Public Economics

Lecture 2: Tools of Welfare Analysis

Julien Grenet
Paris School of Economics

September 13 & 14, 2022

M2 APE / PPD
Practical Information

E-mail: [email protected]

Office Hours:
– Monday, 13:30–14:30
– PSE, Office R3-12 (third floor)
– Please send an email to make sure I’m there
– You can also schedule an appointment

Lectures:
– Today and tomorrow: Tools of Welfare Analysis
– 20 & 21 September: Externalities
– 27 & 28 September: Public Goods
– 29 & 30 November: Social Insurance

1 / 111
This Lecture

Welfare economics provides the basis for judging the

achievements of markets and policy makers in allocating resources

In this lecture, we will review some important conceptual tools by

which economists assess the desirability of market outcomes and
the welfare and distributional effects of government intervention
– economic surplus
– deadweight loss
– tax incidence

These tools will be illustrated by examples drawn from recent

empirical studies (e.g., consumer surplus generated by Uber)

They will be used throughout the course to study the effects of

specific policy interventions (e.g., taxes/subsidies, regulations)

2 / 111
Tools of Welfare Analysis: Outline

1. The Concept of Economic Surplus

2. Competitive Equilibrium and Social Efficiency
3. Measuring Inefficiency: Deadweight Loss
4. The Efficiency Cost of Taxation
5. Tax Incidence
6. Appendix

4 / 111
The Concept of Economic
Surplus
The Concept of Economic Surplus
Economic surplus is a measure of the amount by which buyers
and sellers benefit from participating in the market

The total economic surplus is the sum of:

– Consumer surplus: benefit consumers derive from consuming a
good above and beyond the price they pay for the good
– Producer surplus: benefit derived by producers from selling a
good above and beyond the cost of producing that good
– Government revenue (if relevant)

Can be represented graphically in the supply/demand diagram

– Consumer surplus: area below the demand curve and above
the market price of the good.
– Producer surplus: area above the supply curve (marginal cost)
and below the market price of the good

6 / 111
 Producer and Consumer Surplus

Price

Consumer
Surplus

Supply S(p)

p0 E0

Demand D(p)

Producer
Surplus

Q0
7 / 111
Consumer Surplus
How to evaluate consumer welfare?

Problem: using utility levels is not very useful (“utils”)

What we would like is to express consumer’s welfare in a monetary

equivalent: difference between the amount of money you would be
willing to pay for a good and how much you actually pay for it

Consumer surplus (area between demand curve and market price)

is an approximate measure of consumer welfare

Two exact measures of changes in consumer welfare:

– Compensating variation (CV)
– Equivalent variation (EV)

∆CS, EV and CV coincide only in the case of quasilinear utility

8 / 111
Consumer Surplus: Model
Suppose that a consumer has m euros to spend on two goods:
– a commodity X whose price is p
– a numeraire good Y whose price is normalized to 1
(= amount of money spent on other goods)

Assume that the consumer’s utility function is quasilinear:

U(x, y ) = v (x) + y

where x is the quantity of a good X and y is the amount of money

(euros) spent on other goods. v 0 > 0, v 00 < 0 and v (0) = 0

The consumer solves:

max v (x) + y
x,y

s.t. p·x +y =m

9 / 111
Consumer Surplus: Model
Substituting out for y , the consumer’s program simplifies to:

max v (x) + m − p · x
x
FOC:
v 0 (x) = p

The FOC implicity defines the consumer’s Marshallian demand

for good X as a function of its price p:
x = D(p)
(Note: quasilinear utility → no income effects)

The FOC directly defines the consumer’s inverse demand function,

which can be interpreted as the consumer’s Marginal willingness
to pay (MWTP) for an additional unit of good X :
p(x) = v 0 (x)

10 / 111
 Marshallian Demand

Price

Demand D(p)

Quantity
x=D(p)
11 / 111
 Marginal Willingness to Pay

Price

p(x)=v'(x)

MWTP = v'(x)

Quantity
x
12 / 111
Consumer Surplus: Model

Suppose that the price of good X is p0

At this price, the consumer’s demand x0 of good X is such that

v 0 (x0 ) = p0

We would like to evaluate the consumer’s net benefit (in monetary

terms) from consuming a quantity x0 of good X at price p0

This net benefit is the difference between

– the consumer’s willingness to pay (WTP) for that quantity
– the consumer’s expenditure (CE) on the good

13 / 111
Consumer Surplus: WTP

The consumer’s willingness to pay (WTP) for x0 units of good X

is the amount of income that the consumer would be willing to
sacrifice to enjoy x0 units of good X instead of zero

It can be defined as:

U(x0 , m − WTP) = U(0, m)

⇔ v (x0 ) + m − WTP = v (0) + m
Rx
⇔ WTP = v (x0 ) − v (0) = 0 0 v 0 (x)dx

Graphically, the consumer’s WTP is the area below the demand

curve and to the left of x0

14 / 111
 Willingness to Pay

Price

Willingness
to pay for
quantity x0

MWTP = v'(x)

Quantity
x0
15 / 111
Consumer Surplus

The consumer’s expenditure (CE) when buying a quantity x0 of

good X at price p0 is:
CE = p0 x0

Hence the consumer’s net surplus from consuming a quantity x0

at price p0 is:
Z x0
CS = WTP − CE = v (x0 ) − p0 x0 = v 0 (x)dx − p0 x0
0

Graphically, this is the area below the inverse demand curve and
above the price p0

16 / 111
 Expenditure

Price

Expenditure

p0

MWTP = v'(x)

Quantity
x0
17 / 111
 Consumer Surplus

Price

Consumer
surplus

p0

MWTP = v'(x)

Quantity
x0
18 / 111
Changes in Consumer Welfare

How does a change in prices affect the consumer’s welfare?

Consider a decrease in the price of good X : p0 → p1 , with p1 < p0

The change in consumer surplus ∆CS can be measured as:

Z p0
∆CS = D(p)dp Details
p1

Graphically, ∆CS is measured by the area below the Marshallian

demand and between the old and new prices

19 / 111
 Change in Consumer Surplus

Price

Change in
Consumer
surplus

p0

p1

MWTP = v'(x)

Quantity
x0 x1
20 / 111
Equivalent and Compensating Variations
Problem: changes in CS are an exact measure of a consumer’s
welfare change only if utility is quasilinear (no income effects)

There are only two exact measures of changes in consumer welfare:

compensating variation (CV) and equivalent variation (EV)

Consider a decrease in the price of a good:

– CV: amount of money that must be taken away from
consumer after the change to restore her original utility level
– EV: amount of money that must be given to the consumer
before the change to leave her as well off as with the change

Both measures are evaluated using different reference points:

– CV: old utility, new prices
– EV: new utility, old prices

21 / 111
 Compensating Variation
Quantity of
good 2
(price=1)

e(p0,u0)
As well off as
before
CV
e(p1,u0)
A
B

C U1

U0

Quantity
of good 1
22 / 111
 Equivalent Variation
Quantity of
good 2
(price=1)

e(p0,u1) As well off as

after

EV
e(p1,u1)

A
B

U1

U0

Quantity
of good 1
23 / 111
Equivalent and Compensating Variations
CV: amount of money to subtract from the consumer’s income
after the price fall to make her just as well off as she was before

Using Shephard’s lemma:

Z p0
CV = e(p0 , u0 ) − e(p1 , u0 ) = h(p, u0 )dp
p1

e(p, u): expenditure function

h(p, u): compensated demand (Hicksian demand)

EV: amount of money to add to the consumer’s income before the

price fall to leave her just as well off as she would be after
Z p0
EV = e(p0 , u1 ) − e(p1 , u1 ) = h(p, u1 )dp
p1

24 / 111
Changes in Consumer Surplus as an Approximation
Problem: because utility is not observable, CV and EV are difficult
to measure empirically

Changes in CS are easier to measure because they require only the

knowledge of the demand curve

∆CS is an approximation of welfare change. For a normal good

and a price fall:
CV < ∆CS < EV

∆CS is an exact measure of welfare change only when the

consumer’s preferences can be represented by a quasilinear utility
function (because there are no income effects). In this case:

∆CS = EV = CV

25 / 111
Comparing the Three Measures of Welfare Change (Normal Good)

Price
Hicksian Demand

h(p,u0) h(p,u1)

A D
p0

CV

p1 B
C

x(p,m) Marshallian
Demand

Quantity
x0 x1

26 / 111
Comparing the Three Measures of Welfare Change (Normal Good)

Price
Hicksian Demand

h(p,u0) h(p,u1)

A D
p0

EV

p1 B
C

x(p,m) Marshallian
Demand

Quantity
x0 x1

26 / 111
Comparing the Three Measures of Welfare Change (Normal Good)

Price
Hicksian Demand

h(p,u0) h(p,u1)
For a normal good:
CV < ∆CS < EV
A D
p0

∆CS

p1 B
C

x(p,m) Marshallian
Demand

Quantity
x0 x1

26 / 111
Equivalent and Compensating Variations

Which measure to use?

– CV assumes that agents are entitled to their current level of
utility → Appropriate to arrange for some compensation
scheme at new price.
– EV assumes that agents are entitled to some alternative level
of utility → More suitable for comparisons among a variety of
policies.

Changes in CS are a good approximation of welfare changes if

income effects are negligible or if budget share of the considered
good is small

27 / 111
Measuring Welfare Changes: Numerical Example
Application: what would be the impact of reducing the housing
rent on a tenant’s welfare?

Suppose that the tenant’s utility depends on the size of the rented
flat (x: number of square meters) and on the quantity y of a
composite good whose price is normalized to 1:

U(x, y ) = ln x + ln y

(N.B. this utility function is not quasilinear)

The tenant’s monthly income m is 2,000 euros and the rental price
per square meter is denoted p.

The tenant’s budget constraint is:

p·x +y =m

28 / 111
Measuring Welfare Changes: Numerical Example
Consider the following policy change: a 50% rent reduction, from
p0 =20 euros/m2 to p1 =10 euros/m2

How would this rent reduction affect the tenant’s welfare?

Alternative welfare measures:

– CV: how many euros would the tenant be willing to pay to
enjoy the lower rent? (WTP)
– EV: how many euros would the tenant require to accept to
forego the rent reduction? (WTA)
– Change in consumer surplus (∆CS)

Numerical application Details

– CV = 586 euros
– EV = 828 euros
– ∆CS = 693 euros
29 / 111
Cohen et al. (2016)
Cohen, Hahn, Hall, Levitt and Metcalfe (2016): “Using Big Data
to Estimate Consumer Surplus: the Case of Uber”, NBER working
paper No. 22627

Starting point: obtaining convincing estimates of consumer

surplus is empirically challenging
– requires identification of the entire demand curve
– problem: we only observe equilibrium points between
supply/demand

Most of the existing literature exploits exogenous shifts in the

supply curve to generate local estimates of demand elasticities

Main limitation: local elasticities are not sufficient for estimating

consumer surplus without strong functional form restrictions
→ need to know the entire demand curve to measure surplus

30 / 111
Cohen et al. (2016)
Cohen et al. (2016) exploit the richness of data generated by Uber
to estimate consumer surplus under less restrictive assumptions

Key feature of Uber: uses real-time pricing (“surge” pricing) to

balance local short-term supply and demand

A consumer wishing to take a particular trip can face prices

ranging from the base price (“no surge” or “1.0x” price) to five
times or higher price

Data: over 50 million UberX sessions from 4 large U.S. cities

(Chicago, Los Angeles, New York and San Francisco)

Empirical strategy:
– Rely on Uber’s “surge” pricing algorithm to estimate demand
elasticities at several points along the demand curve
– Use elasticity estimates to measure CS from Uber
31 / 111
ure 1: Uber mobile application request screens
Uber Mobile Application Request Screens

Panel A Panel B

e: These figures illustrate what the Uber app looks like43 when a rider is requesting transportation. Panel
Source: Cohen et al. (2016), Figure 1.
cts the period preceding a request when users are asked to choose a product and set a pick-up location. Pane
32 / 111
Step 1: Estimation of Demand Elasticities

If surge prices were randomly assigned to sessions, identification of

price elasticities of demand D would be straightforward: regress
log of number of purchases Q on log of price p
∆Q/Q
D = ∆p/p : % change in purchases for a 1% increase in price

Problem: surge varies systematically with observed factors (e.g.,

week day) and unobserved ones (e.g., availability of taxi cabs, rain)

The papers’ identification strategy is to use the discontinuous

pricing induced by Uber’s business rules regarding surges
– surge pricing algorithm generates a continuous measure
– but actual prices take discrete value (e.g., 1.0x, 1.2x, 1.3x...)

Use RD methods to estimate price elasticities around “jump” points

33 / 111
 Example of UberX Purchase Rate Changes at Price Discontinuity
Figure 4: Example of purchase rate changes at price discontinuity
(1.3x vs. 1.2x Surge)

Note: This figure illustrates how purchase rates vary as a function of the surge generator over the range
1.15x to 1.35x. The vertical line when the surge generator equals 1.25 identifies the point at which the
surge price changes from 1.2x to 1.3x.

Source: Cohen et al. (2016), Figure 4.

34 / 111
 UberX Request Rate Drops at Pricing Discontinuities
Figure 5: Request rate drops at pricing discontinuities

Note: This figure illustrates how purchase rates vary as a function of the surge generator when the surge
generator is less than 2.4x. Red bars identify all observations within .01 units to the left of a price
discontinuity. Yellow bars identify all observations within .01 units to the right of a price discontinuity.
All observations not within these windows are depicted in gray.

Source: Cohen et al. (2016), Figure 5.

35 / 111
 Table 3: Estimated Price Elasticities at various Points along the Demand Curve
Estimated Price Elasticities at various Points along the Demand Curve

Note: Price elasticity = (% ∆ in number of purchases)/(% ∆ in prices) = ([θ/Purchase

rate]/% ∆ in prices), where θ is the drop in purchase rates at the price discontinuity.
Source: Cohen et al. (2016), Table 3.
36 / 111
Step 2: Estimation of Consumer Surplus
Consumer surplus (CS) is estimated using the the price elasticity
estimates at various points along the demand curve

Baseline approach: compute CS separately each group of sessions:

– CS of sessions that made a purchase at a 1.0x surge
– CS of sessions that made a purchase at 1.2x surge
...
– maximum willingness to pay is assumed to be at 4.9x surge
(conservative)

CS for each group of session is calculated as difference between

willingness to pay (based on demand curve generated from
underlying elasticities) and actual expenditure

Total CS is the sum of CS across all surge price groups

37 / 111
 Figure 6: Visual representation of demand curve for transactions at 1.0x
Visual Representation of Demand Curve for Transactions at 1.0x

Note: This figure presents a piecewise linear demand curve with jumps at each price discontinuity. The
curve is generated from the underlying elasticities estimated for each price discontinuity and for
consumers facing transactions at 1.0x.

Source: Cohen et al. (2016), Figure 6.

38 / 111
Step 2: Estimation of Consumer Surplus
Example: CS of sessions that made a purchase at 1.0x surge
– Q1.0 : number of sessions that led to a purchase at 1.0x surge
– CS up to price 1.2x for those who paid 1.0x:
CS(1.0 → 1.2) ≈ Q̂1.2 × (1.2P − P) = Q̂1.2 × 0.2P
Q̂1.2 : estimated number of sessions with purchase at 1.0x
surge that would have continued to buy at 1.2x surge
P: average fare paid by sessions at 1.0x surge
– Q̂1.2 is estimated from price elasticity at 1.2x surge, using:
 
Q̂1.2 − Q1.0 D 1.2P − P
≈ ε̂1.2
Q1.0 P
ε̂D
1.2 : estimated price elasticity of demand at 1.2x surge
– Analogous calculations for CS(1.2 → 1.3), etc.

Repeat above steps for sessions that made purchase at 1.2x, etc.
39 / 111
Step 2: Estimation of Consumer Surplus
Underlying identifying assumption: price changes are not correlated
with other determinants of demand
Problem: users who face high surge might not be comparable to
no-surge users (e.g., more/less elastic?)
Sensitivity checks:
– Control for changes in the composition of sessions over the
surge distribution: reweight the data so that observable
characteristics (e.g., location, time, intensity of Uber use) at a
particular surge level match sessions where surplus is measured
– The degree of surge could be correlated with unobservable
determinants of demand (e.g., availability of taxi cabs, rain):
the authors exploit the fact that Uber limits the speed at
which the surge price can increase (i.e., some sessions see a
price of 1.5x even if the surge generator says 2x or 3x).
Estimates of CS are very similar across the different specifications
40 / 111
 Figure 8: Elasticity estimates with and without matching on observables
Elasticity Estimates with and without Matching on Observables

Note: This figure presents two demand curves generated via different approaches. The blue demand curve
(also presented in Figures 6 and 7) is linear with jumps at each price discontinuity while the green
demand curve is based on elasticity estimates derived from data that were re-weighted to match the
distribution of observables found at price 1.0x.

Source: Cohen et al. (2016), Figure 8.

41 / 111
 Purchase Rates as a Function of Underlying Market Conditions when
Figure 10: Purchase rates as a function of underlying market conditions when the surge
the Surge Price is Artificially Restricted to 1.5x
price is artificially restricted to 1.5x

Note: This figure presents the purchase rate by surge generator where the actual price observed by the
rider is constrained to be 1.5x.

Source: Cohen et al. (2016), Figure 10.

43 / 111
Step 2: Estimation of Consumer Surplus
Estimated consumer surplus from Uber:
– $2.88 billion per year in the four US cities covered by study
– $6.76 billion per year if extrapolated to all UberX trips in US

For each $1 spent on an UberX ride at 1.0x, estimate that the

consumer receives $1.57 in extra surplus

Argue that these estimates are large relative to the likely losses
experienced by taxi drivers as a consequence of Uber

Main limitations:
– Short-run elasticities: if Uber was to disappear permanently, a
long-run elasticity would be more appropriate
– Misses consumer surplus associated with other ride-sharing
products and consumer benefit/harm from responses of the
taxi industry to Uber’s entry
– Externalities not factored in (e.g., increased congestion)
45 / 111
Producer Surplus
Measuring the impact of changes in prices on the producer’s
welfare is more straightforward than for the consumer

Consider a firm selling x units of output at price p. Cost function

c(·) assumed to be increasing and convex (c 0 > 0, c 00 > 0)

FOC from profit maximization:

p = c 0 (x) Details

Implicitly defines the producer’s supply curve:

x S = S(p)

The producer’s inverse supply curve (marginal cost curve) can be

interpreted as the minimum price at which a producer is willing to
sell every extra unit of good:

p(x) = c 0 (x)
46 / 111
 Supply Curve

Price

Supply = c'(x)

Quantity
x=S(p)
47 / 111
 Inverse Supply Curve

Price

Supply = c'(x)

p=c'(x)

Quantity
x
48 / 111
Producer Surplus
The producer’s willingness to sell (WTS): minimum amount of
money required to produce a given quantity x0 of good. It is equal
to the sum of its marginal costs of production (total variable costs):
Z x0
WTS = c 0 (x)dx = TVC
0

What the producer actually receives when selling a quantity x0 of

good at price p0 is given by its total revenues TR:

TR = p0 x0

The producer surplus (PS) is the difference between total

revenues and total variable costs:
Z x0
PS = TR − TVC = p0 x0 − c 0 (x)dx
0

Graphically, it is the area above the inverse supply curve and

below the price p0
49 / 111
 Wilingness to Sell = Sum of Marginal Costs

Price

Supply = c'(x)
Sum of
marginal
costs

p0

Quantity
x0
50 / 111
 Total Revenues

Price

Supply = c'(x)
Total
revenues

p0

Quantity
x0
51 / 111
 Producer Surplus = Total Revenues − Total Variable Costs

Price

Supply = c'(x)
Producer
surplus

p0

Quantity
x0
52 / 111
Producer Surplus
Producer surplus is closely related to the profits Π of the firm:
Z x0
PS = p0 x0 − c 0 (x)dx
0
⇔ PS = p0 x0 − c(x0 ) + c(0)
⇔ PS = Π + FC

where FC = c(0) are the firm’s fixed costs of production

Thus: Producer surplus = Profits + Fixed costs of production

Firms produce output only if the producer surplus is positive. If

not, they shut down.

In a long-run equilibrium, producer surplus measures the economic

rents that are captured by the owners of scarce industry-specific
inputs
53 / 111
Measuring Changes in Producer Welfare

How does a change in prices affect producer surplus?

Consider a price increase: p0 → p1 , with p1 > p0

The change in producer surplus ∆PS can be defined as:

Z p1
∆PS = ∆Π = S(p)dp Details
p0

Graphically, ∆PS is measured by the area above the supply

curve S(p) and between the old and new prices

54 / 111
 Change in Producer Surplus

Price

Change in
Producer
surplus Supply = c'(x)

p1

p0

Quantity
x0 x1
55 / 111
Competitive Equilibrium and
Social Efficiency
Competitive Equilibrium and Efficiency
The social surplus framework can be used to illustrate the First
Fundamental Theorem of Welfare: the competitive equilibrium
maximizes social efficiency

Intuition:
– social efficiency is created whenever a trade occurs such that
has benefits that exceed its costs
– this occurs for every transaction to the left of the point where
supply equals demand
– for each of these transaction, the benefits (willingness to pay,
i.e., demand) exceed the costs (marginal cost, i.e., supply)

Formal analysis using a partial equilibrium framework (assumes

no feedback effects to other markets when changes occur in the
market under study)

57 / 111
Competitive Equilibrium and Efficiency

Setup:
– Two goods: good X (price p) and a composite commodity Y
(numeraire)
– I consumers with quasilinear preferences over X and Y . Each
consumer is initially endowed with a certain amount of the
numeraire.
– J firms. Each firm can produce q units of good X by using
cj (q) units of the numeraire (cj0 > 0 and cj00 > 0)

A Pareto-efficient allocation is an allocation that maximizes the

total surplus (Marshallian aggregate surplus)

The competitive equilibrium achieves this Details

58 / 111
 Aggregate Demand Curve

Price

Aggregate
p demand
curve

Individual
demand
curves

v'(x) v'(x)
2 D(p)
1

Quantities
x1(p) x2(p) x(p)=Σxj(p)
59 / 111
 Aggregate Supply Curve

Price

c'(q) c'(q) Aggregate S(p)

1 2
supply
Individual curve
supply
curves

Quantities
q1(p) q2(p) q(p)=Σqj(p)
59 / 111
 Competitive equilibrium

Price

Aggregate
supply

p*

Aggregate
demand

Quantities
Q*
59 / 111
Marshallian Aggregate Surplus for a Given Quantity of the Traded Good

Price

Marshallian
aggregate Aggregate
surplus supply

Aggregate
demand

Quantities
Q
60 / 111
Competitive Equilibrium Maximizes Marshallian Aggregate Surplus

Price

Marshallian
aggregate Aggregate
surplus supply

p*

Aggregate
demand

Quantities
Q*
61 / 111
 Surplus of Consumer 1

Price

Surplus of
consumer 1

Aggregate
p demand
curve

Individual
demand
curves

v'(x) v'(x)
2 D(p)
1

Quantities
x1(p) x2(p) x(p)=Σxj(p)
61 / 111
 Surplus of Consumer 2

Price

Surplus of
consumer 2

Aggregate
p demand
curve

Individual
demand
curves

v'(x) v'(x)
2 D(p)
1

Quantities
x1(p) x2(p) x(p)=Σxj(p)
61 / 111
 Total Consumer Surplus

Price

Total
consumer
surplus

Aggregate
p demand
curve

Individual
demand
curves

v'(x) v'(x)
2 D(p)
1

Quantities
x1(p) x2(p) x(p)=Σxj(p)
61 / 111
 Surplus of Producer 1

Price

c'(q) c'(q) Aggregate S(p)

1 2
supply
Individual curve
supply
curves

Surplus of
producer 1

Quantities
q1(p) q2(p) q(p)=Σqj(p)
61 / 111
 Surplus of Producer 2

Price

c'(q) c'(q) Aggregate S(p)

1 2
supply
Individual curve
supply
curves

Surplus of
producer 2
p

Quantities
q1(p) q2(p) q(p)=Σqj(p)
61 / 111
 Total Producer Surplus

Price

c'(q) c'(q) Aggregate S(p)

1 2
supply
Individual curve
supply
curves

Total
producer
surplus
p

Quantities
q1(p) q2(p) q(p)=Σqj(p)
61 / 111
Measuring Inefficiency:
Deadweight Loss
Measuring Inefficiency: Deadweight Loss

Market failures can cause the market economy to deliver an

outcome that does not maximize efficiency (e.g., imperfect
competition, public goods, externalities, asymmetric information)

Inefficient outcomes can also result from government

intervention (e.g., price ceiling/floor, taxes/subsidies)

The loss in producer and consumer surplus due to an inefficient

level of production can be measured using the concept of
deadweight loss

Two examples:
– Monopoly pricing
– Rent control

63 / 111
Deadweight Loss of Monopoly Pricing
Consider a monopolist firm producing a good X

Contrary to a competitive firm, the monopolist firm has market

power: it choose both the quantity of output to supply and the
price of that output, the relationship between price and quantity
being given by the demand curve that it faces

The monopolist’s profit maximization problem can be written as:

max p · q − c(q)
p,q

s.t. q = D(p)

where D(·) is the aggregate demand function for good X

Let p(q) denote the inverse demand function

64 / 111
Deadweight Loss of Monopoly Pricing

Rewrite the monopolist’s profit maximization problem as:

max p(q)q − c(q)

q

FOC:
p(q) + p 0 (q)q = c 0 (q)
| {z } | {z }
Marginal Revenue Marginal Cost

The marginal revenue takes into account the fact that producing
and selling one more unit will decrease the price at which the firm
can sell all of its output

since p 0 (q) < 0, the marginal revenue curve is below the demand
curve p(q)

65 / 111
 Surplus under Competitive Equilibrium

Price

Marginal Cost
Consumer = c'(q)
Surplus

pM

p0 E0

Producer Demand
Surplus Marginal Revenue = p(q)
= p(q) + p'(q)q
Quantity
QM Q0
67 / 111
 Surplus under Monopoly Pricing

Price

Marginal Cost
Consumer = c'(q)
Surplus

pM

p0 E0

Producer Demand
Surplus Marginal Revenue = p(q)
= p(q) + p'(q)q
Quantity
QM Q0
67 / 111
 Deadweight Loss of Monopoly Pricing

Price

Marginal Cost
= c'(q)

Deadweight
Loss
pM

p0 E0

Demand
Marginal Revenue = p(q)
= p(q) + p'(q)q
Quantity
QM Q0
67 / 111
Government Intervention can Induce Inefficiencies

The potential gain from government intervention to correct market

failures (e.g., antitrust laws) can be measured by the deadweight
loss avoided

However, government intervention may itself induce economic

inefficiency welfare losses

The government has to trade off the benefits of its intervention

against the distortions that it might create

Example: deadweight loss from rent control

68 / 111
Rental Housing Market: Competitive Equilibrium of Housing Market

Price

Consumer
Surplus

Supply of Rental
Housing

p0 E0

Demand for Rental

Housing
Producer
Surplus

Quantities
Q0
69 / 111
 Rental Housing Market: Deadweight Loss from Rent Control

Price

Consumer
Surplus

Deadweight
Retained
consumer
Loss Supply of Rental
surplus Housing
Lost
consumer
surplus
Lost producer
E0
Transfer from producers
to consumers surplus
Price
cap ER
Retained Demand for Rental
producer
surplus
Housing
Producer
Surplus Shortage

Quantities
QR QD
69 / 111
Glaeser and Luttmer (2003)

Glaeser and Luttmer (2003) argue that the classical analysis of the
welfare loss from rent control underestimates true deadweight loss:
– assumes that the rationing under rent control ensures that
apartments go to the consumers who value them the most
→ deadweight loss driven purely by undersupply
– but price controls will lead to a misallocation of apartments
among consumers because of rationing → extra loss in
consumer surplus

For small price caps, allocation inefficiency dwarfs undersupply

inefficiency

Example: suppose that apartments are randomly allocated across

the consumers who want them at the rent-controlled price

70 / 111
 Rental Housing Market: Misallocation

Price

Consumer
Surplus
Lost consumer
surplus due to Deadweight
missalocation
Average
Loss Supply of Rental
consumer Housing
Lost
valuation
Retained consumer consumer
of rental surplus due to
unit surplus
undersupply
Lost producer surplus E0
Transfer from producers due to under-
to consumers supply
Price
cap ER
Retained Demand for Rental
producer
surplus
Housing
Producer
Surplus Shortage

Quantities
QR QD
71 / 111
Glaeser and Luttmer (2003)

Glaeser and Luttmer (2003) aim to quantify the welfare losses from
misallocation by comparing consumption patterns in rent-controlled
city (NYC) vs. free-market cities across demographic groups

Predict apartment size using number in family, income, education,

age, etc., in 103 cities without rent control

Test if actual apartment allocations in NYC match predictions

Strong assumption: preferences are stable across cities

Placebo tests using Chicago and Hartford (no rent control)

72 / 111
 Average Overlap in Housing Consumption between Population Groups

Probability that rooms for household from group A

⬎ rooms for household from group B
New York City renters U.S. free-market renters
Observations Overlap Observations Overlap
Group A: High school dropoutb 3,174 0.470 4,554 0.316
Group B: College or more 2,450 (0.008) 5,123 (0.005)
Group A: Households without children 6,794 0.229 16,027 0.200
Group B: Households with children 3,206 (0.005) 4,573 (0.004)
Group A: Age ⱕ 35a 2,859 0.279 10,456 0.343
Group B: Age ⬎ 35 and ⱕ 60 4,280 (0.006) 5,381 (0.005)
Group A: 1 person households 3,758 0.150 10,261 0.150
Group B: 3⫹ person households 3,621 (0.005) 4,483 (0.004)
Group A: Per capita income in bottom 1⁄3c 3,338 0.457 6,798 0.351
Group B: Per capita income in top 1⁄3 3,300 (0.007) 6,795 (0.005)

Source: Glaeser and Luttmer (2003), Table 2.

73 / 111
of apartments in buildings with five units or more living in PUMAs with at least 10 percent of the population living in
buildings with five units or more.
a
Age refers to the average of the household head and his/her spouse.
b
Maximum of educational attainment of household head and his/her spouse.
c
Top and bottom 1⁄3 of the per capita income distribution are determined relative to indicated sample.

Actual and Efficient Allocation of Apartments in NYC

TABLE 3—ACTUAL AND EFFICIENT ALLOCATION

Allocation of New York City households across apartments

(fraction of households)
Efficient apartment size (number of rooms):
Actual apartment size
(number of rooms): 1 2 3 4 5 Marginal
1 0.0688 0.0206 0.0000 0.0000 0.0000 0.0894
2 0.0204 0.0766 0.0375 0.0000 0.0000 0.1345
3 0.0002 0.0373 0.2667 0.0465 0.0001 0.3508
4 0.0000 0.0000 0.0466 0.2126 0.0243 0.2835
5 0.0000 0.0000 0.0000 0.0243 0.1175 0.1418
Marginal 0.0894 0.1345 0.3508 0.2835 0.1418 1.0000

Note: The table shows the joint distribution of the actual and efficient allocation of households to apartments in the baseline
treatment group (10,000 households in New York City).
Source: Glaeser and Luttmer (2003), Table 3.

ments but are living in studios. As the matrix measure, M. The first row shows the gross
→ 26% of NYC renters live in apartments
indicates, we find few cases where misalloca-
that are the “wrong” size
misallocation, which does not correct for mis-
(sum
tion is of
off off-diagonal
by more than oneelements)
room. Naturally, allocation due to sampling error. According to
since we assume that allocation is perfectly this measure, the overall percentage of New
efficient within subgroups, we should not be York renters that are living in apartments that
surprised to find so few cases of major misal- are the wrong size is 25.8 percent, which is the
location. The basic fact driving this table is that, sum of the off-diagonal elements in Table
as shown in Table 2, the connection between 3. The second row uses the bootstrap procedure 74 / 111
The Efficiency Cost of Taxation
Efficiency Cost of Taxation
The Government raises taxes for two main reasons:
– to raise revenue to finance government expenditure
– to redistribute income

Problem: because lump-sum taxes are not feasible, the government

has to rely on distortionary taxes which generate inefficiencies.
To generate $1 of revenue, the welfare of those taxed is reduced by
more than $1 because the tax distorts incentives and behavior

Intuition: suppose the government imposes a large tax on

imported cars, leading consumers to completely switch to cars
produced domestically. This tax raises no revenue but the
consumers are clearly made worse off and would be willing to pay
some amount to get the government to abolish this tax

Core theory of public finance: how to implement policies that

minimize these efficiency costs?
76 / 111
Efficiency Cost of Taxation: Model
Partial equilibrium analysis with two goods X and Y (numeraire)
– only one relative price → partial and general eq. are the same
– can be viewed as an approximation of a multi-good model if
the market being taxed in small, and there are no close
substitutes/complements in the utility function

Suppose the government introduces an excise tax τ on a

commodity X (e.g., a tax on fuel)
– Excise tax: levied on a quantity (e.g., gallon, ton)
Tax = quantity × tax per unit ($)
– Ad-valorem tax: fraction of prices (e.g., VAT)
Tax = quantity × price × tax rate (%)

The price of the numeraire good Y is 1. The pre-tax price of

commodity X is p. The after-tax price is q = p + τ
77 / 111
Efficiency Cost of Taxation: Demand
Consumer has income m, quasilinear utility v (x) + y , and solves:

max v (x) + y
x,y

s.t. (p + τ )x + y = m

First-order condition yields the consumer’s demand function:

p + τ = v 0 (x) ⇒ x D = D(p + τ )

dD/D qD 0 (q)
Let εD = = < 0 denote the price elasticity of
dq/q D(q)
demand (consumer faces q = p + τ )

– % change in demand when (post-tax) price changes by 1%

– useful because elasticities are independent of scaling

78 / 111
Efficiency Cost of Taxation: Supply
The producer uses c(x) units of the numeraire good to produce x
units of good X

Marginal cost is increasing and convex (c 0 > 0 and c 00 ≥ 0)

The producer’s profit at pre-tax price p and level of supply x is:

π = p · x − c(x)

Under perfect competition, the producer’s supply function is

implicitly defined by:

p = c 0 (x) ⇒ x S = S(p)

dS/S pS 0 (p)
Let εS = = > 0 denote the price elasticity of
dp/p S(p)
supply (% change in supply when pre-tax price changes by 1%)

79 / 111
Efficiency Cost of Taxation: Equilibrium

The equilibrium in this market is the point where supply equals

demand, i.e. the pre-tax price (producer price) p(τ ) such that:

Q = D (p + τ ) = S (p)

Defines an equation p(τ ) → equilibrium pre-tax price depends on τ

The tax causes a reduction in the equilibrium quantity of

good X below the competitive quantity

80 / 111
 Impact of Excise Tax on Equilibrium

Pre-tax
price (p)

Supply S(p)
p(τ)+τ

p0 E0

p(τ) Eτ
τ Demand D(p)

D(p+τ)

Quantities
Qt Q0
81 / 111
Efficiency Cost of Taxation

The reduction in quantity causes the consumer and producer

surplus to fall

Much of this reduced social surplus is transferred to the

government in the form of higher tax revenue, but some of it
disappears because some surplus-producing trades are not made

The deadweight loss (“excess burden”) measures the amount of

social surplus which is lost and not recaptured in the form of tax
revenues

Graphically, the DWL is measured by Harberger’s Triangle

(Hines, 1999)

82 / 111
 Deadweight Loss of Taxation

Pre-tax
price (p)

Consumer
Surplus

Tax
Revenue
Supply S(p)
p(τ)+τ

p0 E0

p(τ) Eτ
τ Demand D(p)

Producer D(p+τ)
Surplus

Quantities
Qτ Q0
83 / 111
 Deadweight Loss of Taxation

Pre-tax
price (p)

DWL of
tax Supply S(p)
p(τ)+τ

p0 E0

p(τ) Eτ
τ Demand D(p)

D(p+τ)

Quantities
Qτ Q0
83 / 111
Efficiency Cost of Taxation: Harberger Formula
Harberger formula to compute deadweight loss of introducing a
small tax dτ > 0:
1
DWL = − dQ · dτ |dτ| |dQ|
2
   2
1 εS · εD dτ
DWL = − pQ Details
2 εS − εD p

Important implications for tax policy:

1. DWL proportional to the size of the market (Q)
2. DWL rises with elasticities of supply (εS ) and demand (εD )
⇒ should tax more inelastic goods (e.g., medical drugs, food)
3. DWL rises with square of the tax rate dτ (i.e. marginal DWL
rises with tax rate) ⇒ spread taxes across all goods so as to
keep tax rates relatively low on each good (broad tax base)
84 / 111
 DWL Rises with the Elasticities of Supply and Demand

(a) Inelastic Supply and Demand (b) Elastic Supply and Demand

Pre-tax Pre-tax
price (p) Price (p)

S(p)

p(τ)+τ
p(τ)+τ
E0
S(p)
p0 E0 p0
D(p)
p(τ) Eτ p(τ) τ
Eτ

D(p+τ)
τ D(p)

D(p+τ)
Q Q
Qτ Q0 Qτ Q0
85 / 111
 Deadweight Loss Increases with Square of Tax

Pre-tax
price (p)

DWL of
tax Supply S(p)
p(τ)+τ

p0 E0

p(τ) Eτ
τ
Demand D(p)

D(p+τ)

Quantities
Qτ Q0
86 / 111
 Deadweight Loss Increases with Square of Tax

Pre-tax
price (p)

Change
p(2τ)+2τ in DWL
Supply S(p)

p0 E0

Eτ

p(2τ) E 2τ Demand D(p)

τ D(p+τ)

D(p+2τ)
Quantities
Q 2τ Qτ Q0
86 / 111
Efficiency Cost of Taxation: Implementable Formula
Alternative implementable formula for excess burden: in
terms of total change in equilibrium quantity caused by
introduction of small tax dτ > 0:
 2
1 dτ
DWL = ηQ (pQ) Details
2 p

dQ/Q
where ηQ = −
dτ /p0

ηQ : effect of a 1% increase in initial price (via a tax change) on

equilibrium quantity, taking into account endogenous price change

This is the coefficient β in a regression of the form:

τ
log Q = α + β +
p0
The coefficient β can be identified from exogenous variation in τ
87 / 111
Efficiency Cost of Taxation
Suppose that we start with an initial tax rate of τ . Then:
 2
1 τ
DWL = ηQ (pQ)
2 p
The marginal DWL (to a first-order approximation) is:
dDWL τ
≈ ηQ Q
dτ p
The marginal change in government revenue R = Q · τ is:
dR d(Q · τ ) dQ τ
= =Q +τ = Q + QηQ
dτ dτ dτ p
Hence MDWL of tax per dollar of net government revenue is:

dDWL/dτ ηQ pτ
≈
dR/dτ 1 − ηQ pτ
88 / 111
Marion and Muehlegger (2008)

Marion and Muehlegger (2008) study the deadweight loss from

taxing diesel fuels, focusing on evasion

Diesel fuel used for business purposes (e.g. trucking) is taxed, but
is not for residential purposes (e.g. heating homes)

→ Substantial opportunity to evade tax

1993 Reform: government added red dye to residential diesel fuel

→ easy to monitor cheating by opening gas tank of a truck

First, document effect of dye reform on evasion

89 / 111
 U.S. Sales of Diesel Fuels
Thousands of gallons/day
200000

Diesel dye program implemented

180000

160000

140000

120000

100000
All No 2 Distillate
No 2 Fuel oil (untaxed)
80000
No 2 Diesel (taxed)

60000

40000

20000

0
1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003

Source: Marion and Muehlegger (2008), Figure 2

90 / 111
Marion and Muehlegger (2008)
Then use reform to assess the DWL of evasion and taxation

Harder to evade → elasticity of behavior with respect to tax much

lower after the reform

Estimate price and tax elasticities of diesel fuel before and after
reform, using cross-state variation in tax rates and price variation
from world market (IV):
 
τit
ln Qit = β0 + β1 ln(pit ) + β2 ln 1 + + Xit γ + f (t) + αi + it
pit

Qit : quantity of diesel fuel sold in state i in month t

pit : pre-tax price of diesel (same for business/residential)
τit : diesel tax (per gallon) – only for business
|β1 |: price elasticity; |β2 |: tax elasticity.

If no tax evasion then β1 = β2 (∆tax and ∆price have same effect)

91 / 111
 Price and Tax Elasticity Estimates

Log Diesel Sales

OLS IV
(1) (2)
tax Log(1 ⫹ tax rate) ⫺1.409 ⫺1.333
elasticity ȕ2 (.506)*** (.511)**
Postdye#log(1 ⫹ tax rate) .579 .597
(.930) (.965)
price Log(price) ⫺.521 ⫺.486
elasticity ȕ1 (.148)*** (.144)***
Postdye#log(price) .241 .258
(.335) (.359)
Observations 10,236 10,236
R2 .96 .96
Source: Marion and Muehlegger (2008), Table 6

|β2 | > |β1 |: tax elasticity much higher than price elasticity before
reform → evidence of tax evasion
Tax elasticity falls considerably after reform: from −1.409 to
−0.830 (caveat: difference is not statistically significant)
92 / 111
Marion and Muehlegger (2008)
Use OLS estimates of β2 (tax elasticity) to compute DWL of tax
Implied value of ηQ (impact of a 1% increase in price via a tax
change on equilibrium quantity):
 
p
ηQ = −β2
p+τ
  
dQ/Q
(Derivation: β2 = dln Q 
= dτ /p
1+ τ
p
= −ηQ p+τ
p
)
d ln 1+ τp

– pre-dye: β̂2 = −1.409; τ = 30.54 cts; p = 75.49 cts ⇒ ηQ =1.003

– post-dye: β̂2 = −0.830; τ = 44.47 cts; p = 81.17 cts ⇒ ηQ =0.536
Implied marginal DWL of diesel tax (using formula )
– MDWL=68 cents per dollar of net tax revenue before reform
– MDWL=42 cents per dollar after reform → 40% reduction!
Lesson: DWL depends not just on preferences/costs but also on
enforcement technology
93 / 111
Tax Incidence
Tax Incidence

Deadweight loss is about how taxes affect the size of the pie

Tax incidence is about how taxes affect the distribution of the pie:
impact on equilibrium prices and distribution of economic welfare

What happens to market prices when a tax is introduced/changed?

– what happens when we impose a 1 euro per pack tax on
cigarettes? Increase the VAT? Provide a subsidy for food (e.g.,
food stamps)? Raise Social Security contributions (SSCs)?
– these price interventions affect prices, and hence have
distributional effects on smokers, profits of producers,
shareholders, workers...

Tax incidence is an important first step when evaluating how

policies affect social welfare

95 / 111
Tax Incidence: Terminology

Legal liability: what the law says about who should pay the tax
(also called statutory or formal incidence)
– e.g., households should pay the income tax
– e.g., employees should pay employees’ payroll tax

Remittance responsibility: who is responsible for remitting the

amount of tax to the tax authorities
– e.g., value-added tax (VAT) is remitted by producers
– e.g., employees’ payroll tax is remitted by employers
– e.g., capital income tax can be remitted by banks

Economic/effective incidence: describes who actually bears the

tax burden, i.e., who is worse off (in terms of utility) as a result of
the tax

96 / 111
Economic Incidence
Who pays taxes? The naive answer is that it is obvious
– e.g., consumers pay the VAT
– e.g., employers pay employers’ payroll tax
– e.g., firms pay the corporate tax

Who actually bears the tax burden? Taxes can be shifted

because they affect prices and factor returns

Example: increase in VAT

– if after-tax prices increase by the amount of the tax
→ consumers pay the tax
– if after-tax prices stay constant and profits go down
→ capital owners pay the tax
– if after-tax prices stay constant and wages go down
→ wage earners pay the tax

97 / 111
Partial Equilibrium Incidence

Key reference: Kotlikoff and Summers (1987)

Two good economy: X and Y (numeraire)

Partial equilibrium model

Government levies an excise tax τ on good X

Let p denote the pre-tax price of X and q = p + τ denote the

tax-inclusive price of X (statutory incidence is on consumer)

(Note: if tax is levied on producers, consumers face pre-tax price p

while producers receive p − τ )

98 / 111
Partial Equilibrium Incidence
Demand for good X is D(q), where q = p + τ

Supply for good X is S(p)

Equilibrium condition: Q = S (p) = D (p + τ )

Defines equation for equilibrium pre-tax price p(τ )

Start from τ = 0 and S(p) = D(p). We want to characterize dp dτ :

effect of a tax increase on pre-tax price, which determines who
bears the effective burden of the tax
dp
Fully differentiating eq. condition w.r.t. τ and solving for dτ gives:

dp εD
= Details
dτ εS − εD

εD : price-elasticity of demand; εS : price-elasticity of supply

99 / 111
The Three Rules of Tax Incidence

Incidence formula:

dp εD
=
dτ ε S − εD

Key implications:
1. The legal incidence of a tax does not describe who really bears
the tax
2. The side of the market on which a tax is imposed is irrelevant
to the distribution of the tax burdens
3. The more inelastic factor bears more of the tax

These are robust conclusions that hold is more complicated models

100 / 111
Rule 1: Legal Incidence is not Economic Incidence
Example: Tax Levied on Consumers

Pre-tax
price (p)

Consumer
burden
pτ + τ Supply S(p)

p0
E0
pτ
Producer
burden

τ
Demand D(p)
D(p+τ)
Quantities
Qτ Q0

Although the statutory burden of the tax is on consumers, the real

burden of the tax is shared by producers and consumers
101 / 111
Rule 2: The Side of the Market is Irrelevant
Example: Tax Levied on Producers

Pre-tax
price (p) S(p−τ)

Consumer
burden
pτ τ Supply S(p)

p0
E0
pτ − τ
Producer
burden

Demand D(p)

Quantities
Qτ Q0

Same distribution of the tax burden as when tax is levied on consumers

102 / 111
Rule 3: More Inelastic Factor Bears More of the Tax

Incidence on consumers (i.e., on after-tax price q):

dq d(p + τ ) dp εS
= =1+ =
dτ dτ dτ εS − εD

Who bears the burden of the tax?

d(p+τ )
– εS  εD ⇒ dτ ≈ 1 (tax fully shifted on consumers)
d(p+τ )
– εD  εS ⇒ dτ ≈ 0 (tax fully shifted on producers)

The inelastic factor tends to bear more of the tax:

– e.g., producers with fixed quantity supplied (housing)
– e.g., consumers with no substitute untaxed commodity (gas)

Tax shifting occurs through pre-tax price changes

103 / 111
Rule 3: More Inelastic Factor Bears More of the Tax
Example: Perfectly Inelastic Demand

Pre-tax
price (p) S(p−τ)

pτ
τ Supply S(p)

p0=pτ−τ
E0
Consumer
burden

Demand D(p)

Quantities
Qτ = Q 0

A tax on producers of a inelastically demanded good is fully reflected in

increased prices, so consumers bear the full tax
104 / 111
Rule 3: More Inelastic Factor Bears More of the Tax
Example: Perfectly Elastic Demand

S(p−τ)

Supply S(p)

τ
Producer
burden

p0=pτ Demand D(p)

E0

p τ −τ

Quantities
Qτ Q0

A tax on producers of a perfectly elastically demanded good cannot be

passed along to consumers through an increase in price
105 / 111
Application: VAT Reforms in France
Carbonnier (2007): exploits two VAT reforms in France to estimate
the incidence of VAT

Two large reductions in VAT tax rates:

– Sep 1987: rate on car sales = 33.3% → 18.6%
– Sep 1999: rate on housing repair services = 20.6% → 5.5%

Data on prices:
– Insee Indice des prix à la consommation (IPC) monthly series
– COICOP commodity classification

Estimation strategy: Difference-in-differences

Findings: consumer share of the tax burden

– 57% for car sales
– 77% for housing repair services
106 / 111
 Car C.
Prices around September 1987 VAT Reform
Carbonnier / Journal of Public Economics 91 (2007) 1219–1229 1223

Source: Carbonnier (2007), Figure 1.

107 / 111
 Housing Repair
1224
Service Prices around September 1999 VAT Reform
C. Carbonnier / Journal of Public Economics 91 (2007) 1219–1229

Source: Carbonnier (2007), Figure 2.

108 / 111
 VAT Reforms: Estimates of Shifting Parameter
Consumer share estimates
OLS White method
(1) House repair (2) Car sales (3) House repair (4) Car sales
Number of observations 36 36 36 36
R2 99% 67% 99% 67%
VAT rate shifting during the 1st month (α1) 0.169⁎⁎⁎ 0.173⁎⁎⁎ 0.169⁎⁎⁎ 0.173⁎⁎⁎
(0.009) (0.043) (0.002) (0.008)
VAT rate shifting during the 2nd month (α2) 0.472⁎⁎⁎ 0.272⁎⁎⁎ 0.472⁎⁎⁎ 0.272⁎⁎⁎
(0.009) (0.045) (0.002) (0.024)
VAT rate shifting during the 3rd month (α3) 0.072⁎⁎⁎ 0.025 0.072⁎⁎⁎ 0.025⁎⁎
(0.009) (0.044) (0.002) (0.010)
VAT rate shifting during the 4th month (α4) 0.024⁎⁎⁎ 0.032 0.024⁎⁎⁎ 0.032⁎⁎⁎
(0.009) (0.044) (0.002) (0.009)
Consumer share+ 77%⁎⁎⁎ 57%⁎⁎⁎ 77%⁎⁎⁎ 57%⁎⁎⁎
(2%) (6%) (0.4%) (2.9%)
⁎⁎⁎: Significant at 1%, ⁎⁎: significant at 5%, ⁎: significant at 10%. Standard errors in parentheses.
Source: Carbonnier
+: Calculated as the sum of(2007), Table
the coefficients until2.the last significant, according to Eq. (5).
Note: This table presents the results of the regressions of housing repair service prices in the one hand and of new car sale
prices in the other hand, on VAT rate changes. Columns (1) and (2) present the result of the regressions with a standard
OLS method. Columns (3) and (4) present the result of the regressions with the White method for estimate the standard
errors.

109 / 111
References
[Atkinson and Stiglitz], chap. 6, 7 and 11.
[Gruber], chap. 2, 3, 19 and 20.
[Rosen and Gayer], chap. 2, 3 and 15.
[Stiglitz and Rosengard], chap. 3 and 18.
Auerbach, A. (1985). “The Theory of Excess Burden and Optimal Taxation”, in Auerbach,
A. and Feldstein, M. (eds), Handbook of Public Economics, vol. 3., chap. 2 .
Carbonnier, C. (2007). “Who Pays Sales Taxes? Evidence from French VAT Reforms,
1987–1999”, Journal of Public Economics, 91(5-6), pp. 1219–29.
Cohen, P., Hahn, R., Hall, J., Levitt, S. and Metcalfe, R. (2016). “Using Big Data to
Estimate Consumer Surplus: The Case of Uber”, NBER Working Paper No. 22627.
Fullerton, D. and Metcalf, G. (2002). “Tax Incidence”, in Auerbach, A. and Feldstein, M.
(eds), Handbook of Public Economics, vol. 4., chap. 26.
Glaeser, E. and Luttmer, E. (2003). “The Misallocation of Housing Under Rent Control”,
American Economic Review, 93(4), pp. 1027–1046.
Hines, J. (1999). “Three Sides of Harberger Triangles,” Journal of Economic Perspectives,
13(2), pp. 167–188.
Kotlikoff, L. and Summers, L. (1987). “Tax Incidence” in Auerbach, A. and Feldstein, M.
(eds), Handbook of Public Economics, North Holland, Vol. 2, pp. 1043–92.

110 / 111
References (continued)
Marion, J. and Muehlegger, E. (2008). “Measuring Illegal Activity and the Effects of
Regulatory Innovation: Tax Evasion and the Dyeing of Untaxed Diesel”, Journal of Political
Economy, 116(4), pp. 633–666.

111 / 111
Appendix
Utility Maximization
Utility maximization problem: household chooses the optimal
bundle of goods to consume, subject to a budget constraint:

max u(x)
x
s.t. p·x6m
Optimal bundle satisfies:

 ∂u(x)/∂xj = pj

∀j, k (i.e. MRSjk =
pj
)
∂u(x)/∂xk pk pk
p·x = m



The solution is called the Marshallian demand function:

x(p, m)
Indirect utility function:
v (p, m) = u (x(p, m))
A.1
Expenditure Minimization
Expenditure minimization problem: minimize expenditure
subject to the constraint that some minimum utility level u is
reached (“dual” problem):

min p·x
x
s.t. u(x) > u
Optimal bundle satisfies:
pj
MRSjk =
pk
u(x) = u
The solution is called the Hicksian demand function:
h(p, u)
Expenditure function:
e(p, u) = p.h(p, u)
A.2
Useful Identities

Roy’s Identity: if x(p, m) is the Marshallian demand function,

then:
∂v (p, m)/∂pi
xi (p, m) = ∀i
∂v (p, m)/∂m
∂v (p,m)
where ∂m is the marginal utility of income

Shephard’s Lemma: if h(p, u) is the Hicksian demand function,

then:
∂e(p, u)
hi (p, u) = ∀i
∂pi
Both identities are direct consequences of the envelope theorem

A.3
Comparative Statics
A change in the price of a good has two effects:
– a substitution effect: consumers react to an increase in the
relative price of goods by reducing consumption of the more
expensive good and increasing consumption of the cheaper
good
– an income effect: a price increase indirectly affects
consumption through a reduction in income

Slutsky equation:
∂x(p, m) ∂h(p, m) ∂x(p, m)
= − x(p, m)
∂p ∂p | ∂m {z }
| {z } | {z }
change in consumption substitution effect income effect

Sign:
– Normal goods: both effects < 0
– Inferior goods: substitution effect < 0, income effect > 0
A.4
 Income and Substitution Effects

Quantity of
good 2

A
C

B U1

UO

BC0 BCs BC1

Quantity of
Substitution Income good 1
effect effect
A.5
Producer Theory: Cost Minimization
Production function: Quantity of output produced as a function
of inputs:
q = F (x) = F (x1 , x2 , . . . , xn )

Cost minimization problem: minimizing the cost of producing a

given output target q given input prices wi :

min w·x
x
s.t. F (x) > q

Solution of this problem yields the cost function:

c(q)

Diminishing marginal productivity ( ∂F∂x(x)

i
< 0 ∀i) implies increasing
marginal cost (c 0 (q) > 0)
A.6
Producer Theory: Profit Maximization

Profit maximization: The producer solves

max p · q − c(q)
y

FOC:
c 0 (q) = p

The FOC implicitly defines the producer’s individual supply curve:

q = S(p)

Back

A.7
Change in Consumer Surplus
The change in consumer surplus ∆CS that follows a fall in the
price of good g (p0 → p1 , p1 < p0 ) can be written:
Z p0
∆CS = D(p)dp
p1

Proof:



∆CS = v (x1 ) − p1 x1 − v (x0 ) − p0 x0
Z x1
= v 0 (x)dx − p1 x1 + p0 x0
x0

We make the following substitution:

1
p = v 0 (x); dp = dx
D 0 (p)
Note: the second equality holds because the functions D(·) and
v 0 (·) are inverse to one another and the derivative of an inverse
function f −1 is equal to 1/f 0
A.8
Change in Consumer Surplus
Proof (continued):
Z p1
∆CS = p · D 0 (p)dp − p1 x1 + p0 x0
p0

Using the method of integration by parts, we get:

 p
Z p1
∆CS = p · D(p) p1 − D(p)dp − p1 x1 + p0 x0
0
p0
Z p1
∆CS = − D(p)dp
p0
Z p0
∆CS = D(p)dp
p1

which completes the proof.

Back

A.9
Change in Producer Surplus
The change in producer surplus ∆PS that follows a increase in
the price of good g (p0 → p1 , p1 > p0 ) can be written:
Z p1
∆PS = S(p)dp
p0

Proof:



∆PS = p1 x1 − c(x1 ) − p0 x0 − c(x0 )
Z y1
= p1 x1 − p0 x0 − c 0 (x)dx
y0

We make the following substitution:

1
p = c 0 (x); dp = dx
S 0 (p)
Note: the second equality holds because the functions S(·) and
c 0 (·) are inverse to one another and the derivative of an inverse
function f −1 is equal to 1/f 0
A.10
Change in Producer Surplus

Proof (continued):
Z p1
∆PS = p1 x1 − p0 x0 − p · S 0 (p)dp
p0

Using the method of integration by parts, we get:

 p
Z p1
∆PS = p1 x1 − p0 x0 − p · S(p) p1 + S(p)dp
0
p0
Z p1
∆PS = S(p)dp
p0

which completes the proof.

Back

A.11
Competitive Equilibrium and Efficiency
Consider I consumers with quasilinear preferences over a good G
(whose price is p) and a composite commodity (numeraire):

Ui (xi , yi ) = vi (xi ) + yi

with v 0 > 0, v 00 < 0 and v (0) = 0

Each consumer i initially owns an amount wi of the numeraire good

Consider J firms. Each firm can produce q units of good G by

using cj (q) units of the numeraire good

The firms’ marginal costs of production are increasing and convex

(c 0 > 0 and c 00 > 0) and there are no fixed costs (c(0) = 0)

Each
P consumer i owns a share θij of the profits of firm j
( i θij = 1)

A.12
Competitive Equilibrium
Market demand:
Individual demand curves: vi0 (xi ) = p ∀i ⇒ xi = Di (p)
P
Market demand curve: D(p) = i Di (p)

Market supply:
Individual supply curves: cj0 (qj ) = p ∀j ⇒ qj = Sj (p)
P
Market supply curve: S(p) = j Sj (p)

Competitive equilibrium: price p ∗ such that market supply equals

market demand
S(p ∗ ) = D(p ∗ )

the (I + J + 1) conditions determine the (I + J + 1) equilibrium

values (x1∗ , . . . , xI∗ , q1∗ , . . . , qJ∗ , p ∗ ), which are independent from
initial endowments and ownership shares of firms

A.13
Efficiency of Competitive Equilibrium
We now ask the question: is the competitive equilibrium efficient?
With quasilinear utility functions, the set of Pareto optimal
allocations is the set of allocations (x1o , . . . , xIo , q1o , . . . , qJo ) that
solves: X X
max vi (xi ) − cj (qj )
x1 ,...,xI ,q1 ,...,qJ
i j
X X
s.t. xi = qj
i j

[see Mas-Colell et al. (1995), chap. 10]

P P
The maximand i vi (xi ) − j cj (qj ) is called the Marshallian
aggregate surplus: it measures the net benefit to society from
producing and consuming good G
An efficient allocation is therefore an allocation that maximizes the
total surplus: the competitive equilibrium achieves this
A.14
Efficiency of Competitive Equilibrium

Letting λ denote the multiplier on the constraint, we have the

following FOCs:
vi0 (xio ) = λ ∀i
cj0 (qjo ) = λ ∀j
X X
xio = qjo
i j

These conditions are met by the competitive equilibrium (simply

replace λ by p ∗ )

⇒ The competitive equilibrium is Pareto optimal (First welfare

theorem)

Back

A.15
Measuring Welfare Changes: Numerical Example
The tenant’s (Marshallian) demand for housing x(p, m) is the
solution to the following maximization problem:

max ln x + ln(m − p · x)
x

FOC yields:
m
x(p, m) =
2p
The Marshallian demand for the composite good is:
m
c(p, m) =
2
The tenant’s indirect utility V(r,m) is:
 
  m m
V (p, m) = U x(p, m), c(p, m) = ln + ln
2p 2

A.16
Measuring Welfare Changes: Numerical Example

Plugging the numerical values into these formulas yields:

– Period 0 (p0 =20 euros/m2 ):

x0 = 50 m2 ; y0 = 1, 000 euros; V0 = 10.82 “utils”

The tenant spends half of her income on housing (renting a

50 m2 flat) and spends the other half on other goods
– Period 1 (p1 =10 euros/m2 ):

x1 = 100 m2 ; y1 = 1, 000 euros; V1 = 11.51 “utils”

The tenant spends half of her income on housing (renting a

100 m2 flat) and spends the other half on other goods. The
overall utility that she derives from her consumption is higher
than in period 0.

A.17
Measuring Welfare Changes: Numerical Example
Compensating variation CV is the solution to:

V (p1 , m − CV ) = V0
 
⇔ ln 2000−CV + ln 2000−CV


2×10 2 = 10.82
which yields: CV = 586 euros
Equivalent variation EV is the solution to:

V (p0 , m + EV ) = V1
 
⇔ ln 2000+EV + ln 2000+EV


2×20 2 = 11.51
which yields: EV = 828 euros
Change in consumer surplus:
Rp R 20 2000
 20
∆CS = p10 x(p, m)dp = 10 2p dp = 1000 ln p 10
which yields: ∆CS = 693 euros Back

A.18
Tax Incidence Formula
Equilibrium condition: D (p + τ ) = S (p)

Implicitly defines equilibrium price p(τ )

Differentiate both sides w.r.t. τ :

dD (p + τ ) dS (p)
=
dτ dτ
dp dp
⇒ D 0 (p) + 1 = S 0 (p)
dτ dτ
0
dp D (p)
⇒ = 0
dτ S (p) − D 0 (p)
dp εD
⇒ =
dτ εS − εD
qD 0 (q) pS 0 (p)
(using εD = D(q) and εS = S(p) , evaluated at τ = 0)
Back

A.19
DWL of Taxation: Harberger Formula
DWL of tax is the area of the triangle | τ| d
|dQ|

1
DWL = − dQ · dτ
2
1
DWL = − S 0 (p)dp · dτ
2
1 pS 0
  
S εD
DWL = − dτ 2
2 S p εS − εD
   2
1 εS · εD dτ
DWL = − (pQ)
2 εS − εD p

2nd line uses Q = S(p) ⇒ dQ = S 0 (p)dp

 
3rd line uses incidence formula dp = εSε−ε
D
D
dτ Incidence Formula

pS 0 (p)
4th line uses definition of εS = S(p) Back

A.20
DWL of Taxation: Implementable Formula

DWL of tax is the area of the triangle | τ|

d
|dQ|

1
DWL = − dQ · dτ
2  
1 dQ p
DWL = − dτ · dτ
2 dτ Q
 2
1 dτ
DWL = ηQ (pQ)
2 p

dQ p0
where ηQ = −
dτ Q
(ηQ : effect of a 1% increase in initial price via a tax change on
equilibrium quantity, taking into account endogenous price change)
Back

A.21