ,p1 × q1 xold
1
2 q1
 2 
q1 xold

1 t1 c 1
= − × x1 ,p1 − ηx1 · × q1 xold
1
2 q1 y
 2
1 t1 q1 xold
1
= EB c + · ηx1 · × q1 xold
1
2 q1 y

Therefore, as long as good 1 is a normal good (i.e. ηx1 > 0), we have
EB M > EB c . Therefore, if the use the Marshallian demand function, we
will overestimate the excess burden. At the same time, if the income effect
is small, then both answers will be similar.

25